A Survey of Large Models in Sports

# A Survey of Large Models in Sports
Source: [https://arxiv.org/html/2608.14377](https://arxiv.org/html/2608.14377)
Yichen Xu Jianzhe Ma11footnotemark:1Chuhan Wang Zhonghao CaoThanks:These authors contributed equally\.Affiliation:Renmin University of ChinaAffiliation:Renmin University of ChinaAffiliation:Sichuan UniversityAffiliation:Beijing University of Posts and Telecommunications\{xu\_yichen, majianzhe, liangyuchen, wangwenxuan, qjin\}@ruc\.edu\.cnwangchuhan51@stu\.scu\.edu\.cncaozhonghao@bupt\.edu\.cnLiangyu ChenWenxuan WangQin Jin22footnotemark:2Thanks:Qin Jin and Wenxuan Wang are corresponding authors\.Affiliation:Renmin University of ChinaAffiliation:Renmin University of ChinaAffiliation:Renmin University of China

###### Abstract

Sports have witnessed growing global enthusiasm in recent years, serving as a vital force for physical health, cultural exchange, social connection, and economic growth\. The rapid advancement of large models, particularly \(multimodal\) large language models \(M\)LLMs, has demonstrated transformative potential to reshape sports understanding, analysis, and interaction across diverse domains\. This paper presents a comprehensive survey of large models in sports, including \(i\) an overview of tasks and applications across different participant groups; \(ii\) a detailed analysis of sports\-related datasets and benchmarks; and \(iii\) a critical discussion of current challenges and future directions\. Our goal is to establish a foundation for advancing research and practical development of large\-model\-driven sports intelligence\. An open\-source GitHub repository is maintained at:[https://github\.com/Road2Redemption/Awesome\_Large\_Models\_In\_Sports1](https://github.com/Road2Redemption/Awesome_Large_Models_In_Sports1)\.

## 1Introduction

In recent years, the global enthusiasm for sports has continued to rise, with more and more people actively participating in it, and the sports industry has also flourished\. To further drive this development, modern sports increasingly rely on massive data support\([103](https://arxiv.org/html/2608.14377#bib.bib3)\), while the introduction of Artificial Intelligence \(AI\) has greatly accelerated this trend\([321](https://arxiv.org/html/2608.14377#bib.bib4)\)\. A pivotal pillar of this transformation is the ability to process and generate sports\-related language, which serves as a vital bridge translating raw athletic data into actionable insights for participants and fans alike\.

Early interdisciplinary research in sports and AI focused on natural language processing and computer vision, with applications in tasks such as sports data processing\([49](https://arxiv.org/html/2608.14377#bib.bib5)\)and video analysis\([178](https://arxiv.org/html/2608.14377#bib.bib6)\)\. As shown in Figure[1](https://arxiv.org/html/2608.14377#S1.F1), the transition to the era of large models—underpinned by the rapid evolution of Large Language Models \(LLMs\) and Multimodal Large Language Models \(MLLMs\) like GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)and Gemini\([248](https://arxiv.org/html/2608.14377#bib.bib1)\)—has brought new opportunities and challenges to the sports domain\. With linguistic intelligence at their core, these models not only generate language effectively but also process multiple data modalities, enabling broader applications in sports\. Tasks that were previously difficult—such as designing athlete training plans\([234](https://arxiv.org/html/2608.14377#bib.bib15)\), developing coaching strategies\([24](https://arxiv.org/html/2608.14377#bib.bib14)\), and generating sports game summarization\([100](https://arxiv.org/html/2608.14377#bib.bib13)\)—have been greatly enhanced by large models\. Moreover, leveraging their vast knowledge bases, these models can generate more comprehensive and personalized content\([144](https://arxiv.org/html/2608.14377#bib.bib92)\)\. The number of papers on large models in sports has grown rapidly, from just 1 in 2020 to 78 in 2024, and continues to increase in 2025 \(see Figure[4](https://arxiv.org/html/2608.14377#A0.F4)in the Appendix\)\.

![Refer to caption](https://arxiv.org/html/2608.14377v1/sports_survey_figure1_v5_2026.png)Figure 1:Large models have shown powerful applications across 6 sports stakeholder groups: athletes and trainers, coaches and educators, referees, researchers, fans and social media, and the sports industry, enabling diverse tasks\.A growing body of review literature has examined the use of AI and deep learning in sports\([321](https://arxiv.org/html/2608.14377#bib.bib4);[320](https://arxiv.org/html/2608.14377#bib.bib11)\)\. The most relevant work on large models includes studies on their applications to exercise recommendations\([124](https://arxiv.org/html/2608.14377#bib.bib7)\), sports science and medicine\([46](https://arxiv.org/html/2608.14377#bib.bib8);[180](https://arxiv.org/html/2608.14377#bib.bib9)\), and the sports industry\([275](https://arxiv.org/html/2608.14377#bib.bib10)\), along with surveys of datasets for language and multimodal models\([286](https://arxiv.org/html/2608.14377#bib.bib12)\)\. However, these studies are still limited in scope, lacking comprehensive coverage of the diverse sports\-related tasks and datasets where large models can be applied\.

To ensure a comprehensive and rigorous survey, we adopted a systematic snowballing methodology\([281](https://arxiv.org/html/2608.14377#bib.bib304)\), adhering to the PRISMA statement\([188](https://arxiv.org/html/2608.14377#bib.bib305)\)\. Starting from the aforementioned review papers, we performed iterative forward and backward searches to capture the latest advancements in the era of large models \(Jan 2020–July 2025\)\. This process resulted in a final collection of241core academic papers addressing large models in sports\. The detailed selection methodology is provided in the Appendix[A](https://arxiv.org/html/2608.14377#A1)\.

We first systematically categorize and summarize existing applications of large models in sports across 6 key groups \(§[2](https://arxiv.org/html/2608.14377#S2)\)\. Then, we review and conduct an in\-depth analysis of the relevant datasets and benchmarks in sports \(§[3](https://arxiv.org/html/2608.14377#S3)\)\. Subsequently, we discuss the current challenges in this field and, finally, outline the potential future directions \(§[4](https://arxiv.org/html/2608.14377#S4)\)\.

\{forest\}Figure 2:Taxonomy of applications, tasks, and approaches of large models in sports\.
## 2Large Model Applications in Sports

The fast growth of large models has brought big chances for their use in sports\. As shown in Figure[2](https://arxiv.org/html/2608.14377#S1.F2), we categorize these applications into a taxonomy with6 stakeholder groupsand19 specific tasks\. In this section, beyond merely listing existing literature, we conducta detailed analysisof the impact of large models on each task, focusing on defining the task, analyzing technical paradigms, and summarizing common evaluation metrics\. For comprehensive reviews of specific works associated with each task, see Appendix[B](https://arxiv.org/html/2608.14377#A2)\.

### 2\.1Applications for Athletes and Trainers

Exercise and Training Plans\.Large models help athletes and trainers create exercise prescriptions, translating sports science into practice to improve performance\([196](https://arxiv.org/html/2608.14377#bib.bib16);[263](https://arxiv.org/html/2608.14377#bib.bib17)\)\. Recent AI coaches powered by LLMs significantly facilitate the generation of personalized training plans across a wide spectrum of health conditions and fitness goals, ranging from general weight management\([221](https://arxiv.org/html/2608.14377#bib.bib24)\)to chronic disease guidance\([184](https://arxiv.org/html/2608.14377#bib.bib53)\)\. As shown in Table[1](https://arxiv.org/html/2608.14377#S2.T1), in the YourSkatingCoach dataset\([36](https://arxiv.org/html/2608.14377#bib.bib233)\), a fine\-tuned T5 model\([206](https://arxiv.org/html/2608.14377#bib.bib239)\)achieves a BLEU\-4 score of0\.27, while a vanilla Transformer\([260](https://arxiv.org/html/2608.14377#bib.bib306)\)trained from scratch only reaches 0\.04\([300](https://arxiv.org/html/2608.14377#bib.bib34)\)\. This highlights that LLMs leverage pre\-trained knowledge to address sports data scarcity and excel at open\-ended generation, outperforming traditional rule\-based or smaller deep learning models\. Additionally, strategies like Retrieval\-Augmented Generation \(RAG\)\([314](https://arxiv.org/html/2608.14377#bib.bib39)\)and agentic paradigms\([256](https://arxiv.org/html/2608.14377#bib.bib30)\)have been explored to enhance reliability and personalization\. Common evaluation metrics include BLEU\-4, METEOR, and ROUGE\-L\.

Sports Injury and Rehabilitation\.Large models assist athletes and trainers throughout the entire lifecycle of sports injury management, spanning prevention, diagnosis, and rehabilitation, with applications expanding from providing preventive advice\([326](https://arxiv.org/html/2608.14377#bib.bib54)\)to aiding clinical decision\-making for surgical treatments\([216](https://arxiv.org/html/2608.14377#bib.bib63)\)\. While LLMs possess the interdisciplinary knowledge required for orthopedics and rehabilitation\([165](https://arxiv.org/html/2608.14377#bib.bib59)\), most current applications rely on the direct deployment of pre\-trained large models for Question Answering \(QA\) and classification\. Deep technical integration remains limited, with only early exploration of efficient fine\-tuning methods like GaLore\([319](https://arxiv.org/html/2608.14377#bib.bib307)\)to tailor models for sports medicine[326](https://arxiv.org/html/2608.14377#bib.bib54)\. This indicates the field is in its infancy, lacking unified evaluation metrics\.

Sports Psychology and Behavior\.Sports psychology enhances athletes’ training performance and mental well\-being through behavioral interventions\. Recent LLM applications range from general cognitive assessment\([329](https://arxiv.org/html/2608.14377#bib.bib69)\)to targeted interventions for specific behavioral issues\([162](https://arxiv.org/html/2608.14377#bib.bib67)\)\. Recent advances move beyond text generation by integrating multimodal physiological data from wearable sensors—such as heart rate and IMU signals—to deliver personalized interventions\([105](https://arxiv.org/html/2608.14377#bib.bib73);[168](https://arxiv.org/html/2608.14377#bib.bib70)\)\. However, task definitions remain ambiguous, and the area lacks standardized benchmarks, requiring further exploration\.

DatasetModelArchitectureParadigmMetricPerformanceExercise and Training PlansYourSkatingCoach\([36](https://arxiv.org/html/2608.14377#bib.bib233)\)MAAIG\([300](https://arxiv.org/html/2608.14377#bib.bib34)\)T5 \(pretrained\)\([206](https://arxiv.org/html/2608.14377#bib.bib239)\)fine\-tuningBLEU\-40\.27Transformer\([260](https://arxiv.org/html/2608.14377#bib.bib306)\)vanilla Transformertrain from scratch0\.04Action Spotting and RecognitionSoccerNet\-v2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)Soccer\-CLIP\([233](https://arxiv.org/html/2608.14377#bib.bib148)\)CLIP\([205](https://arxiv.org/html/2608.14377#bib.bib308)\)fine\-tuningt\-AmAP75\.7COMEDIAN\([53](https://arxiv.org/html/2608.14377#bib.bib309)\)spatiotemporal Transformertrain from scratch73\.1Llama 3\.1\-8B\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)LLM w/ textual commentary\([28](https://arxiv.org/html/2608.14377#bib.bib153)\)few\-shot60\.8Sports Action Quality AssessmentFineFS\([109](https://arxiv.org/html/2608.14377#bib.bib267)\)Beats\-to\-Scores\([264](https://arxiv.org/html/2608.14377#bib.bib169)\)Video\-Audio \(V\-A\) fusion Transformertrain from scratchSpearman’sρ\\rho0\.88InternVL2\([38](https://arxiv.org/html/2608.14377#bib.bib268)\)InternViT \+ MLP \+ InternLM2fine\-tuning0\.86Qwen2\-VL\([268](https://arxiv.org/html/2608.14377#bib.bib276)\)ViT \+ MLP \+ Qwen2fine\-tuning0\.75Sports Commentary GenerationSoccerNet\-Caption\([173](https://arxiv.org/html/2608.14377#bib.bib107)\)MatchVoice\([209](https://arxiv.org/html/2608.14377#bib.bib90)\)ViT \+ Aggregator & MLP \+ Llama 3fine\-tuningCIDEr38\.42SoccerComment\([136](https://arxiv.org/html/2608.14377#bib.bib101)\)MLLM \+ memory unitfine\-tuning36\.58SN\-Caption\([173](https://arxiv.org/html/2608.14377#bib.bib107)\)encoder\-decoder Transformertrain from scratch23\.74Video\-LLaMA\([311](https://arxiv.org/html/2608.14377#bib.bib284)\)V\-A encoder \+ Q\-Former \+ LLaMAzero\-shot3\.44

Table 1:Quantitative comparison of modeling paradigms across 4 representative sports tasks\.
### 2\.2Applications for Coaches and Educators

Action Spotting and Recognition\.Action spotting and recognition in sports involves temporally localizing and classifying fine\-grained player movements or events to provide reliable match facts for downstream analytics\([320](https://arxiv.org/html/2608.14377#bib.bib11)\)\. Traditional methods relied on specific deep learning architectures trained from scratch, whereas recent MLLMs leverage pre\-training alignment for enhanced semantic understanding\. As shown in Table[1](https://arxiv.org/html/2608.14377#S2.T1), on the SoccerNet\-v2 action spotting benchmark\([52](https://arxiv.org/html/2608.14377#bib.bib247)\), fine\-tuned Soccer\-CLIP achieves a state\-of\-the\-art75\.7%t\-AmAP\([233](https://arxiv.org/html/2608.14377#bib.bib148)\), slightly surpassing specialized Transformers \(73\.1%\)\([53](https://arxiv.org/html/2608.14377#bib.bib309)\)and highlighting the importance of pre\-training methods\. In contrast, relying solely on language\-centric LLMs through textual commentary prompts yields significantly lower results \(60\.8%\)\([28](https://arxiv.org/html/2608.14377#bib.bib153)\), underscoring the necessity of fine\-grained visual alignment rather than merely injecting textualized visual information\. Common metrics include mAP, top\-1 accuracy, and F1 score\.

Sports Action Quality Assessment\.Sports Action Quality Assessment \(AQA\) quantifies the execution of athletic movements for coaching and officiating\([323](https://arxiv.org/html/2608.14377#bib.bib168);[320](https://arxiv.org/html/2608.14377#bib.bib11)\)\. Methodologies have evolved from simple regression to fine\-tuning MLLMs for personalized evaluation\([58](https://arxiv.org/html/2608.14377#bib.bib170)\)and developing unified agents\([247](https://arxiv.org/html/2608.14377#bib.bib171)\)\. On the FineFS benchmark\([109](https://arxiv.org/html/2608.14377#bib.bib267)\)\(see Table[1](https://arxiv.org/html/2608.14377#S2.T1)\), specialized small\-scale Transformers currently outperform general MLLMs \(0\.88versus 0\.86 Spearman’sρ\\rho\) by explicitly aligning audio\-visual features\([264](https://arxiv.org/html/2608.14377#bib.bib169)\)\. This indicates that high\-precision scoring still depends on domain\-specific traditional structural designs\. Moreover, architectural choices within MLLMs remain pivotal, as seen in models like InternVL2\([38](https://arxiv.org/html/2608.14377#bib.bib268)\)and Qwen2\-VL[268](https://arxiv.org/html/2608.14377#bib.bib276); model design and training details can lead to noticeably different performance on this task\. Common metrics include Spearman’s rank correlation, mean square error, and accuracy\.

Sports Tactics and Strategies\.Sports tactics and strategy analysis models on\-field interactions to extract actionable strategic patterns using large models\([26](https://arxiv.org/html/2608.14377#bib.bib154)\)\. Current methodologies employ large model\-based frameworks for tactical analysis and visualization \(processing structured and unstructured data\)\([106](https://arxiv.org/html/2608.14377#bib.bib160);[171](https://arxiv.org/html/2608.14377#bib.bib157)\), and tactical exploration and design\([151](https://arxiv.org/html/2608.14377#bib.bib155)\)\. Technically, current research mainly relies on prompt engineering with pre\-trained large models, rather than extensive post\-training, due to the scarcity of high\-quality tactical datasets\. This limits the depth of tactical discovery to the capabilities of the frozen base model, indicating that the field is still nascent and requires future exploration to address these data and methodological constraints\.

Game and Player Performance Prediction\.Game and player performance prediction utilizes historical, contextual, and multimodal data to forecast match outcomes and individual behaviors, thereby providing valuable insights for strategic planning and preparation\([286](https://arxiv.org/html/2608.14377#bib.bib12)\)\. Methodologies have advanced from BERT\-based specific action forecasting\([104](https://arxiv.org/html/2608.14377#bib.bib163)\)to LLM\-driven approaches that integrate diverse data sources for more holistic and interpretable predictions\([18](https://arxiv.org/html/2608.14377#bib.bib167)\)\. Common metrics include accuracy and F1 score\.

Sports Education\.Recent applications of large models in sports education have demonstrated their versatility for educators and teachers\. Current research primarily uses general\-purpose LLMs to generate and analyze pedagogical data and content\([316](https://arxiv.org/html/2608.14377#bib.bib179);[71](https://arxiv.org/html/2608.14377#bib.bib180)\)\. However, a gap exists in high\-level applications\. Professional athlete guidance, in particular, demands deep domain\-specific expertise that general models often lack, presenting a promising direction for exploration\.

### 2\.3Applications for Referees

Sports Refereeing\.Large models improve sports refereeing by supporting decision\-making and enhancing fairness and transparency\. Key tasks include QA, captioning, and action recognition\. For example, X\-VARS\([93](https://arxiv.org/html/2608.14377#bib.bib181)\)uses QLoRA\([56](https://arxiv.org/html/2608.14377#bib.bib310)\)fine\-tuning on MLLMs to accurately understand video content while following soccer rules, representing the first step toward explainable LLMs for refereeing\.

### 2\.4Applications for Fans and Social Media

Sports Commentary Generation\.Sports commentary generation creates natural\-language narratives that integrate factual event descriptions, tactical analysis, and emotionally resonant insights, setting it apart from standard video captioning\([74](https://arxiv.org/html/2608.14377#bib.bib97)\)\. Recent methods have advanced from end\-to\-end fine\-tuning of MLLMs for better temporal alignment and coherence\([209](https://arxiv.org/html/2608.14377#bib.bib90);[277](https://arxiv.org/html/2608.14377#bib.bib91)\)to agentic frameworks that dynamically adapt by prompting LLMs with key events and tracking data\([7](https://arxiv.org/html/2608.14377#bib.bib96);[261](https://arxiv.org/html/2608.14377#bib.bib95)\)\. As shown in Table[1](https://arxiv.org/html/2608.14377#S2.T1), in this semantic\-rich task, adapted MLLM architectures like MatchVoice \(CIDEr:38\.42\)\([209](https://arxiv.org/html/2608.14377#bib.bib90)\)decisively outperform traditional encoder\-decoder models \(23\.74\)\([173](https://arxiv.org/html/2608.14377#bib.bib107)\)due to temporal aggregators that handle long\-form narratives\. Conversely, the near\-failure of zero\-shot Video\-LLaMA \(3\.44\)\([311](https://arxiv.org/html/2608.14377#bib.bib284)\)confirms that practical utility requires domain\-specific fine\-tuning or RAG\([136](https://arxiv.org/html/2608.14377#bib.bib101)\)to bridge the linguistic gap between raw visual signals and professional terminology\. Key metrics include METEOR, ROUGE\-L, and CIDEr\.

Sports Highlight Generation\.Sports highlight generation aims to automatically identify and compile significant match moments into concise summaries for social media\. Existing approaches typically employ hybrid frameworks combining computer vision and LLMs to perform sub\-tasks like key frame extraction, event localization, video clipping, and captioning\([129](https://arxiv.org/html/2608.14377#bib.bib311);[172](https://arxiv.org/html/2608.14377#bib.bib111)\)\. However, the field faces ambiguous definitions and variance in practical applications, requiring future research to clarify task boundaries and standardize protocols\.

Sports News Generation\.Sports news generation automatically produces factual match reports in a standard journalistic style, summarizing key outcomes, events, and statistics\. Existing techniques primarily rely on comprehensive frameworks powered by large models\. Recent advances include knowledge retrieval[267](https://arxiv.org/html/2608.14377#bib.bib116), in\-context learning[40](https://arxiv.org/html/2608.14377#bib.bib117), and Chain\-of\-Thought \(CoT\) prompting[43](https://arxiv.org/html/2608.14377#bib.bib120)to enhance quality\. Notably, specialized methods like Tree\-of\-Report address table\-to\-text challenges, ensuring accurate conversion of structured data into coherent narratives\([44](https://arxiv.org/html/2608.14377#bib.bib119)\)\. Common metrics include ROUGE\-L, F1 score, and LLM\-based metrics\.

Sports Narratives and Storytelling\.Sports narratives and storytelling aim to create long\-form, multimodal stories that combine match events with contextual details to engage fans\. Current work typically inputs keyframe information, commentator narration, and other relevant data into LLMs to generate engaging tactical analyses and personalized narratives for social media\([224](https://arxiv.org/html/2608.14377#bib.bib125);[143](https://arxiv.org/html/2608.14377#bib.bib124)\)\. While LLMs excel at crafting narratives, they often struggle with the intricacies of specific sports domains\. To address this, meticulous prompt engineering is crucial to prevent factual inaccuracies and enhance personalization and engagement\. Additionally, the field urgently needs unified evaluation metrics to effectively assess narrative quality\.

Public Opinion Analysis in Sports\.Public opinion and sentiment analysis in sports involves detecting, classifying, and measuring public attitudes toward sporting events or related issues\. However, achieving high accuracy in sports sentiment analysis is challenging due to the complexity of sport\-specific contexts\. Recent works have employed strategies such as fine\-tuning on sports corpora\([202](https://arxiv.org/html/2608.14377#bib.bib128)\)and inference\-time techniques like in\-context learning and CoT prompting\([211](https://arxiv.org/html/2608.14377#bib.bib129)\)\. Yet, current research is mostly limited to small\-scale analyses\. Scaling these approaches to larger datasets is a crucial future direction to establish the generalizability and practical significance of the findings\. Common evaluation metrics include accuracy and F1 score\.

Sports Models and Systems\.Unlike task\-specific research, work on sports models and systems targets general\-purpose infrastructures\. These include:\(1\) sports\-related chatbots and modelsthat utilize dialogue state tracking[237](https://arxiv.org/html/2608.14377#bib.bib134), domain\-specific fine\-tuning[208](https://arxiv.org/html/2608.14377#bib.bib143), and multi\-agent frameworks to coordinate specialized reasoning[207](https://arxiv.org/html/2608.14377#bib.bib142)and knowledge graph integration[31](https://arxiv.org/html/2608.14377#bib.bib188); and\(2\) search engines and retrieval systemsthat employ RAG architectures and offline query understanding to ground LLM outputs in verified sports facts\([115](https://arxiv.org/html/2608.14377#bib.bib132);[243](https://arxiv.org/html/2608.14377#bib.bib136)\)and facilitate fine\-grained video retrieval\([80](https://arxiv.org/html/2608.14377#bib.bib145)\)\.

### 2\.5Applications for Researchers

Sports Academic Writing\.Sports academic writing entails crafting and refining scholarly content in fields like sports science and medicine\. The advent of LLMs like ChatGPT has revolutionized this area, shifting the writing process from traditional manual methods to AI\-assisted collaboration\. Current LLMs excel at generating structured text, such as research outlines and abstract summaries\([125](https://arxiv.org/html/2608.14377#bib.bib78)\)\. However, their reliability for scientific accuracy is undermined by model hallucinations, which often compromise factual integrity and calculation precision\([170](https://arxiv.org/html/2608.14377#bib.bib75);[54](https://arxiv.org/html/2608.14377#bib.bib77)\)\. Thus, while these models can be efficient writing partners, their outputs need thorough human verification and cautious use\.

### 2\.6Applications for the Sports Industry

Sports Management\.Large models are gaining traction in sports management, covering areas like financial, database, and facility management\. Unlike traditional tools, large models excel at processing both structured and unstructured data, such as PDF reports and long interview transcripts\([167](https://arxiv.org/html/2608.14377#bib.bib82);[82](https://arxiv.org/html/2608.14377#bib.bib80)\)\. However, empirical research in this domain is still limited and requires further exploration\.

Sports Talent Scouting\.Sports talent scouting is vital for clubs to identify, evaluate, and predict player potential, thus building successful teams\. Large models, capable of processing vast data, can make this process more objective and data\-driven\([163](https://arxiv.org/html/2608.14377#bib.bib88)\)\. Recent research has used RAG to search unstructured data, speeding up football talent scouting\([210](https://arxiv.org/html/2608.14377#bib.bib87);[160](https://arxiv.org/html/2608.14377#bib.bib86)\)\. Yet, there is still much room to define and expand this task to other sports\.

Sports Tourism\.Sports tourism integrates travel services with athletic activities and major events, enriching the experiences of fans and participants\. In this realm, large models play key roles, such as analyzing tourism trends and enhancing community engagement\([302](https://arxiv.org/html/2608.14377#bib.bib84)\)\. This shift transforms traditional, static travel planning into a dynamic, real\-time interactive experience\. Nevertheless, current models encounter several challenges, including privacy and data security concerns, as well as managing fan expectations and trust\([166](https://arxiv.org/html/2608.14377#bib.bib85)\)\. Tackling these issues will be essential for future advancements\.

## 3Datasets for Large Models in Sports

![Refer to caption](https://arxiv.org/html/2608.14377v1/figure3a_revised_fixed.png)\(a\)Analysis of sports types
\(b\)Analysis of application distribution![Refer to caption](https://arxiv.org/html/2608.14377v1/figure3c_revised_fixed.png)\(c\)Analysis of modality composition

Figure 3:Data analysis from 3 different perspectives\. Sports and tasks abbreviations are listed in Table[6](https://arxiv.org/html/2608.14377#A3.T6)\.In this section, we first categorize the landscape of sports datasets for large models into two main types based on their design objectives:task\-specific datasetsandsports understanding datasets\. We then conduct a comprehensive multi\-dimensional analysis of theirdataset distributionsacross various facets to identify key trends and research gaps\. More details are provided in the Appendix[C](https://arxiv.org/html/2608.14377#A3)\.

BenchmarkSportsModal\# Video\# QABIG\-bench\-SU\([241](https://arxiv.org/html/2608.14377#bib.bib190)\)SC, BK, etc\.text\-986SportQA\([285](https://arxiv.org/html/2608.14377#bib.bib185)\)TN, AF, etc\.text\-70592SPORTU\([287](https://arxiv.org/html/2608.14377#bib.bib184)\)BB, IH, etc\.video, text170112948Sports\-3K\-QA\([34](https://arxiv.org/html/2608.14377#bib.bib100)\)49 Sportsvideo, text4121174FSBench\([70](https://arxiv.org/html/2608.14377#bib.bib183)\)FSvideo, text7834000FBBench\([88](https://arxiv.org/html/2608.14377#bib.bib187)\)BMvideo, text25632563Gym\-QA\([31](https://arxiv.org/html/2608.14377#bib.bib188)\)GYvideo, text603127469Diving\-QA\([31](https://arxiv.org/html/2608.14377#bib.bib188)\)DVvideo, text∼\\sim1001055Sports\-QA\([134](https://arxiv.org/html/2608.14377#bib.bib189)\)GY, VB, etc\.video, text596794073Table 2:Overview of specialized sports understanding benchmarks related to large models\. Sports abbreviations are listed in Table[6](https://arxiv.org/html/2608.14377#A3.T6)\.### 3\.1Landscape and Categorization

Task\-Specific Datasets\.Task\-specific datasets are created to support the practical applications of large models in various sports contexts, covering the 6 stakeholder groups and 19 tasks outlined in Section[2](https://arxiv.org/html/2608.14377#S2)\. These datasets offer detailed annotations and evaluation metrics, facilitating model training, fine\-tuning, and performance assessment\. They bridge the gap between the general capabilities of large models and real\-world sports applications, enabling customized system development, reproducible evaluation, and advancing research on model deployment\. Further details are provided in the Appendix[C\.1](https://arxiv.org/html/2608.14377#A3.SS1), and Tables[3](https://arxiv.org/html/2608.14377#A3.T3)and[4](https://arxiv.org/html/2608.14377#A3.T4)\.

Sports Understanding Datasets\.Sports are fast\-paced, diverse, and strategically complex, presenting unique challenges for large models\([285](https://arxiv.org/html/2608.14377#bib.bib185)\)\. To enhance models’ comprehension and reasoning in sports contexts, researchers have developed sports understanding datasets\. These datasets fall into two main categories:\(1\) datasets specifically for sports understanding, with Table[2](https://arxiv.org/html/2608.14377#S3.T2)providing an overview of relevant benchmarks; and\(2\) general video understanding datasets containing sports content, covering tasks like video captioning\([272](https://arxiv.org/html/2608.14377#bib.bib195)\), multi\-view understanding\([79](https://arxiv.org/html/2608.14377#bib.bib197)\), and fine\-grained analysis\([148](https://arxiv.org/html/2608.14377#bib.bib196)\)\. More details are in the Appendix[C\.2](https://arxiv.org/html/2608.14377#A3.SS2)and Table[5](https://arxiv.org/html/2608.14377#A3.T5)\.

### 3\.2Comprehensive Analysis

Dataset Distribution by Sport Type\.As illustrated in Figure[3\(a\)](https://arxiv.org/html/2608.14377#S3.F3.sf1), the availability of datasets is notably skewed toward popular invasion team sports \(e\.g\., soccer leading with 74 datasets\), racket and table sports, and bat\-and\-ball sports\. Conversely, individual disciplines such as cycling and boxing remain underrepresented with only a single dataset each, highlighting a significant coverage imbalance\. Furthermore, the prevalence of fitness\-related datasets \(21\) and the emergence of esports datasets \(3\) reflect the growing scholarly interest in these evolving domains\.

Dataset Distribution by Application\.Analysis of the 6 stakeholder groups \(Figure[3\(b\)](https://arxiv.org/html/2608.14377#S3.F3.sf2)\) reveals that data representation is robust for athletes, coaches, and fans, aligning with the commercial popularity of these segments\. Tasks such as sports commentary generation, exercise prescription, and action recognition have garnered substantial attention\. In contrast, data for referees, sports researchers, and the sports industry remain scarce, with referee\-relevant datasets accounting for a mere 2%\. This disparity underscores an urgent need to develop specialized datasets to bridge these application gaps\.

Dataset Distribution by Modality\.In the era of large\-scale models, video and text remain the dominant modalities \(Figure[3\(c\)](https://arxiv.org/html/2608.14377#S3.F3.sf3)\), while audio\-centric studies are beginning to demonstrate their significance\([289](https://arxiv.org/html/2608.14377#bib.bib216)\)\. However, specialized modalities such as IMU sensor data and skeletal poses are relatively rare and primarily confined to athlete\-focused motion analysis\. Expanding the diversity of these less common modalities is essential to strengthening the cross\-modal reasoning capabilities of MLLMs in complex sports scenarios\.

Dataset Distribution by Annotation Source\.Based on Table[5](https://arxiv.org/html/2608.14377#A3.T5), we categorized the datasets by their annotation origin \(automatic, manual, and expert\)\. While manual annotation remains the standard to minimize noise, a critical deficit exists in expert\-level labels\. This is particularly evident in general datasets, where expert annotation accounts for only 6%, compared to 41% in specialized sports understanding benchmarks\. This lack of high\-quality, professional\-grade labels poses a key bottleneck for the fine\-tuning and reliability of models intended for elite\-level sports analysis\.

Dataset Distribution by Modeling Paradigm\.Sports datasets can be categorized based on their supported tasks, falling into two paradigms: discriminative and generative\.Discriminative\-oriented datasets\(21\.5%\), such as action recognition and game prediction, benefit from large models’ ability to capture spatio\-temporal contexts—a notable improvement over traditional methods’ handling of complex multimodal inputs\. Conversely,generative tasksdominate the landscape \(78\.5%\), spanning from descriptive applications like commentary generation to reasoning\-intensive tactical synthesis\. This transition underscores a paradigm shift in sports AI: moving beyond simple categorical labeling toward open\-ended synthesis and logical reasoning facilitated by large models\.

## 4Discussion

In the preceding sections, we have examined the landscape of large models in sports and their emerging capabilities across a range of stakeholders\. Building upon these insights, this discussion distills the key barriers to practical deployment and outlines promising directions for future research\.

### 4\.1Challenges

Despite rapid progress, existing large models in sports still encounter 4 fundamental challenges that hinder robust and trustworthy real\-world adoption\.

Bias, Fairness, and Privacy\.Current sports datasets are heavily skewed toward a small set of popular sports such as soccer and basketball\([52](https://arxiv.org/html/2608.14377#bib.bib247);[283](https://arxiv.org/html/2608.14377#bib.bib103)\), leaving many less popular sports largely underrepresented\. Moreover, existing research disproportionately focuses on data from elite, mainstream competitions, while settings such as the Paralympics, youth development programs, and school sports remain underexplored\. Current datasets and models also predominantly focus on men’s sports, while women’s leagues and competitions remain substantially underrepresented\([19](https://arxiv.org/html/2608.14377#bib.bib230)\)\. These biases may lead to unfair model behaviors and limit generalization across diverse sporting contexts\([54](https://arxiv.org/html/2608.14377#bib.bib77);[191](https://arxiv.org/html/2608.14377#bib.bib231)\)\.

Real\-Time and High\-FPS Understanding\.Sports applications such as live officiating and broadcast commentary are highly latency\-sensitive, yet current Video LLMs still incur substantial inference overhead\([304](https://arxiv.org/html/2608.14377#bib.bib89)\), limiting their use in time\-critical workflows\. Moreover, long\-video understanding remains difficult: sports broadcasts often last for hours and demand sustained temporal reasoning over extended contexts\([328](https://arxiv.org/html/2608.14377#bib.bib229)\)\. Finally, most generic video pipelines are not designed for high\-frame\-rate inputs\. In sports, decisive cues \(e\.g\., ball contact, offside timing, foul initiation\) can unfold within milliseconds; aggressive temporal downsampling removes these fine\-grained dynamics, degrading event localization and rule\-level judgments\([138](https://arxiv.org/html/2608.14377#bib.bib152)\)\.

Hallucination and Interpretability\.In practical sports workflows, stakeholders require explanations that directly support decisions, rather than descriptive summaries\. For example, coaches and analysts need to identify actionable causes and detailed explanations, which remains challenging for current black\-box models\([169](https://arxiv.org/html/2608.14377#bib.bib317)\)\. Meanwhile, hallucinations in sports often invent key events, actors, or causal links, producing plausible narratives that can directly mislead downstream decisions\. Such errors are especially harmful in high\-stakes settings as they can quickly erode user trust\([204](https://arxiv.org/html/2608.14377#bib.bib46);[93](https://arxiv.org/html/2608.14377#bib.bib181)\)\.

Practicability and Real\-World Deployment\.Recent research still falls short of real sports workflows in terms of ecological validity\. Although models can perform well on curated clips, they often break down in the wild due to heavy occlusion, shifting camera viewpoints, and low\-quality footage common in sports scenarios\([181](https://arxiv.org/html/2608.14377#bib.bib203)\)\. At the same time, real\-world deployment remains largely underexplored: practical setups such as deploying models on the sidelines or on edge devices like wearables and drone cameras remain challenging and are rarely validated in real\-world settings\([121](https://arxiv.org/html/2608.14377#bib.bib324);[15](https://arxiv.org/html/2608.14377#bib.bib325)\)\.

### 4\.2Future Directions

Building on the challenges discussed above, we outline 4 future directions to advance large models in sports toward robust real\-world use\.

Trustworthy Sport AI\.Future research should address these ethical and reliability challenges by integrating advanced technical safeguards into model development\. Key directions include implementing rigorous data balancing and cleaning\([13](https://arxiv.org/html/2608.14377#bib.bib328)\)alongside sport\-specific alignment via Reinforcement Learning from Human Feedback \(RLHF\) to minimize bias\([307](https://arxiv.org/html/2608.14377#bib.bib329);[69](https://arxiv.org/html/2608.14377#bib.bib327)\)\.

Streaming and Long Video Mechanisms\.Future work should better handle sports’ temporal demands\. For low latency, explore streaming inference with more efficient KV\-cache and attention mechanisms\([32](https://arxiv.org/html/2608.14377#bib.bib312);[60](https://arxiv.org/html/2608.14377#bib.bib314);[293](https://arxiv.org/html/2608.14377#bib.bib313)\)\. For long matches, adapt long\-video modeling via memory, parallelism, and token compression, or architectures like Mamba\([212](https://arxiv.org/html/2608.14377#bib.bib315);[304](https://arxiv.org/html/2608.14377#bib.bib89);[37](https://arxiv.org/html/2608.14377#bib.bib219);[271](https://arxiv.org/html/2608.14377#bib.bib316)\)\. For high\-FPS events, develop vision backbones that support dense frames without losing fine\-grained dynamics\([138](https://arxiv.org/html/2608.14377#bib.bib152)\)\.

Knowledge Grounding and Tool Use\.Future work should improve factual reliability via explicit retrieval from structured knowledge \(e\.g\., RAG\)\([242](https://arxiv.org/html/2608.14377#bib.bib135);[231](https://arxiv.org/html/2608.14377#bib.bib138)\), grounding claims in visual evidence\([284](https://arxiv.org/html/2608.14377#bib.bib326)\), and producing explicit rationales\([93](https://arxiv.org/html/2608.14377#bib.bib181)\)\. Tool\-enabled models that query live databases, rule engines, or match\-tracking APIs can further make outputs verifiable and logically consistent\([207](https://arxiv.org/html/2608.14377#bib.bib142)\)\.

In\-the\-Wild Evaluation and Edge Deployment\.Future work should prioritize practical deployment by developing ecologically valid, workflow\- and latency\-aware in\-the\-wild benchmarks that stress\-test models under real sports conditions\([276](https://arxiv.org/html/2608.14377#bib.bib318);[15](https://arxiv.org/html/2608.14377#bib.bib325)\)\. In parallel, enabling on\-device inference requires efficiency advances such as quantization\([327](https://arxiv.org/html/2608.14377#bib.bib319);[142](https://arxiv.org/html/2608.14377#bib.bib320)\), knowledge distillation\([294](https://arxiv.org/html/2608.14377#bib.bib321)\), and mobile \(M\)LLMs\([155](https://arxiv.org/html/2608.14377#bib.bib322);[257](https://arxiv.org/html/2608.14377#bib.bib323)\)to support local analysis on wearables or mobile platforms\.

## 5Conclusion

This survey reviews the emerging landscape of large models in sports, establishing a structured taxonomy that spans 6 stakeholder groups\. We provide a deep analysis of relevant datasets, and highlight fundamental challenges\. By consolidating these disparate research efforts, we aim to establish a solid framework for future exploration\. We hope this work serves as a foundation for advancing large\-model\-driven sports intelligence and provides a practical resource for research and development\.

## Limitations

Although this survey strives to provide a comprehensive overview of large models in sports, several limitations remain\. Firstly, given the rapid development of this field, our survey may not be able to timely reflect the latest progress before and after the survey\. Secondly, our literature selection primarily follows standard protocols focused on English\-language publications\. This may naturally limit the coverage of domestic research in other regions or studies published in other languages\. Thirdly, our analysis objectively reflects the current research imbalance in the field, which is heavily skewed toward a few dominant sports\. Consequently, this leads to a lack of in\-depth coverage for underrepresented or niche sporting scenarios in our survey\. Fourthly, as some studies span multiple application domains, minor overlaps are inevitable; we categorize each work based on its primary research focus while cross\-referencing related sections when appropriate\. Finally, our analysis primarily centers on academic research, and the discussion of commercial systems or industrial applications remains limited\. Despite these limitations, this survey provides a valuable and timely overview of the field, offering a solid reference for subsequent research and development\.

## Acknowledgments

We thank all reviewers for their insightful comments and suggestions\. This work was partially supported by the Beijing Natural Science Foundation \(No\. L233008\)\.

## References

- M\. Abdin, J\. Aneja, H\. Awadalla, A\. Awadallah, A\. A\. Awan, N\. Bach, A\. Bahree, A\. Bakhtiari, J\. Bao, H\. Behl,et al\.Phi\-3 technical report: a highly capable language model locally on your phone\.arXiv preprint arXiv:2404\.14219\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.15.5)\.
- Achiamet al\.\(2023\)J\. Achiam, S\. Adler, S\. Agarwal, L\. Ahmad, I\. Akkaya, F\. L\. Aleman, D\. Almeida, J\. Altenschmidt, S\. Altman, S\. Anadkat,et al\.Gpt\-4 technical report\.arXiv preprint arXiv:2303\.08774\.Cited by:[§C\.1](https://arxiv.org/html/2608.14377#A3.SS1.p5.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.14.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.17.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.26.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.30.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.8.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.26.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.28.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.36.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.43.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.45.6),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.48.5),[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- Ahsan \(2023\)M\. AhsanChatbot generative pre\-trained transformer and artificial intelligence in sports physical therapy and rehabilitation\.Saudi Journal of Sports Medicine23\(2\),pp\. 61–62\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1)\.
- Akrimiet al\.\(2025\)S\. Akrimi, L\. Schwensfeier, P\. Düking, T\. Kreutz, and C\. BrinkmannChatGPT\-4o\-generated exercise plans for patients with type 2 diabetes mellitus—assessment of their safety and other quality criteria by coaching experts\.Sports13\(4\),pp\. 92\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Andersonet al\.\(2023\)N\. Anderson, D\. L\. Belavy, S\. M\. Perle, S\. Hendricks, L\. Hespanhol, E\. Verhagen, and A\. R\. MemonAI did not write this manuscript, or did it? can we trick the ai text detector into generated texts? the potential future of chatgpt and ai in sports & exercise medicine manuscript generation\.Vol\.9,BMJ Specialist Journals\.Cited by:[§B\.5](https://arxiv.org/html/2608.14377#A2.SS5.p1.1)\.
- Andrewset al\.\(2024a\)P\. Andrews, O\. E\. Nordberg, N\. Borch, F\. Guribye, and M\. FjeldDesigning for automated sports commentary systems\.InProceedings of the 2024 ACM International Conference on Interactive Media Experiences,pp\. 75–93\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1)\.
- Andrewset al\.\(2024b\)P\. Andrews, O\. E\. Nordberg, S\. Zubicueta Portales, N\. Borch, F\. Guribye, K\. Fujita, and M\. FjeldAiCommentator: a multimodal conversational agent for embedded visualization in football viewing\.InProceedings of the 29th International Conference on Intelligent User Interfaces,pp\. 14–34\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.6.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1)\.
- Anguitaet al\.\(2013\)D\. Anguita, A\. Ghio, L\. Oneto, X\. Parra, J\. L\. Reyes\-Ortiz,et al\.A public domain dataset for human activity recognition using smartphones\.\.InEsann,Vol\.3,pp\. 3–4\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.30.1)\.
- Anthropic \(2024\)AnthropicClaude 3 opus model card\.Note:[https://www\.anthropic\.com/claude\-3\-model\-card](https://www.anthropic.com/claude-3-model-card)Accessed: October 5, 2025Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.65.6)\.
- Anthropic \(2025\)AnthropicClaude 3\.5 sonnet\.Note:[https://www\.anthropic\.com/news/claude\-3\-5\-sonnet](https://www.anthropic.com/news/claude-3-5-sonnet)Accessed: October 5, 2025Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.14.5.1)\.
- Argan and Dinç \(2025\)M\. Argan and H\. DinçINVESTIGATING the factors influencing adoption intentions of chatgpt for sport events\.SPORMETRE Beden Eğitimi ve Spor Bilimleri Dergisi23\(2\),pp\. 77–97\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p5.1)\.
- Ashutoshet al\.\(2025\)K\. Ashutosh, T\. Nagarajan, G\. Pavlakos, K\. Kitani, and K\. GraumanExpertAF: expert actionable feedback from video\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 13582–13594\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.9.4)\.
- Baiet al\.\(2022\)Y\. Bai, S\. Kadavath, S\. Kundu, A\. Askell, J\. Kernion, A\. Jones, A\. Chen, A\. Goldie, A\. Mirhoseini, C\. McKinnon,et al\.Constitutional ai: harmlessness from ai feedback\.arXiv preprint arXiv:2212\.08073\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p2.1)\.
- Banet al\.\(2022\)K\. Ban, J\. See, J\. Abdullah, and Y\. P\. LohBadmintondb: a badminton dataset for player\-specific match analysis and prediction\.InProceedings of the 5th international ACM workshop on multimedia content analysis in sports,pp\. 47–54\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.48.1)\.
- Bandraupalliet al\.\(2025\)S\. Bandraupalli, M\. Gajera, A\. A\. Saifee, and A\. ParwarVLMs\-in\-the\-wild: bridging the gap between academic benchmarks and enterprise reality\.In2025 5th International Conference on AI\-ML\-Systems \(AIMLSystems\),pp\. 299–311\.Cited by:[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p5.1),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p5.1)\.
- Banuet al\.\(2025\)S\. Banuet al\.Survey paper on ai based sports highlight generation for social media\.Journal of Scientific Research and Technology,pp\. 30–38\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.31.5)\.
- Baughmanet al\.\(2024\)A\. Baughman, E\. Morales, R\. Agarwal, G\. Akay, R\. Feris, T\. Johnson, S\. Hammer, and L\. KarlinskyLarge scale generative ai text applied to sports and music\.InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining,pp\. 4784–4792\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.9.1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.9.4.1)\.
- Bhatnagar and Bhatnagar \(2025\)M\. Bhatnagar and M\. BhatnagarAnalyzing key factors influencing ipl cricket scores using explainability and multimodal data\.Journal of Quantitative Analysis in Sports21\(3\),pp\. 253–267\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p4.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.51.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.51.4),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p4.1)\.
- Biester \(2025\)L\. BiesterSports and women’s sports: gender bias in text generation with olympic data\.InProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies \(Volume 2: Short Papers\),pp\. 195–205\.Cited by:[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p2.1)\.
- Blacket al\.\(2022\)S\. Black, S\. Biderman, E\. Hallahan, Q\. Anthony, L\. Gao, L\. Golding, H\. He, C\. Leahy, K\. McDonell, J\. Phang,et al\.GPT\-neox\-20b: an open\-source autoregressive language model\.InProceedings of BigScience Episode\# 5–Workshop on Challenges & Perspectives in Creating Large Language Models,pp\. 95–136\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.38.6)\.
- Broglyet al\.\(2025\)C\. Brogly, S\. Rjaibi, C\. Liang, E\. Lam, E\. Wang, S\. Paleczny, A\. Levitan, and M\. D\. CusimanoEvaluation of the phi\-3\-mini slm for identification of texts related to medicine, health, and sports injuries\.In2025 IEEE 4th International Conference on Computing and Machine Intelligence \(ICMI\),pp\. 1–5\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.15.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.15.4)\.
- Brownet al\.\(2020\)T\. Brown, B\. Mann, N\. Ryder, M\. Subbiah, J\. D\. Kaplan, P\. Dhariwal, A\. Neelakantan, P\. Shyam, G\. Sastry, A\. Askell,et al\.Language models are few\-shot learners\.Advances in neural information processing systems33,pp\. 1877–1901\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.40.6),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.17.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.3.6)\.
- Bullardet al\.\(2025\)E\. Bullard, N\. Khan, and N\. SarkarEnhancing athletic performance through ai: an iterative prompt engineering approach for llm\-based coaching feedback\.InInternational Conference on Human\-Computer Interaction,pp\. 251–255\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Bunker and Susnjak \(2022\)R\. Bunker and T\. SusnjakThe application of machine learning techniques for predicting match results in team sport: a review\.Journal of Artificial Intelligence Research73,pp\. 1285–1322\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- CANUZAKOVet al\.\(2025\)K\. CANUZAKOV, C\. ABDIRAHMANOVA, B\. DEMİRHAN,et al\.Digital coaches: an alternative to expert coaches for men’s fitness goals\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Caron and Müller \(2023\)M\. Caron and O\. MüllerTacticalGPT: uncovering the potential of llms for predicting tactical decisions in professional football\.InStatsBomb Conference,pp\. 1–11\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.38.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.38.5),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p3.1)\.
- Cavazzottoet al\.\(2024\)T\. G\. Cavazzotto, D\. B\. Dantas, and M\. R\. QueirogaChatGPT and exercise prescription: human vs\. machine or human plus machine?\.Journal of Sport and Health Science13\(5\),pp\. 661–662\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Chakrabortyet al\.\(2025\)R\. Chakraborty, R\. Chakraborty, A\. Dasgupta, and S\. ChaurasiaDo we need large vlms for spotting soccer actions?\.InThe 14th International Joint Conference on Natural Language Processing and The 4th Conference of the Asia\-Pacific Chapter of the Association for Computational Linguistics,pp\. 59–65\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.28.4),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.8.2)\.
- Chan Changet al\.\(2021\)S\. Chan Chang, R\. Walmsley, J\. Gershuny, T\. Harms, E\. Thomas, K\. Milton, P\. Kelly, C\. Foster, A\. Wong, N\. Gray,et al\.Capture\-24: activity tracker dataset for human activity recognition\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.17.1)\.
- Changet al\.\(2025\)S\. Chang, S\. Chen, and C\. ChangExploring opportunities and challenges toward chatgpt for inclusion in sport education\.Journal of Hospitality, Leisure, Sport & Tourism Education37,pp\. 100572\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1)\.
- Chenet al\.\(2025a\)H\. Chen, H\. Huang, X\. Yin, and D\. ShaoFineQuest: adaptive knowledge\-assisted sports video understanding via agent\-of\-thoughts reasoning\.InProceedings of the 33rd ACM International Conference on Multimedia,pp\. 2909–2918\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.61.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.61.4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.62.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.16.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.17.1),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p6.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.8.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.9.1)\.
- Chenet al\.\(2024a\)J\. Chen, Z\. Lv, S\. Wu, K\. Q\. Lin, C\. Song, D\. Gao, J\. Liu, Z\. Gao, D\. Mao, and M\. Z\. ShouVideollm\-online: online video large language model for streaming video\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 18407–18418\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Chenet al\.\(2024b\)J\. Chen, Z\. Lv, S\. Wu, K\. Q\. Lin, C\. Song, D\. Gao, J\. Liu, Z\. Gao, D\. Mao, and M\. Z\. ShouVideollm\-online: online video large language model for streaming video\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 18407–18418\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.24.5)\.
- Chenet al\.\(2025b\)J\. Chen, Z\. Zeng, Y\. Lin, W\. Li, Z\. Ma, and M\. Z\. ShouLivecc: learning video llm with streaming speech transcription at scale\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 29083–29095\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.18.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.18.4),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.10.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.5.1)\.
- Chenet al\.\(2021\)M\. Chen, J\. Tworek, H\. Jun, Q\. Yuan, H\. P\. D\. O\. Pinto, J\. Kaplan, H\. Edwards, Y\. Burda, N\. Joseph, G\. Brockman,et al\.Evaluating large language models trained on code\.arXiv preprint arXiv:2107\.03374\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.50.5)\.
- Chenet al\.\(2024c\)W\. Chen, Y\. Lin, Y\. Su, W\. Yeh, and L\. KuYourSkatingCoach: a figure skating video benchmark for fine\-grained element analysis\.arXiv preprint arXiv:2410\.20427\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.3.2),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.3.1.1)\.
- Chenet al\.\(2025c\)Y\. Chen, F\. Xue, D\. Li, Q\. Hu, L\. Zhu, X\. Li, Y\. Fang, H\. Tang, S\. Yang, Z\. Liu,et al\.LongVILA: scaling long\-context visual language models for long videos\.InThe Thirteenth International Conference on Learning Representations,Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.29.1),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Chenet al\.\(2024d\)Z\. Chen, J\. Wu, W\. Wang, W\. Su, G\. Chen, S\. Xing, M\. Zhong, Q\. Zhang, X\. Zhu, L\. Lu,et al\.Internvl: scaling up vision foundation models and aligning for generic visual\-linguistic tasks\.InProceedings of the IEEE/CVF conference on computer vision and pattern recognition,pp\. 24185–24198\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.32.6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.7.5),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.11.1)\.
- Chenget al\.\(2023\)K\. Cheng, Q\. Guo, Y\. He, Y\. Lu, R\. Xie, C\. Li, and H\. WuArtificial intelligence in sports medicine: could gpt\-4 make human doctors obsolete?\.Annals of Biomedical Engineering51\(8\),pp\. 1658–1662\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1)\.
- Chenget al\.\(2024a\)L\. Cheng, D\. Deng, X\. Xie, R\. Qiu, M\. Xu, and Y\. WuSNIL: generating sports news from insights with large language models\.IEEE Transactions on Visualization and Computer Graphics\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.35.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p3.1)\.
- Chenget al\.\(2024b\)Z\. Cheng, S\. Leng, H\. Zhang, Y\. Xin, X\. Li, G\. Chen, Y\. Zhu, W\. Zhang, Z\. Luo, D\. Zhao,et al\.Videollama 2: advancing spatial\-temporal modeling and audio understanding in video\-llms\.arXiv preprint arXiv:2406\.07476\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.23.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.23.5)\.
- Chenget al\.\(2025\)Z\. Cheng, J\. Hu, Z\. Liu, C\. Si, W\. Li, and S\. GongV\-star: benchmarking video\-llms on video spatio\-temporal reasoning\.arXiv preprint arXiv:2503\.11495\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.40.1)\.
- Chianget al\.\(2024\)S\. Chiang, L\. Chao, K\. Wang, C\. Wang, and W\. PengBADGE: badminton report generation and evaluation with llm\.arXiv preprint arXiv:2406\.18116\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.36.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p3.1)\.
- Chianget al\.\(2025\)S\. Chiang, T\. Yang, K\. Wang, W\. Wang, A\. Yen, and W\. PengTree\-of\-report: table\-to\-text generation for sports game reports with tree\-structured prompting\.InACL 2025 Student Research Workshop,Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.38.4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.40.1),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p3.1)\.
- Chianget al\.\(2023\)W\. Chiang, Z\. Li, Z\. Lin, Y\. Sheng, Z\. Wu, H\. Zhang, L\. Zheng, S\. Zhuang, Y\. Zhuang, J\. E\. Gonzalez,et al\.Vicuna: an open\-source chatbot impressing gpt\-4 with 90%\* chatgpt quality\.See https://vicuna\. lmsys\. org \(accessed 14 April 2023\)2\(3\),pp\. 6\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.18.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.45.6),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.19.5)\.
- Connor and O’Neill \(2023\)M\. Connor and M\. O’NeillLarge language models in sport science & medicine: opportunities, risks and considerations\.arXiv preprint arXiv:2305\.03851\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p3.1)\.
- Cook and Karakuş \(2024\)A\. Cook and O\. KarakuşLLM\-commentator: novel fine\-tuning strategies of large language models for automatic commentary generation using football event data\.Knowledge\-Based Systems300,pp\. 112219\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.8.4)\.
- Cosentinoet al\.\(2024\)J\. Cosentino, A\. Belyaeva, X\. Liu, Z\. Yang, Y\. Liu, S\. A\. Tailor, T\. Althoff, J\. B\. Hernandez, Y\. Matias, G\. Corrado,et al\.Towards a personal health large language model\.InAdvancements In Medical Foundation Models: Explainability, Robustness, Security, and Beyond,Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.5.4)\.
- Cossichet al\.\(2023\)V\. R\. Cossich, D\. Carlgren, R\. J\. Holash, and L\. KatzTechnological breakthroughs in sport: current practice and future potential of artificial intelligence, virtual reality, augmented reality, and modern data visualization in performance analysis\.Applied Sciences13\(23\),pp\. 12965\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- Cuiet al\.\(2025\)B\. Cui, W\. Jiao, S\. Gui, Y\. Li, and Q\. FangInnovating physical education with artificial intelligence: a potential approach\.Frontiers in Psychology16,pp\. 1490966\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1)\.
- Davidset al\.\(2025\)D\. M\. Davids, A\. A\. E\. Raj, and C\. S\. ChristopherSportSummarizer: a unified multimodal fusion transformer for context\-aware sports video summarization\.Neurocomputing,pp\. 131011\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p2.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.29.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.29.4.1)\.
- Deliegeet al\.\(2021\)A\. Deliege, A\. Cioppa, S\. Giancola, M\. J\. Seikavandi, J\. V\. Dueholm, K\. Nasrollahi, B\. Ghanem, T\. B\. Moeslund, and M\. Van DroogenbroeckSoccernet\-v2: a dataset and benchmarks for holistic understanding of broadcast soccer videos\.InProceedings of the IEEE/CVF conference on computer vision and pattern recognition,pp\. 4508–4519\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.24.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.28.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.29.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.14.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.19.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.54.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.56.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.59.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.6.1.1),[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p2.1)\.
- Denizeet al\.\(2024\)J\. Denize, M\. Liashuha, J\. Rabarisoa, A\. Orcesi, and R\. HéraultCOMEDIAN: self\-supervised learning and knowledge distillation for action spotting using transformers\.InProceedings of the IEEE/CVF Winter Conference on applications of computer vision,pp\. 530–540\.Cited by:[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.7.1)\.
- Dergaaet al\.\(2023\)I\. Dergaa, K\. Chamari, P\. Zmijewski, and H\. B\. SaadFrom human writing to artificial intelligence generated text: examining the prospects and potential threats of chatgpt in academic writing\.Biology of sport40\(2\),pp\. 615–622\.Cited by:[§B\.5](https://arxiv.org/html/2608.14377#A2.SS5.p1.1),[§2\.5](https://arxiv.org/html/2608.14377#S2.SS5.p1.1),[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p2.1)\.
- Dergaaet al\.\(2024\)I\. Dergaa, H\. B\. Saad, A\. El Omri, J\. Glenn, C\. Clark, J\. Washif, N\. Guelmami, O\. Hammouda, R\. Al\-Horani, L\. Reynoso\-Sánchez,et al\.Using artificial intelligence for exercise prescription in personalised health promotion: a critical evaluation of openai’s gpt\-4 model\.Biology of Sport41\(2\),pp\. 221–241\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Dettmerset al\.\(2023\)T\. Dettmers, A\. Pagnoni, A\. Holtzman, and L\. ZettlemoyerQlora: efficient finetuning of quantized llms\.Advances in neural information processing systems36,pp\. 10088–10115\.Cited by:[§2\.3](https://arxiv.org/html/2608.14377#S2.SS3.p1.1)\.
- Devlinet al\.\(2019\)J\. Devlin, M\. Chang, K\. Lee, and K\. ToutanovaBert: pre\-training of deep bidirectional transformers for language understanding\.InProceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 \(long and short papers\),pp\. 4171–4186\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p4.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.46.6),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.47.5.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.7.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.12.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.25.6)\.
- Dibenedettoet al\.\(2025\)G\. Dibenedetto, E\. Musacchio, M\. Polignano, and P\. LopsFine\-tuning large multimodal models for fitness action quality assessment\.InAdjunct Proceedings of the 33rd ACM Conference on User Modeling, Adaptation and Personalization,pp\. 39–44\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.35.4),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1)\.
- Dindorfet al\.\(2025\)C\. Dindorf, J\. Dully, E\. Bartaguiz, T\. Menges, C\. Reidick, J\. Seibert, and M\. FröhlichCharacteristics and perceived suitability of artificial intelligence\-driven sports coaches: a pilot study on psychological and perceptual factors\.Frontiers in Sports and Active Living7,pp\. 1548980\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Dinget al\.\(2025\)X\. Ding, H\. Wu, Y\. Yang, S\. Jiang, Q\. Zhang, D\. Bai, Z\. Chen, and T\. CaoStreammind: unlocking full frame rate streaming video dialogue through event\-gated cognition\.InProceedings of the IEEE/CVF International Conference on Computer Vision,pp\. 13448–13459\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.23.4),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Dosovitskiyet al\.\(2021\)A\. Dosovitskiy, L\. Beyer, A\. Kolesnikov, D\. Weissenborn, X\. Zhai, T\. Unterthiner, M\. Dehghani, M\. Minderer, G\. Heigold, S\. Gelly,et al\.An image is worth 16x16 words: transformers for image recognition at scale\.InInternational Conference on Learning Representations,Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.24.5.1)\.
- Dükinget al\.\(2024\)P\. Düking, B\. Sperlich, L\. Voigt, B\. Van Hooren, M\. Zanini, and C\. ZinnerChatGPT generated training plans for runners are not rated optimal by coaching experts, but increase in quality with additional input information\.Journal of sports science & medicine23\(1\),pp\. 56\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Erol and Arıkan \(2024\)E\. Erol and H\. ArıkanDoes chatgpt provide comprehensive and accurate information regarding the effects, types and programming of core exercises?\.Turkish Journal of Kinesiology10\(3\),pp\. 178–182\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Fayedet al\.\(2023\)A\. M\. Fayed, N\. S\. B\. Mansur, K\. A\. de Carvalho, A\. Behrens, P\. D’Hooghe, and C\. de Cesar NettoArtificial intelligence and chatgpt in orthopaedics and sports medicine\.Journal of Experimental Orthopaedics10\(1\),pp\. 74\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1)\.
- Fazackerleyet al\.\(2025\)L\. A\. Fazackerley, D\. Perrin, and G\. M\. MinettHarnessing generative ai in exercise and sports science education: enhancing real\-world learning and overcoming traditional barriers in data analysis\.Vol\.49,American Physiological Society Rockville, MD\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1)\.
- Felice \(2024\)F\. FeliceAI for handball: predicting and explaining the 2024 olympic games tournament with deep learning and large language models\.arXiv preprint arXiv:2407\.15987\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p4.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.50.4)\.
- Ferrara \(2024\)E\. FerraraLarge language models for wearable sensor\-based human activity recognition, health monitoring, and behavioral modeling: a survey of early trends, datasets, and challenges\.Sensors24\(15\),pp\. 5045\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1)\.
- Fuet al\.\(2025\)C\. Fu, Y\. Dai, Y\. Luo, L\. Li, S\. Ren, R\. Zhang, Z\. Wang, C\. Zhou, Y\. Shen, M\. Zhang,et al\.Video\-mme: the first\-ever comprehensive evaluation benchmark of multi\-modal llms in video analysis\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 24108–24118\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.27.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.31.1)\.
- Gallegoset al\.\(2024\)I\. O\. Gallegos, R\. A\. Rossi, J\. Barrow, M\. M\. Tanjim, S\. Kim, F\. Dernoncourt, T\. Yu, R\. Zhang, and N\. K\. AhmedBias and fairness in large language models: a survey\.Computational Linguistics50\(3\),pp\. 1097–1179\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p2.1)\.
- Gaoet al\.\(2025a\)R\. Gao, X\. Liu, Z\. Hu, B\. Xing, B\. Xia, Z\. Yu, and H\. KälviäinenFsbench: a figure skating benchmark for advancing artistic sports understanding\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 13595–13605\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p2.1),[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.11.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.12.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.13.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.6.1)\.
- Gaoet al\.\(2025b\)X\. Gao, J\. Ruan, J\. Gao, M\. Xie, Z\. Zhang, T\. Liu, and Y\. FuFrom motion signals to insights: a unified framework for student behavior analysis and feedback in physical education classes\.arXiv preprint arXiv:2503\.06525\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.30.4),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p5.1)\.
- Gautamet al\.\(2022\)S\. Gautam, C\. Midoglu, S\. Shafiee Sabet, D\. B\. Kshatri, and P\. HalvorsenSoccer game summarization using audio commentary, metadata, and captions\.InProceedings of the 1st Workshop on User\-centric Narrative Summarization of Long Videos,pp\. 13–22\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.3.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.3.5)\.
- Gautamet al\.\(2025\)S\. Gautam, C\. Midoglu, V\. L\. Thambawita, M\. A\. Riegler, P\. Halvorsen, and M\. ShahSoccerchat: integrating multimodal data for enhanced soccer game understanding\.In2025 International Conference on Content\-Based Multimedia Indexing \(CBMI\),pp\. 1–8\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.53.4.1)\.
- Geet al\.\(2024\)K\. Ge, L\. Chen, K\. Zhang, Y\. Luo, T\. Shi, L\. Fan, X\. Li, G\. Wang, and S\. ZhangSCBench: a sports commentary benchmark for video llms\.arXiv preprint arXiv:2412\.17637\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.7.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.7.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1)\.
- Genç \(2023\)N\. GençArtificial intelligence in physical education and sports: new horizons with chatgpt\.Akdeniz Spor Bilimleri Dergisi6\(1\-Cumhuriyet’in 100\. Yılı Özel Sayısı\),pp\. 17–32\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1)\.
- Giancolaet al\.\(2018\)S\. Giancola, M\. Amine, T\. Dghaily, and B\. GhanemSoccernet: a scalable dataset for action spotting in soccer videos\.InProceedings of the IEEE conference on computer vision and pattern recognition workshops,pp\. 1711–1721\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.31.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.24.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.30.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.48.1)\.
- GLMet al\.\(2024\)T\. GLM, A\. Zeng, B\. Xu, B\. Wang, C\. Zhang, D\. Yin, D\. Zhang, D\. Rojas, G\. Feng, H\. Zhao,et al\.Chatglm: a family of large language models from glm\-130b to glm\-4 all tools\.arXiv preprint arXiv:2406\.12793\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.36.5)\.
- Grattafioriet al\.\(2024\)A\. Grattafiori, A\. Dubey, A\. Jauhri, A\. Pandey, A\. Kadian, A\. Al\-Dahle, A\. Letman, A\. Mathur, A\. Schelten, A\. Vaughan,et al\.The llama 3 herd of models\.arXiv preprint arXiv:2407\.21783\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.22.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.28.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.6.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.9.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.20.5.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.4.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.56.5.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.8.1)\.
- Graumanet al\.\(2024\)K\. Grauman, A\. Westbury, L\. Torresani, K\. Kitani, J\. Malik, T\. Afouras, K\. Ashutosh, V\. Baiyya, S\. Bansal, B\. Boote,et al\.Ego\-exo4d: understanding skilled human activity from first\-and third\-person perspectives\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 19383–19400\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.10.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.9.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.21.1),[§3\.1](https://arxiv.org/html/2608.14377#S3.SS1.p2.1)\.
- Guptaet al\.\(2025\)A\. Gupta, J\. Parmar, I\. R\. Dave, and M\. ShahFrom play to replay: composed video retrieval for temporally fine\-grained videos\.InThe Thirty\-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track,Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.51.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.51.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p6.1)\.
- Haghparastet al\.\(2024\)M\. Haghparast, M\. S\. Hoseini, and D\. N\. EsfahaniA financial management maturity model in sports organizations: a novel approach using artificial intelligence\.Journal of New Studies in Sport Management\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p1.1)\.
- Haghparastet al\.\(2025\)M\. Haghparast, M\. Soltan Hoseini, and D\. Nasr EsfahaniForesight in sports businesses: exploring emerging scenarios based on ai\-language models and financial management strategies\.Sports Business Journal\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p1.1),[§2\.6](https://arxiv.org/html/2608.14377#S2.SS6.p1.1)\.
- Hakamet al\.\(2024\)H\. T\. Hakam, R\. Prill, L\. Korte, B\. Lovreković, M\. Ostojić, N\. Ramadanov, and F\. MuehlensiepenHuman\-written vs ai\-generated texts in orthopedic academic literature: comparative qualitative analysis\.JMIR formative research8,pp\. e52164\.Cited by:[§B\.5](https://arxiv.org/html/2608.14377#A2.SS5.p1.1)\.
- Han \(2025\)Y\. HanIntent\-aware personalized feedback generation from coach\-athlete dialogues in sports training\.Journal of King Saud University Computer and Information Sciences37\(6\),pp\. 1–17\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.7.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.7.4)\.
- Hasnainet al\.\(2023\)M\. Hasnain, B\. Mehboob, and S\. ImranThe role of chatgpt in sports trauma: a mini review on strengths and limits of open ai application\.Discover Artificial Intelligence3\(1\),pp\. 40\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1)\.
- Haverset al\.\(2025\)T\. Havers, L\. Masur, E\. Isenmann, S\. Geisler, C\. Zinner, B\. Sperlich, and P\. DükingReproducibility and quality of hypertrophy\-related training plans generated by gpt\-4 and google gemini as evaluated by coaching experts\.Biology of Sport42\(2\),pp\. 289–329\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Heet al\.\(2025a\)X\. He, W\. Feng, K\. Zheng, Y\. Lu, W\. Zhu, J\. Li, Y\. Fan, J\. Wang, L\. Li, Z\. Yang,et al\.MMWorld: towards multi\-discipline multi\-faceted world model evaluation in videos\.InThe Thirteenth International Conference on Learning Representations,Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.28.1)\.
- Heet al\.\(2025b\)X\. He, W\. Liu, S\. Ma, Q\. Liu, C\. Ma, and J\. WuFinebadminton: a multi\-level dataset for fine\-grained badminton video understanding\.InProceedings of the 33rd ACM International Conference on Multimedia,pp\. 12776–12783\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.14.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.15.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.7.1)\.
- Heet al\.\(2025c\)Y\. He, Y\. Huang, G\. Chen, B\. Pei, J\. Xu, T\. Lu, and J\. PangEgoExoBench: a benchmark for first\-and third\-person view video understanding in mllms\.InThe Thirty\-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track,Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.48.1)\.
- Hegdeet al\.\(2024\)N\. Hegde, M\. Vardhan, D\. Nathani, E\. Rosenzweig, C\. Speed, A\. Karthikesalingam, and M\. SeneviratneInfusing behavior science into large language models for activity coaching\.PLOS Digital Health3\(4\),pp\. e0000431\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.4.4)\.
- Heldet al\.\(2025\)J\. Held, A\. Cioppa, S\. Giancola, E\. Almahmoud, K\. M\. Collins, U\. Bhatt, B\. Ghanem, and M\. Van DroogenbroeckEnhancing football refereeing with\{\\\{ai\}\\\}:\{\\\{vars\}\\\}and\{\\\{x\-vars\}\\\}for assisted decision\-making\.InMathSport Conference,Cited by:[§B\.3](https://arxiv.org/html/2608.14377#A2.SS3.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.53.5)\.
- Heldet al\.\(2023\)J\. Held, A\. Cioppa, S\. Giancola, A\. Hamdi, B\. Ghanem, and M\. Van DroogenbroeckVARS: video assistant referee system for automated soccer decision making from multiple views\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 5086–5097\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.58.1)\.
- Heldet al\.\(2024\)J\. Held, H\. Itani, A\. Cioppa, S\. Giancola, B\. Ghanem, and M\. Van DroogenbroeckX\-vars: introducing explainability in football refereeing with multi\-modal large language models\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 3267–3279\.Cited by:[§B\.3](https://arxiv.org/html/2608.14377#A2.SS3.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.53.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.53.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.53.1),[§2\.3](https://arxiv.org/html/2608.14377#S2.SS3.p1.1),[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p4.1),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p4.1)\.
- Honget al\.\(2026\)J\. Hong, S\. Yan, J\. Cai, X\. Jiang, Y\. Hu, and W\. XieWorldSense: evaluating real\-world omnimodal understanding for multimodal LLMs\.InThe Fourteenth International Conference on Learning Representations,External Links:[Link](https://openreview.net/forum?id=YxsfxAvJv4)Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.50.1)\.
- Honget al\.\(2025\)W\. Hong, Y\. Cheng, Z\. Yang, W\. Wang, L\. Wang, X\. Gu, S\. Huang, Y\. Dong, and J\. TangMotionbench: benchmarking and improving fine\-grained video motion understanding for vision language models\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 8450–8460\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.32.1)\.
- Huet al\.\(2024a\)S\. Hu, X\. Li, X\. Li, J\. Zhang, Y\. Wang, X\. Zhao, and K\. H\. CheongFIOVA: a multi\-annotator benchmark for human\-aligned video captioning\.arXiv preprint arXiv:2410\.15270\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.24.1)\.
- Huet al\.\(2024b\)Y\. Hu, K\. Song, S\. Cho, X\. Wang, H\. Foroosh, D\. Yu, and F\. LiuCan large language models do analytical reasoning?\.arXiv preprint arXiv:2403\.04031\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.41.5)\.
- Huet al\.\(2024c\)Y\. Hu, K\. Song, S\. Cho, X\. Wang, H\. Foroosh, D\. Yu, and F\. LiuSportsMetrics: blending text and numerical data to understand information fusion in LLMs\.InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),L\. Ku, A\. Martins, and V\. Srikumar \(Eds\.\),Bangkok, Thailand,pp\. 267–278\.External Links:[Link](https://aclanthology.org/2024.acl-long.17/),[Document](https://dx.doi.org/10.18653/v1/2024.acl-long.17)Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.41.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.41.5)\.
- Huet al\.\(2024d\)Y\. Hu, K\. Song, S\. Cho, X\. Wang, W\. Yao, H\. Foroosh, D\. Yu, and F\. LiuWhen reasoning meets information aggregation: a case study with sports narratives\.InProceedings of the 2024 conference on empirical methods in natural language processing,pp\. 4293–4308\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.43.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.43.5)\.
- Huanget al\.\(2020\)K\. Huang, C\. Li, and K\. ChangGenerating sports news from live commentary: a chinese dataset for sports game summarization\.InProceedings of the 1st Conference of the Asia\-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing,pp\. 609–615\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.32.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.34.1),[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- Hudl \(2024\)HudlWyscout: football data and analytics platform\.Note:[https://wyscout\.com](https://wyscout.com/)Accessed: October 5, 2025Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.15.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.60.1)\.
- Hurstet al\.\(2024\)A\. Hurst, A\. Lerer, A\. P\. Goucher, A\. Perelman, A\. Ramesh, A\. Clark, A\. Ostrow, A\. Welihinda, A\. Hayes, A\. Radford,et al\.Gpt\-4o system card\.arXiv preprint arXiv:2410\.21276\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.10.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.21.5),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.42.6),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.43.6),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.51.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.16.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.22.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.38.5.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.44.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.49.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.52.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.66.6)\.
- Hutchins \(2016\)B\. HutchinsTales of the digital sublime: tracing the relationship between big data and professional sport\.Convergence22\(5\),pp\. 494–509\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p1.1)\.
- Ibhet al\.\(2024\)M\. Ibh, S\. Graßhof, and D\. W\. HansenA stroke of genius: predicting the next move in badminton\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 3376–3385\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p4.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.47.4.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p4.1)\.
- Imranet al\.\(2024\)S\. A\. Imran, M\. N\. H\. Khan, S\. Biswas, and B\. IslamLlasa: large multimodal agent for human activity analysis through wearable sensors\.arXiv preprint arXiv:2406\.144983\(4\)\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.18.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.18.4),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p3.1)\.
- Janssenset al\.\(2024\)B\. Janssens, M\. Bogaert, and S\. VerstocktLarge language models on race commentary: towards granular data in cycling analytics\.InInternational Workshop on Machine Learning and Data Mining for Sports Analytics,pp\. 14–25\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.42.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.42.5),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p3.1)\.
- Jardimet al\.\(2023\)P\. C\. Jardim, L\. M\. P\. Moraes, and C\. D\. AguiarQasports: a question answering dataset about sports\.InDataset Showcase Workshop \(DSW\),pp\. 1–12\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.3.1)\.
- Jiet al\.\(2024\)S\. Ji, X\. Zheng, and C\. WuHargpt: are llms zero\-shot human activity recognizers?\.In2024 IEEE International Workshop on Foundation Models for Cyber\-Physical Systems & Internet of Things \(FMSys\),pp\. 38–43\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.17.4)\.
- Jiet al\.\(2023\)Y\. Ji, L\. Ye, H\. Huang, L\. Mao, Y\. Zhou, and L\. GaoLocalization\-assisted uncertainty score disentanglement network for action quality assessment\.InProceedings of the 31st ACM International Conference on Multimedia,pp\. 8590–8597\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.34.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.10.1.1)\.
- Jianget al\.\(2023\)A\. Q\. Jiang, A\. Sablayrolles, A\. Mensch, C\. Bamford, D\. S\. Chaplot, D\. de las Casas, F\. Bressand, G\. Lengyel, G\. Lample, L\. Saulnier, L\. R\. Lavaud, M\. Lachaux, P\. Stock, T\. L\. Scao, T\. Lavril, T\. Wang, T\. Lacroix, and W\. E\. SayedMistral 7b\.External Links:2310\.06825,[Link](https://arxiv.org/abs/2310.06825)Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.50.5)\.
- Jianget al\.\(2025a\)D\. Jiang, R\. Zhang, Z\. Guo, Y\. Wu, P\. Qiu, P\. Lu, Z\. Chen, G\. Song, P\. Gao, Y\. Liu,et al\.Mmsearch: unveiling the potential of large models as multi\-modal search engines\.InThe Thirteenth International Conference on Learning Representations,Cited by:[§C\.2\.3](https://arxiv.org/html/2608.14377#A3.SS2.SSS3.p1.1)\.
- Jianget al\.\(2025b\)T\. Jiang, H\. Wang, M\. S\. Salekin, P\. Atighehchian, and S\. ZhangDomain adaptation of vlm for soccer video understanding\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 6111–6121\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.29.4),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.14.4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.59.4.1)\.
- Jörkeet al\.\(2025\)M\. Jörke, S\. Sapkota, L\. Warkenthien, N\. Vainio, P\. Schmiedmayer, E\. Brunskill, and J\. A\. LandayGPTCoach: towards llm\-based physical activity coaching\.InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems,pp\. 1–46\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Kanget al\.\(2025\)J\. Kang, S\. Kwon, J\. Lee, and B\. KimDIAMOND: an llm\-driven agent for context\-aware baseball highlight summarization\.InProceedings of the 1st Workshop for Research on Agent Language Models \(REALM 2025\),pp\. 386–400\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p2.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.27.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.27.4)\.
- Karatet al\.\(2025\)A\. Karat, A\. Tibrewal, N\. Kotian, M\. Dang, R\. Valluri, A\. Ravi Teja Marineni, S\. Sahni, R\. Sundaresan, A\. Kumar, A\. Mehndiratta,et al\.A system for triggering sports instant answers on search engines\.InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval,pp\. 4304–4308\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.49.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.49.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p6.1)\.
- Kauppinen \(2024\)A\. KauppinenProactive autonomous assignments as pedagogical responses to the rise of artificial intelligence solutions in sport management teaching practice\.Sport Management Education Journal19\(1\),pp\. 54–58\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1)\.
- Keiperet al\.\(2023\)M\. C\. Keiper, G\. Fried, J\. Lupinek, and H\. NordstromArtificial intelligence in sport management education: playing the ai game with chatgpt\.Journal of Hospitality, Leisure, Sport & Tourism Education33,pp\. 100456\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1)\.
- Kimet al\.\(2025\)K\. Kim, H\. Song, J\. Ryu, C\. Oh, and B\. SuhBleacherBot: ai agent as a sports co\-viewing partner\.InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems,pp\. 1–31\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1)\.
- Koet al\.\(2025\)K\. Ko, M\. Oh, M\. Seong, and S\. KimLEGOLAS: learning & enhancing golf skills through llm\-augmented system\.InProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems,pp\. 1–10\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Kodathalaet al\.\(2025\)S\. V\. Kodathala, Y\. R\. Vutukoori, and R\. VunnamSV3\. 3b: a sports video understanding model for action recognition\.arXiv preprint arXiv:2507\.17844\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.22.4)\.
- Kohet al\.\(2021\)P\. W\. Koh, S\. Sagawa, H\. Marklund, S\. M\. Xie, M\. Zhang, A\. Balsubramani, W\. Hu, M\. Yasunaga, R\. L\. Phillips, I\. Gao,et al\.Wilds: a benchmark of in\-the\-wild distribution shifts\.InInternational conference on machine learning,pp\. 5637–5664\.Cited by:[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p5.1)\.
- Konget al\.\(2025a\)F\. Kong, J\. Zhang, H\. Zhang, S\. Feng, D\. Wang, L\. Yu, X\. Ji, Y\. Tian, V\. W\., and F\. ZhangTUNA: comprehensive fine\-grained temporal understanding evaluation on dense dynamic videos\.InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),W\. Che, J\. Nabende, E\. Shutova, and M\. T\. Pilehvar \(Eds\.\),Vienna, Austria,pp\. 1810–1839\.External Links:[Link](https://aclanthology.org/2025.acl-long.91/),[Document](https://dx.doi.org/10.18653/v1/2025.acl-long.91),ISBN 979\-8\-89176\-251\-0Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.39.1)\.
- Konget al\.\(2025b\)F\. Kong, W\. Zu, X\. Chen, Y\. Yang, S\. Zhu, and X\. FengSIV\-bench: a video benchmark for social interaction understanding and reasoning\.arXiv preprint arXiv:2506\.05425\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.45.1)\.
- Laiet al\.\(2025\)X\. Lai, J\. Chen, Y\. Lai, S\. Huang, Y\. Cai, Z\. Sun, X\. Wang, K\. Pan, Q\. Gao, and C\. HuangUsing large language models to enhance exercise recommendations and physical activity in clinical and healthy populations: scoping review\.JMIR Medical Informatics13\(1\),pp\. e59309\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p3.1)\.
- Latzel and Glauner \(2024\)R\. Latzel and P\. GlaunerArtificial intelligence in sport scientific creation and writing process\.InArtificial Intelligence in Sports, Movement, and Health,pp\. 15–29\.Cited by:[§B\.5](https://arxiv.org/html/2608.14377#A2.SS5.p1.1),[§2\.5](https://arxiv.org/html/2608.14377#S2.SS5.p1.1)\.
- Ledermanet al\.\(2025\)O\. Lederman, A\. Llana, J\. Murray, R\. Stanton, R\. Chugh, D\. Haywood, A\. Burdett, G\. Warman, J\. Walker, and N\. H\. HartPromises and perils of generative artificial intelligence: a narrative review informing its ethical and practical applications in clinical exercise physiology\.BMC Sports Science, Medicine and Rehabilitation17\(1\),pp\. 131\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Leeet al\.\(2024\)C\. Lee, T\. Lin, H\. Pfister, and C\. Zhu\-TianSportify: question answering with embedded visualizations and personified narratives for sports video\.IEEE Transactions on Visualization and Computer Graphics\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.42.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.42.4)\.
- Leeet al\.\(2025\)J\. Lee, Y\. Yu, and D\. LeeHIPPO\-video: simulating watch histories with large language models for history\-driven video highlighting\.InSecond Conference on Language Modeling,Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p2.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.28.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.28.4)\.
- Leeet al\.\(2020\)Y\. Lee, H\. Jung, C\. Yang, and J\. LeeHighlight\-video generation system for baseball games\.In2020 IEEE International Conference on Consumer Electronics\-Asia \(ICCE\-Asia\),pp\. 1–4\.Cited by:[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p2.1)\.
- Liet al\.\(2025a\)B\. Li, Y\. Zhang, D\. Guo, R\. Zhang, F\. Li, H\. Zhang, K\. Zhang, P\. Zhang, Y\. Li, Z\. Liu,et al\.LLaVA\-onevision: easy visual task transfer\.Transactions on Machine Learning Research\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.27.5)\.
- Liet al\.\(2025b\)C\. Li, S\. Padmanabhuni, M\. S\. Cheema, H\. Seifi, and P\. FazliVideoa11y: method and dataset for accessible video description\.InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems,pp\. 1–29\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.38.1)\.
- Liet al\.\(2024a\)F\. Li, R\. Zhang, H\. Zhang, Y\. Zhang, B\. Li, W\. Li, Z\. Ma, and C\. LiLlava\-interleave: tackling multi\-image, video, and 3d in large multimodal models\.InThe Thirteenth International Conference on Learning Representations,Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.29.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.59.5.1)\.
- Liet al\.\(2025c\)G\. Li, H\. Li, Y\. Su, Y\. Li, S\. Jiang, and G\. ZhangGPT\-4 as a virtual fitness coach: a case study assessing its effectiveness in providing weight loss and fitness guidance\.BMC Public Health25\(1\),pp\. 2466\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Liet al\.\(2026a\)H\. Li, A\. Deng, J\. Liu, H\. Rahmani, Y\. Guo, B\. Schiele, M\. Bennamoun, and Q\. KeSports\-qa: a large\-scale video question answering benchmark for complex and professional sports\.International Journal of Computer Vision134\(5\),pp\. 196\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.18.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.10.1)\.
- Liet al\.\(2022\)J\. Li, D\. Li, C\. Xiong, and S\. HoiBlip: bootstrapping language\-image pre\-training for unified vision\-language understanding and generation\.InInternational conference on machine learning,pp\. 12888–12900\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.51.5)\.
- Liet al\.\(2025d\)X\. Li, Y\. He, S\. Zu, Z\. Li, T\. Shi, Y\. Xie, and K\. ZhangMulti\-modal large language model with rag strategies in soccer commentary generation\.In2025 IEEE/CVF Winter Conference on Applications of Computer Vision \(WACV\),pp\. 6197–6206\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.19.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.15.1)\.
- Liet al\.\(2026b\)X\. Li, X\. Li, S\. Hu, K\. Huang, and W\. ZhangCausalstep: a benchmark for explicit stepwise causal reasoning in videos\.InProceedings of the AAAI Conference on Artificial Intelligence,Vol\.40,pp\. 6530–6538\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.52.1)\.
- Liet al\.\(2025e\)Y\. Li, C\. Tang, J\. Zhuang, Y\. Yang, G\. Sun, W\. Li, Z\. MA, and C\. ZhangImproving llm video understanding with 16 frames per second\.InForty\-second International Conference on Machine Learning,Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.27.4),[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p3.1),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Liet al\.\(2024b\)Y\. Li, X\. Chen, B\. Hu, L\. Wang, H\. Shi, and M\. ZhangVideovista: a versatile benchmark for video understanding and reasoning\.arXiv preprint arXiv:2406\.11303\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.23.1)\.
- Liet al\.\(2025f\)Z\. Li, Y\. Lai, C\. Lin, C\. Chou, and P\. HanVisualizing exercise data from combat exergame for exploring the insight from personal informatics with large language models\.InProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems,pp\. 1–8\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Linet al\.\(2024a\)B\. Lin, Y\. Ye, B\. Zhu, J\. Cui, M\. Ning, P\. Jin, and L\. YuanVideo\-llava: learning united visual representation by alignment before projection\.InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing,pp\. 5971–5984\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.61.5.1)\.
- Linet al\.\(2024b\)J\. Lin, J\. Tang, H\. Tang, S\. Yang, W\. Chen, W\. Wang, G\. Xiao, X\. Dang, C\. Gan, and S\. HanAwq: activation\-aware weight quantization for on\-device llm compression and acceleration\.Proceedings of machine learning and systems6,pp\. 87–100\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p5.1)\.
- Linet al\.\(2025\)T\. Lin, R\. Xiang, G\. Liu, D\. Tiwari, M\. Chiang, C\. Ye, H\. Pfister, and C\. Zhu\-TianSportsBuddy: designing and evaluating an ai\-powered sports video storytelling tool through real\-world deployment\.In2025 IEEE 18th Pacific Visualization Conference \(PacificVis\),pp\. 214–223\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.44.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.44.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p4.1)\.
- Linet al\.\(2024c\)X\. Lin, A\. Zare, S\. Huang, M\. Yang, S\. Chang, and L\. ZhangPersonalized video comment generation\.InFindings of the Association for Computational Linguistics: EMNLP 2024,pp\. 16806–16820\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- Lingruiet al\.\(2025\)X\. Lingrui, L\. Mandi, and Z\. LeiTacticExpert: spatial\-temporal graph language model for basketball tactics\.arXiv preprint arXiv:2503\.10722\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.45.5)\.
- Liuet al\.\(2024a\)A\. Liu, B\. Feng, B\. Xue, B\. Wang, B\. Wu, C\. Lu, C\. Zhao, C\. Deng, C\. Zhang, C\. Ruan,et al\.Deepseek\-v3 technical report\.arXiv preprint arXiv:2412\.19437\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.55.5)\.
- Liuet al\.\(2023\)H\. Liu, C\. Li, Q\. Wu, and Y\. J\. LeeVisual instruction tuning\.Advances in neural information processing systems36,pp\. 34892–34916\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.20.6)\.
- Liuet al\.\(2024b\)Y\. Liu, Z\. Ma, Z\. Qi, Y\. Wu, Y\. Shan, and C\. W\. ChenEt bench: towards open\-ended event\-level video\-language understanding\.Advances in Neural Information Processing Systems37,pp\. 32076–32110\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.22.1),[§3\.1](https://arxiv.org/html/2608.14377#S3.SS1.p2.1)\.
- Liuet al\.\(2019\)Y\. Liu, M\. Ott, N\. Goyal, J\. Du, M\. Joshi, D\. Chen, O\. Levy, M\. Lewis, L\. Zettlemoyer, and V\. StoyanovRoberta: a robustly optimized bert pretraining approach\.arXiv preprint arXiv:1907\.11692\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.31.6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.46.5)\.
- Liuet al\.\(2025\)Z\. Liu, K\. Jiang, M\. Ma, Z\. Hou, Y\. Lin, and J\. S\. DongF3\{\}^\{3\}set: towards analyzing fast, frequent, and fine\-grained events from videos\.InThe Thirteenth International Conference on Learning Representations,Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.26.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.26.4)\.
- Liuet al\.\(2024c\)Z\. Liu, X\. Xie, M\. He, W\. Zhao, Y\. Wu, L\. Cheng, H\. Zhang, and Y\. WuSmartboard: visual exploration of team tactics with llm agent\.IEEE Transactions on Visualization and Computer Graphics\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.39.5),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p3.1)\.
- LLC \(2025\)S\. LLCSportVU\.Note:[https://www\.stats\.com/sportvu\-basketball/](https://www.stats.com/sportvu-basketball/)Accessed: October 5, 2025Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.39.2)\.
- Lotfi and Madani \(2024\)N\. Lotfi and M\. MadaniEvaluating the qualitative and quantitative performance of generative ai on knowledge in sports medicine: the case of gpt\.InGeneral Aspects of Applying Generative AI in Higher Education: Opportunities and Challenges,pp\. 103–119\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1)\.
- Love \(2015\)C\. LoveBBC football commentary data – webscraping\.Note:[https://sciolisticramblings\.wordpress\.com/2015/08/24/bbc\-football\-commentary\-data\-webscraping/](https://sciolisticramblings.wordpress.com/2015/08/24/bbc-football-commentary-data-webscraping/)Accessed: October 5, 2025Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.8.1)\.
- Luet al\.\(2024\)Z\. Lu, X\. Li, D\. Cai, R\. Yi, F\. Liu, X\. Zhang, N\. D\. Lane, and M\. XuSmall language models: survey, measurements, and insights\.arXiv preprint arXiv:2409\.15790\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p5.1)\.
- Maet al\.\(2025a\)J\. Ma, J\. Wu, H\. Wang, Y\. Zhang, X\. Xie, Z\. Zhou, H\. Zhang, J\. Wang, and Y\. WuT3Set: a multimodal dataset with targeted suggestions for llm\-based virtual coach in table tennis training\.InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V\. 2,pp\. 5686–5697\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.6.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.6.4)\.
- Maet al\.\(2025b\)W\. Ma, Y\. Liu, Q\. Yi, X\. Liu, W\. Xing, R\. Zhao, H\. Liu, and R\. LiTable tennis coaching system based on a multimodal large language model with a table tennis knowledge base\.PloS one20\(2\),pp\. e0317839\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.8.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.8.4)\.
- Maazet al\.\(2024\)M\. Maaz, H\. Rasheed, S\. Khan, and F\. KhanVideo\-chatgpt: towards detailed video understanding via large vision and language models\.InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),pp\. 12585–12602\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.53.6)\.
- Madanet al\.\(2025\)S\. Madan, S\. Ghosh, L\. R\. Sookha, M\. Ganaie, R\. Subramanian, A\. Dhall, and T\. GedeonMIP\-gaf: a mllm\-annotated benchmark for most important person localization and group context understanding\.In2025 IEEE/CVF Winter Conference on Applications of Computer Vision \(WACV\),pp\. 1467–1476\.Cited by:[§C\.2\.3](https://arxiv.org/html/2608.14377#A3.SS2.SSS3.p1.1)\.
- Martire and Ragazzi \(2025\)F\. A\. Martire and D\. RagazziLeveraging llms and rag for enhanced football talent scouting\.InInternational Conference on Advanced Information Systems Engineering,pp\. 298–309\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p2.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.66.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.66.5),[§2\.6](https://arxiv.org/html/2608.14377#S2.SS6.p2.1)\.
- Masagca \(2025\)R\. C\. MasagcaThe ai coach: a 5\-week ai\-generated calisthenics training program on health\-related physical fitness components of untrained collegiate students\.Journal of Human Sport and Exercise20\(1\),pp\. 39–56\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Masuret al\.\(2025\)L\. Masur, M\. Driller, H\. Suppiah, M\. Matzka, B\. Sperlich, and P\. DükingAssessment of recommendations provided to athletes regarding sleep education by gpt\-4o and google gemini: comparative evaluation study\.JMIR Formative Research9\(1\),pp\. e71358\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p3.1)\.
- Mateuset al\.\(2024\)N\. Mateus, E\. Abade, D\. Coutinho, M\. Gómez, C\. L\. Peñas, and J\. SampaioEmpowering the sports scientist with artificial intelligence in training, performance, and health management\.Sensors25\(1\),pp\. 139\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p2.1),[§2\.6](https://arxiv.org/html/2608.14377#S2.SS6.p2.1)\.
- McBeeet al\.\(2023\)J\. C\. McBee, D\. Y\. Han, L\. Liu, L\. Ma, D\. A\. Adjeroh, D\. Xu, and G\. HuInterdisciplinary inquiry via panelgpt: application to explore chatbot application in sports rehabilitation\.medRxiv\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1)\.
- McBeeet al\.\(2024\)J\. C\. McBee, D\. Y\. Han, L\. Liu, L\. Ma, D\. A\. Adjeroh, D\. Xu, and G\. HuAssessing chatgpt’s competency in addressing interdisciplinary inquiries on chatbot uses in sports rehabilitation: simulation study\.JMIR Medical Education10\(1\),pp\. e51157\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p2.1)\.
- Memonet al\.\(2025\)S\. B\. Memon, J\. A\. Qureshi, and S\. B\. ShahAI\-powered chatgpt in sports tourism: benefits, challenges, and future prospects\.Redefining Tourism With AI and the Metaverse,pp\. 163–188\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p3.1),[§2\.6](https://arxiv.org/html/2608.14377#S2.SS6.p3.1)\.
- Merilehto \(2024\)J\. MerilehtoFrom pdfs to structured data: utilizing llm analysis in sports database management\.arXiv preprint arXiv:2410\.17619\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.65.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.65.5),[§2\.6](https://arxiv.org/html/2608.14377#S2.SS6.p1.1)\.
- Merrillet al\.\(2026\)M\. A\. Merrill, A\. Paruchuri, N\. Rezaei, G\. Kovacs, J\. Perez, Y\. Liu, E\. Schenck, N\. Hammerquist, J\. Sunshine, S\. Tailor,et al\.Transforming wearable data into personal health insights using large language model agents\.Nature Communications\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.16.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.16.5),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p3.1)\.
- Mershaet al\.\(2024\)M\. Mersha, K\. Lam, J\. Wood, A\. K\. Alshami, and J\. KalitaExplainable artificial intelligence: a survey of needs, techniques, applications, and future direction\.Neurocomputing599,pp\. 128111\.Cited by:[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p4.1)\.
- Methnaniet al\.\(2023\)J\. Methnani, I\. Latiri, I\. Dergaa, K\. Chamari, and H\. B\. SaadChatGPT for sample\-size calculation in sports medicine and exercise sciences: a cautionary note\.International Journal of Sports Physiology and Performance18\(10\),pp\. 1219–1223\.Cited by:[§B\.5](https://arxiv.org/html/2608.14377#A2.SS5.p1.1),[§2\.5](https://arxiv.org/html/2608.14377#S2.SS5.p1.1)\.
- Michielssenet al\.\(2024\)S\. Michielssen, A\. Maloof, J\. Haumacher, A\. Dreger, K\. Bonicki, and K\. HallgrenUsing large language models to generate baseball spray charts in the absence of numerical data\.Proceedings of the Institution of Mechanical Engineers, Part P: Journal of Sports Engineering and Technology,pp\. 17543371241257734\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.40.5),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p3.1)\.
- Midogluet al\.\(2024\)C\. Midoglu, S\. S\. Sabet, M\. H\. Sarkhoosh, M\. Majidi, S\. Gautam, H\. M\. Solberg, T\. Kupka, and P\. HalvorsenAi\-based sports highlight generation for social media\.InProceedings of the 3rd Mile\-High Video Conference,pp\. 7–13\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p2.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.26.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.26.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p2.1)\.
- Mkhallatiet al\.\(2023\)H\. Mkhallati, A\. Cioppa, S\. Giancola, B\. Ghanem, and M\. Van DroogenbroeckSoccerNet\-caption: dense video captioning for soccer broadcasts commentaries\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 5074–5085\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.13.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.23.1),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.14.1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.16.1)\.
- Moriet al\.\(2025\)Y\. Mori, C\. Tanaka, A\. Maekawa, S\. Kosugi, T\. Ishigaki, K\. Funakoshi, H\. Takamura, and M\. OkumuraLive football commentary system providing background information\.InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics \(Volume 3: System Demonstrations\),pp\. 394–404\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.16.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.16.4)\.
- Musatet al\.\(2024\)C\. L\. Musat, C\. Mereuta, A\. Nechita, D\. Tutunaru, A\. E\. Voipan, D\. Voipan, E\. Mereuta, T\. V\. Gurau, G\. Gurău, and L\. C\. NechitaDiagnostic applications of ai in sports: a comprehensive review of injury risk prediction methods\.Diagnostics14\(22\),pp\. 2516\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1)\.
- Nagraniet al\.\(2025\)A\. Nagrani, S\. Menon, A\. Iscen, S\. Buch, R\. Mehran, N\. Jha, A\. Hauth, Y\. Zhu, C\. Vondrick, M\. Sirotenko,et al\.Minerva: evaluating complex video reasoning\.InProceedings of the IEEE/CVF International Conference on Computer Vision,pp\. 23968–23978\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.41.1)\.
- Nagraniet al\.\(2024\)A\. Nagrani, M\. Zhang, R\. Mehran, R\. Hornung, N\. B\. Gundavarapu, N\. Jha, A\. Myers, X\. Zhou, B\. Gong, C\. Schmid,et al\.Neptune: the long orbit to benchmarking long video understanding\.arXiv preprint arXiv:2412\.09582\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.25.1)\.
- Naiket al\.\(2022\)B\. T\. Naik, M\. F\. Hashmi, and N\. D\. BokdeA comprehensive review of computer vision in sports: open issues, future trends and research directions\.Applied Sciences12\(9\),pp\. 4429\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- National Strength and Conditioning Association \(NSCA\) \(2025\)National Strength and Conditioning Association \(NSCA\)Certified strength and conditioning specialist \(cscs\) exam description\.Note:[https://www\.nsca\.com/certification/cscs/certified\-strength\-and\-conditioning\-specialist\-exam\-description](https://www.nsca.com/certification/cscs/certified-strength-and-conditioning-specialist-exam-description)Accessed: October 5, 2025Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.5.1)\.
- Naughtonet al\.\(2024\)M\. Naughton, P\. M\. Salmon, H\. R\. Compton, and S\. McLeanChallenges and opportunities of artificial intelligence implementation within sports science and sports medicine teams\.Frontiers in Sports and Active Living6,pp\. 1332427\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p3.1)\.
- Niuet al\.\(2025\)J\. Niu, Y\. Li, Z\. Miao, C\. Ge, Y\. Zhou, Q\. He, X\. Dong, H\. Duan, S\. Ding, R\. Qian,et al\.OVO\-bench: how far is your video\-llms from real\-world online video understanding?\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 18902–18913\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.33.1),[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p5.1)\.
- Nonakaet al\.\(2024\)N\. Nonaka, R\. Fujihira, T\. Koshiba, A\. Maeda, and J\. SeitaRugby scene classification enhanced by vision language model\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 3256–3266\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.20.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.20.5)\.
- Oliver and Guiller \(2025\)A\. Oliver and J\. GuillerGenerative ai in sport and exercise psychology: exploring opportunities and overcoming challenges\.Sport and Exercise Psychology Review19\(2\),pp\. 36–45\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1)\.
- Onanet al\.\(2025\)D\. Onan, H\. Arıkan, İ\. Can, Ş\. Güven, L\. Işıkay, and A\. OzgeExamining the ability of artificial intelligence with chatgpt\-4\.0 to create an exercise program: case scenario examples" lumbar disc herniation, chronic migraine, and urge urinary incontinence"\.Turkish Journal of Kinesiology11\(1\),pp\. 28–44\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1)\.
- OpenAI \(2024\)OpenAIGPT\-4v\(ision\) system card\.Note:[https://cdn\.openai\.com/papers/GPTV\_System\_Card\.pdf?utm\_source=chatgpt\.com](https://cdn.openai.com/papers/GPTV_System_Card.pdf?utm_source=chatgpt.com)Accessed: October 6, 2025Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.39.6)\.
- Ouyanget al\.\(2022\)L\. Ouyang, J\. Wu, X\. Jiang, D\. Almeida, C\. Wainwright, P\. Mishkin, C\. Zhang, S\. Agarwal, K\. Slama, A\. Ray,et al\.Training language models to follow instructions with human feedback\.Advances in neural information processing systems35,pp\. 27730–27744\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.44.6),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.49.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.35.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.47.6),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.6.5)\.
- Ovedet al\.\(2020\)N\. Oved, A\. Feder, and R\. ReichartPredicting in\-game actions from interviews of nba players\.Computational Linguistics46\(3\),pp\. 667–712\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p4.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.46.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.46.5)\.
- Pageet al\.\(2021\)M\. J\. Page, J\. E\. McKenzie, P\. M\. Bossuyt, I\. Boutron, T\. C\. Hoffmann, C\. D\. Mulrow, L\. Shamseer, J\. M\. Tetzlaff, E\. A\. Akl, S\. E\. Brennan,et al\.The prisma 2020 statement: an updated guideline for reporting systematic reviews\.bmj372\.Cited by:[Appendix A](https://arxiv.org/html/2608.14377#A1.p1.1),[§1](https://arxiv.org/html/2608.14377#S1.p4.1)\.
- Pajoet al\.\(2025\)L\. S\. Pajo, R\. Rabuya, A\. Andacao, A\. M\. S\. Tuano, and J\. LoboA 10\-week large language model\-generated \(llm\) versus human\-made volleyball training program on the jumping performance of collegiate volleyball athletes\.Journal of Physical Education36,pp\. e3611\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Papiniet al\.\(2025a\)N\. M\. Papini, M\. Meyer, N\. D\. Squires, and D\. CliffordBalancing act: generative ai tools and scope of practice in health coaching\.American Journal of Health Promotion,pp\. 08901171251340383\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Papiniet al\.\(2025b\)N\. M\. Papini, M\. Meyer, N\. D\. Squires, and D\. CliffordBalancing act: generative ai tools and scope of practice in health coaching\.American Journal of Health Promotion,pp\. 08901171251340383\.Cited by:[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p2.1)\.
- Parmaret al\.\(2022\)P\. Parmar, A\. Gharat, and H\. RhodinDomain knowledge\-informed self\-supervised representations for workout form assessment\.InEuropean Conference on Computer Vision,pp\. 105–123\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.35.1)\.
- Patel \(2022\)S\. PatelAn api client package to access the apis for nba\.com\.Note:[https://github\.com/swar/nba\_api](https://github.com/swar/nba_api)Accessed: October 5, 2023Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.35.1)\.
- Pavlovichet al\.\(2023\)R\. V\. Pavlovich, E\. A\. Tsybulko, K\. N\. Zhigunov, A\. V\. Khelvas, A\. A\. Gilya\-Zetinov, and I\. V\. TykhonovSoccer artificial intelligence commentary service on the base of video analytic and large language models\.In2023 31st Telecommunications Forum \(TELFOR\),pp\. 1–4\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1)\.
- Petti and Gilani \(2021\)B\. Petti and S\. GilaniBaseballr: the sports dataverse’s r package for baseball data\.Note:[https://billpetti\.github\.io/baseballr/](https://billpetti.github.io/baseballr/)Accessed: October 5, 2025Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.40.2)\.
- Phillips and Kennedy \(2012\)E\. M\. Phillips and M\. A\. KennedyThe exercise prescription: a tool to improve physical activity\.Pm&r4\(11\),pp\. 818–825\.Cited by:[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1)\.
- Philueket al\.\(2025\)P\. Philuek, S\. Kusump, T\. Sathianpoonsook, C\. Jansupom, P\. Sawanyawisuth, K\. Sawanyawisuth, and A\. ChainarongThe effects of chat gpt generated exercise program in healthy overweight young adults: a pilot study\.Journal of Human Sport and Exercise20\(1\),pp\. 169–179\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Priyaet al\.\(2024\)M\. Y\. Priya, S\. S\. Kamble, S\. P\. Shendre, S\. Sridhar,et al\.Megan\-a sports chatbot using openai apis and django framework with python\.In2024 IEEE 9th International Conference for Convergence in Technology \(I2CT\),pp\. 1–8\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1)\.
- Puceet al\.\(2025\)L\. Puce, N\. L\. Bragazzi, A\. Curra, and C\. TrompettoHarnessing generative artificial intelligence for exercise and training prescription: applications and implications in sports and physical activity—a systematic literature review\.\.Applied Sciences \(2076\-3417\)15\(7\)\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Puceet al\.\(2024\)L\. Puce, H\. İ\. Ceylan, C\. Trompetto, F\. Cotellessa, C\. Schenone, L\. Marinelli, P\. Zmijewski, N\. Bragazzi, and L\. MoriOptimizing athletic performance through advanced nutrition strategies: can ai and digital platforms have a role in ultraendurance sports?\.Biology of Sport41\(4\),pp\. 305–313\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Puduppullyet al\.\(2019\)R\. Puduppully, L\. Dong, and M\. LapataData\-to\-text generation with entity modeling\.InProceedings of the 57th Annual Meeting of the Association for Computational Linguistics,pp\. 2023–2035\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.39.1)\.
- Qianet al\.\(2025a\)T\. Y\. Qian, W\. Li, H\. Gong, C\. Seifried, and C\. XuExperience is all you need: a large language model application of fine\-tuned gpt\-3\.5 and roberta for aspect\-based sentiment analysis of college football stadium reviews\.Sport Management Review28\(1\),pp\. 1–25\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p5.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.46.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.46.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p5.1)\.
- Qianet al\.\(2025b\)T\. Y\. Qian, B\. Yu, W\. Li, and C\. XuEsports’ debut as a medal event at 2023 asian games: exploring public perceptions with bertopic and gpt\-4 topic fine\-tuning\.In58th Hawaii International Conference on System Sciences, HICSS 2025,pp\. 4303–4312\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p5.1)\.
- Qiu \(2024\)Y\. QiuThe impact of llm hallucinations on motor skill learning: a case study in badminton\.IEEE Access\.Cited by:[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p4.1)\.
- Radfordet al\.\(2021\)A\. Radford, J\. W\. Kim, C\. Hallacy, A\. Ramesh, G\. Goh, S\. Agarwal, G\. Sastry, A\. Askell, P\. Mishkin, J\. Clark,et al\.Learning transferable visual models from natural language supervision\.InInternational conference on machine learning,pp\. 8748–8763\.Cited by:[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.6.3)\.
- Raffelet al\.\(2020\)C\. Raffel, N\. Shazeer, A\. Roberts, K\. Lee, S\. Narang, M\. Matena, Y\. Zhou, W\. Li, and P\. J\. LiuExploring the limits of transfer learning with a unified text\-to\-text transformer\.Journal of machine learning research21\(140\),pp\. 1–67\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.11.5.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.3.6),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.10.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.41.6),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.3.3)\.
- Raoet al\.\(2025a\)J\. Rao, Z\. Li, H\. Wu, Y\. Zhang, Y\. Wang, and W\. XieMulti\-agent system for comprehensive soccer understanding\.InProceedings of the 33rd ACM International Conference on Multimedia,pp\. 3654–3663\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.55.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.55.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p6.1),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p4.1)\.
- Raoet al\.\(2025b\)J\. Rao, H\. Wu, H\. Jiang, Y\. Zhang, Y\. Wang, and W\. XieTowards universal soccer video understanding\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 8384–8394\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.56.4.1),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p6.1)\.
- Raoet al\.\(2024\)J\. Rao, H\. Wu, C\. Liu, Y\. Wang, and W\. XieMatchtime: towards automatic soccer game commentary generation\.InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing,pp\. 1671–1685\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.4.4),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.57.1),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.14.2)\.
- Raskaret al\.\(2025\)S\. Raskar, M\. Thosar, A\. Dandge, and P\. FaleFootyintel: creating an ai scout for better talent recognition\.International Journal of Environmental Sciences,pp\. 99–106\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p2.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.67.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.67.4),[§2\.6](https://arxiv.org/html/2608.14377#S2.SS6.p2.1)\.
- Raucheggeret al\.\(2024\)C\. Rauchegger, S\. M\. Wang, and P\. DelobelleOneLove beyond the field\-a few\-shot pipeline for topic and sentiment analysis during the fifa world cup in qatar\.InProceedings of the 20th Conference on Natural Language Processing \(KONVENS 2024\),pp\. 349–357\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p5.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.45.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.45.5),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p5.1)\.
- Renet al\.\(2024\)S\. Ren, L\. Yao, S\. Li, X\. Sun, and L\. HouTimechat: a time\-sensitive multimodal large language model for long video understanding\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 14313–14323\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Renet al\.\(2025\)W\. Ren, H\. Yang, J\. Min, C\. Wei, and W\. ChenVista: enhancing long\-duration and high\-resolution video understanding by video spatiotemporal augmentation\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 3804–3814\.Cited by:[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.34.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.35.1)\.
- Rocha\-Silvaet al\.\(2025\)R\. Rocha\-Silva, B\. E\. de Lima, T\. G\. Costa, N\. S\. Morais, G\. Jose, D\. F\. Cordeiro, A\. A\. de Almeida, G\. M\. Lopim, R\. B\. Viana, B\. S\. Sousa,et al\.Can people with epilepsy trust ai chatbots for information on physical exercise?\.Epilepsy & Behavior163,pp\. 110193\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Rocha\-Silvaet al\.\(2024\)R\. Rocha\-Silva, B\. E\. de Lima, G\. José, D\. F\. Cordeiro, R\. B\. Viana, M\. S\. Andrade, R\. L\. Vancini, T\. Rosemann, K\. Weiss, B\. Knechtle,et al\.The potential of large language model chatbots for application to epilepsy: let’s talk about physical exercise\.Epilepsy & Behavior Reports27,pp\. 100692\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Saglamet al\.\(2025\)S\. Saglam, V\. Uludag, Z\. O\. Karaduman, M\. Arıcan, M\. O\. Yücel, and R\. E\. DalaslanComparative evaluation of artificial intelligence models gpt\-4 and gpt\-3\.5 in clinical decision\-making in sports surgery and physiotherapy: a cross\-sectional study\.BMC Medical Informatics and Decision Making25\(1\),pp\. 163\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.14.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.14.4),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p2.1)\.
- Salehiet al\.\(2024\)M\. R\. Salehi, J\. S\. Park, A\. Kusupati, R\. Krishna, Y\. Choi, H\. Hajishirzi, and A\. FarhadiActionatlas: a videoqa benchmark for domain\-specialized action recognition\.Advances in Neural Information Processing Systems37,pp\. 137372–137402\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.21.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.21.4)\.
- Salimi Beniet al\.\(2025\)E\. Salimi Beni, M\. Mostahfezian, M\. Khorvash, and D\. Nasr EsfahaniComprehensive site selection model for sports facilities in iran: leveraging ai language models\.Sport Management Journal\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p1.1)\.
- Sameeret al\.\(2025\)N\. M\. Sameer, K\. Jayavardhan, and O\. R\. IyyappanEnhanced cricket commentary using ai vision and multilingual translation\.In2025 IEEE International Conference on Emerging Technologies and Applications \(MPSec ICETA\),pp\. 1–6\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.22.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.22.4)\.
- Sanhet al\.\(2019\)V\. Sanh, L\. Debut, J\. Chaumond, and T\. WolfDistilBERT, a distilled version of bert: smaller, faster, cheaper and lighter\.arXiv preprint arXiv:1910\.01108\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.29.5.1)\.
- Saraçet al\.\(2025\)H\. Saraç, İ\. T\. Ulusoy, J\. Alpay, H\. Ödemiş, and M\. SöğütEvaluating the potential role of ai chatbots in designing personalized exercise programs for weight management\.International Journal of Human–Computer Interaction,pp\. 1–8\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1)\.
- Sarfatiet al\.\(2023\)N\. Sarfati, I\. Yerushalmy, M\. Chertok, and Y\. KellerGenerating factually consistent sport highlights narrations\.InProceedings of the 6th International Workshop on Multimedia Content Analysis in Sports,pp\. 15–22\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.41.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.41.5)\.
- Sarkaret al\.\(2024\)S\. Sarkar, T\. S\. Yashwanth, and A\. GiriAdvancing cricket narratives: ai\-enhanced advanced journaling in the ipl using language models\.In2024 IEEE International Conference on Electronics, Computing and Communication Technologies \(CONECCT\),pp\. 1–6\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.37.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.37.4)\.
- Sarkhooshet al\.\(2024a\)M\. H\. Sarkhoosh, S\. Gautam, C\. Midoglu, S\. S\. Sabet, and P\. HalvorsenMultimodal ai\-based summarization and storytelling for soccer on social media\.InProceedings of the 15th ACM multimedia systems conference,pp\. 485–491\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p4.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.43.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.43.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p4.1)\.
- Sarkhooshet al\.\(2024b\)M\. H\. Sarkhoosh, S\. Gautam, C\. Midoglu, S\. S\. Sabet, T\. Torjusen, and P\. HalvorsenThe soccersum dataset for automated detection, segmentation, and tracking of objects on the soccer pitch\.InProceedings of the 15th ACM Multimedia Systems Conference,pp\. 353–359\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.43.1)\.
- Sattaret al\.\(2023\)H\. Sattar, M\. S\. Umar, E\. Ijaz, and M\. U\. ArshadMulti\-modal architecture for cricket highlights generation: using computer vision and large language model\.In2023 17th International Conference on Open Source Systems and Technologies \(ICOSST\),pp\. 1–6\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p2.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.25.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.25.5)\.
- Schillinget al\.\(2024\)A\. Schilling, J\. Anurathan, J\. Mühlberger, F\. Gerschner, M\. Rössle, A\. Theissler, and M\. KlaiberQuerying football matches for event data: towards using large language models\.InInternational Sports Analytics Conference and Exhibition,pp\. 216–227\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.47.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.47.5)\.
- Secareanu \(2023\)A\. SecareanuFootball events\.Note:[https://www\.kaggle\.com/datasets/secareanualin/football\-events](https://www.kaggle.com/datasets/secareanualin/football-events)Accessed: October 5, 2025Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.8.1)\.
- Seinoet al\.\(2025\)T\. Seino, N\. Saito, T\. Ogawa, S\. Asamizu, and M\. HaseyamaExpert comment generation considering sports skill level using a large multimodal model with video and spatial\-temporal motion features\.Sensors25\(2\),pp\. 447\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.10.4)\.
- Sepasdaret al\.\(2024a\)Z\. Sepasdar, S\. Gautam, C\. Midoglu, M\. A\. Riegler, and P\. HalvorsenEnhancing structured\-data retrieval with graphrag: soccer data case study\.arXiv preprint arXiv:2409\.17580\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1)\.
- Sepasdaret al\.\(2024b\)Z\. Sepasdar, S\. Gautam, C\. Midoglu, M\. A\. Riegler, and P\. HalvorsenSoccer\-graphrag: applications of graphrag in soccer\.InInternational Workshop on Graph\-Based Approaches in Information Retrieval,pp\. 1–10\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p4.1)\.
- Shaoet al\.\(2020\)D\. Shao, Y\. Zhao, B\. Dai, and D\. LinFinegym: a hierarchical video dataset for fine\-grained action understanding\.InProceedings of the IEEE/CVF conference on computer vision and pattern recognition,pp\. 2616–2625\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1)\.
- Shinet al\.\(2025\)Y\. Shin, S\. Park, Y\. Han, B\. Jeon, S\. Lee, and B\. J\. KangSoccer\-clip: vision language model for soccer action spotting\.IEEE Access13,pp\. 44354–44365\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.24.4.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.6.2)\.
- Skeriket al\.\(2018\)T\. Skerik, L\. Chrpa, W\. Faber, and M\. VallatiAutomated training plan generation for athletes\.In2018 IEEE international conference on systems, man, and cybernetics \(SMC\),pp\. 3865–3870\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- Solomon and Laye \(2025\)T\. P\. Solomon and M\. J\. LayeThe sports nutrition knowledge of large language model \(llm\) artificial intelligence \(ai\) chatbots: an assessment of accuracy, completeness, clarity, quality of evidence, and test\-retest reliability\.PloS one20\(6\),pp\. e0325982\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Songet al\.\(2025a\)H\. Song, D\. Hofer, R\. Islambouli, L\. Hawkins, A\. Bhattacharjee, M\. Franklin, and J\. J\. WilliamsInvestigating the relationship between physical activity and tailored behavior change messaging: connecting contextual bandit with large language models\.arXiv preprint arXiv:2506\.07275\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1)\.
- Songet al\.\(2025b\)S\. Song, J\. Park, J\. Choi, J\. Lee, K\. Jin, and Y\. KimKorean football in\-game conversation state tracking dataset for dialogue and turn level evaluation\.Engineering Applications of Artificial Intelligence139,pp\. 109572\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.52.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.52.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p6.1)\.
- SportDevs \(2025\)SportDevsHandball api\.Note:[https://sportdevs\.com/handball](https://sportdevs.com/handball)Accessed: October 5, 2025Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.50.1)\.
- Sportsvision \(2025\)SportsvisionNSVA subset: basketball video\-text dataset\.Note:[https://huggingface\.co/datasets/sportsvision/nsva\_subset](https://huggingface.co/datasets/sportsvision/nsva_subset)Hugging FaceCited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.22.1)\.
- Sprint \(2024\)G\. SprintSocial networks and large language models for division i basketball game winner prediction\.IEEE Access12,pp\. 84774–84784\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p4.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.49.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.49.4)\.
- Srivastavaet al\.\(2023\)A\. Srivastava, A\. Rastogi, A\. Rao, A\. A\. Shoeb, A\. Abid, A\. Fisch, A\. R\. Brown, A\. Santoro, A\. Gupta, A\. Garriga\-Alonso,et al\.Beyond the imitation game: quantifying and extrapolating the capabilities of language models\.Transactions on machine learning research\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.4.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.2.1)\.
- Strandet al\.\(2024a\)A\. T\. Strand, S\. Gautam, C\. Midoglu, and P\. HalvorsenSoccer information retrieval via natural queries using soccerrag\.In2024 International Conference on Content\-Based Multimedia Indexing \(CBMI\),pp\. 1–5\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.48.4),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p4.1)\.
- Strandet al\.\(2024b\)A\. T\. Strand, S\. Gautam, C\. Midoglu, and P\. HalvorsenSoccerrag: multimodal soccer information retrieval via natural queries\.In2024 International Conference on Content\-Based Multimedia Indexing \(CBMI\),pp\. 1–7\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.48.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p6.1)\.
- Szabo \(2023\)A\. SzaboChatGPT is a breakthrough in science and education but fails a test in sports and exercise psychology\.Baltic Journal of Sport and Health Sciences1\(128\),pp\. 25–40\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1)\.
- Tanget al\.\(2025a\)C\. Tang, Y\. Li, Y\. Yang, J\. Zhuang, G\. Sun, W\. Li, Z\. Ma, and C\. ZhangVideo\-salmonn 2: captioning\-enhanced audio\-visual large language models\.arXiv preprint arXiv:2506\.15220\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.26.1)\.
- Tanget al\.\(2024\)C\. Tang, Y\. Li, Y\. Yang, J\. Zhuang, G\. Sun, W\. Li, Z\. Ma, and C\. ZhangEnhancing multimodal llm for detailed and accurate video captioning using multi\-round preference optimization\.arXiv preprint arXiv:2410\.06682\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1)\.
- Tanget al\.\(2025b\)Z\. Tang, J\. Li, P\. Guo, B\. Tian, Q\. Xing, X\. Xing, and P\. WangFitnessAgent: a unified agent framework for open\-set and personalized fitness evaluation\.In2025 IEEE International Conference on Robotics and Automation \(ICRA\),pp\. 12437–12444\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.36.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.36.4),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1)\.
- Teamet al\.\(2023\)G\. Team, R\. Anil, S\. Borgeaud, J\. Alayrac, J\. Yu, R\. Soricut, J\. Schalkwyk, A\. M\. Dai, A\. Hauth, K\. Millican,et al\.Gemini: a family of highly capable multimodal models\.arXiv preprint arXiv:2312\.11805\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.16.6),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.41.6),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.5.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.37.5),[§1](https://arxiv.org/html/2608.14377#S1.p2.1)\.
- Team \(2024\)M\. A\. TeamMistral\-large\-instruct\-2411: a 123b parameter dense language model\.Note:[https://huggingface\.co/mistralai/Mistral\-Large\-Instruct\-2411](https://huggingface.co/mistralai/Mistral-Large-Instruct-2411)Accessed: October 5, 2025Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.27.5)\.
- Teamet al\.\(2024\)Q\. Teamet al\.Qwen2 technical report\.arXiv preprint arXiv:2407\.106712,pp\. 3\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.13.6)\.
- Teo \(2025\)C\. TeoEnhancing sports strategy with video analytics and data mining: assessing the effectiveness of multimodal llms in tennis video analysis\.arXiv preprint arXiv:2507\.02904\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.23.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.23.4)\.
- Thoppilanet al\.\(2022\)R\. Thoppilan, D\. De Freitas, J\. Hall, N\. Shazeer, A\. Kulshreshtha, H\. Cheng, A\. Jin, T\. Bos, L\. Baker, Y\. Du,et al\.Lamda: language models for dialog applications\.arXiv preprint arXiv:2201\.08239\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.4.5)\.
- Touvronet al\.\(2023a\)H\. Touvron, T\. Lavril, G\. Izacard, X\. Martinet, M\. Lachaux, T\. Lacroix, B\. Rozière, N\. Goyal, E\. Hambro, F\. Azhar,et al\.LLaMA: open and efficient foundation language models\.arXiv preprint arXiv:2302\.13971\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.8.5)\.
- Touvronet al\.\(2023b\)H\. Touvron, L\. Martin, K\. Stone, P\. Albert, A\. Almahairi, Y\. Babaei, N\. Bashlykov, S\. Batra, P\. Bhargava, S\. Bhosale,et al\.Llama 2: open foundation and fine\-tuned chat models\.arXiv preprint arXiv:2307\.09288\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.11.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.13.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.5.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.9.5)\.
- Unlu \(2023\)E\. UnluFootGPT: a large language model development experiment on a minimal setting\.arXiv preprint arXiv:2308\.08610\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1)\.
- Vahdatiet al\.\(2025\)M\. Vahdati, K\. Gholizadeh HamlAbadi, F\. Laamarti, and A\. El SaddikA multi\-agent digital twin framework for ai\-driven fitness coaching\.InProceedings of the 2025 ACM International Conference on Interactive Media Experiences,pp\. 380–385\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1)\.
- Van Nguyenet al\.\(2025\)C\. Van Nguyen, X\. Shen, R\. Aponte, Y\. Xia, S\. Basu, Z\. Hu, J\. Chen, M\. Parmar, S\. Kunapuli, J\. Barrow,et al\.A survey on small language models\.InProceedings of the 15th International Conference on Recent Advances in Natural Language Processing\-Natural Language Processing in the Generative AI Era,pp\. 807–821\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p5.1)\.
- Vandelanotteet al\.\(2023\)C\. Vandelanotte, S\. Trost, D\. Hodgetts, T\. Imam, M\. Rashid, Q\. G\. To, and C\. MaherIncreasing physical activity using an just\-in\-time adaptive digital assistant supported by machine learning: a novel approach for hyper\-personalised mhealth interventions\.Journal of Biomedical Informatics144,pp\. 104435\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1)\.
- Vardhanet al\.\(2022\)M\. Vardhan, N\. Hegde, S\. Merugu, S\. Prabhat, D\. Nathani, M\. Seneviratne, N\. Muhammad, P\. Reddy, S\. Lakshminarasimhan, R\. Singh,et al\.Walking with pace\-personalized and automated coaching engine\.InProceedings of the 30th ACM Conference on User Modeling, Adaptation and Personalization,pp\. 57–68\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.4.1)\.
- Vaswaniet al\.\(2017\)A\. Vaswani, N\. Shazeer, N\. Parmar, J\. Uszkoreit, L\. Jones, A\. N\. Gomez, Ł\. Kaiser, and I\. PolosukhinAttention is all you need\.Advances in neural information processing systems30\.Cited by:[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.4.1)\.
- Vijayakumaret al\.\(2025\)A\. Vijayakumar, A\. Toms, and S\. VadivuPlayer tracking\-integrated soccer game commentary generation\.IJSAT\-International Journal on Science and Technology16\(2\)\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.17.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.17.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1)\.
- Wachholzet al\.\(2025\)F\. Wachholz, S\. Manno, D\. Schlachter, N\. Gamper, and M\. SchnitzerAcceptance and trust in ai\-generated exercise plans among recreational athletes and quality evaluation by experienced coaches: a pilot study\.BMC Research Notes18\(1\),pp\. 112\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Wackerhage and Schoenfeld \(2021\)H\. Wackerhage and B\. J\. SchoenfeldPersonalized, evidence\-informed training plans and exercise prescriptions for performance, fitness and health\.Sports Medicine51\(9\),pp\. 1805–1813\.Cited by:[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1)\.
- Wanget al\.\(2025a\)F\. Wang, Q\. Wang, and D\. ChenFrom beats to scores: a multi\-modal framework for comprehensive figure skating assessment\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 5905–5914\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.32.5.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.10.2)\.
- Wanget al\.\(2025b\)H\. Wang, S\. Salekin, J\. Lee, R\. Claytor, S\. Zhang, and M\. ChiAgentic generative ai for media content discovery at the national football league\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1)\.
- Wanget al\.\(2021\)J\. Wang, Z\. Li, Q\. Yang, J\. Qu, Z\. Chen, Q\. Liu, and G\. HuSportssum2\. 0: generating high\-quality sports news from live text commentary\.InProceedings of the 30th ACM International Conference on Information & Knowledge Management,pp\. 3463–3467\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.31.2),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.31.5.1)\.
- Wanget al\.\(2022\)J\. Wang, Z\. Li, T\. Zhang, D\. Zheng, J\. Qu, A\. Liu, L\. Zhao, and Z\. ChenKnowledge enhanced sports game summarization\.InProceedings of the Fifteenth ACM International Conference on Web Search and Data Mining,pp\. 1045–1053\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.33.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.33.4.1),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p3.1)\.
- Wanget al\.\(2024a\)P\. Wang, S\. Bai, S\. Tan, S\. Wang, Z\. Fan, J\. Bai, K\. Chen, X\. Liu, J\. Wang, W\. Ge,et al\.Qwen2\-vl: enhancing vision\-language model’s perception of the world at any resolution\.arXiv preprint arXiv:2409\.12191\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.18.5),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.53.5.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.12.1)\.
- Wanget al\.\(2023\)W\. Wang, Y\. Huang, T\. Ik, and W\. PengShuttleset: a human\-annotated stroke\-level singles dataset for badminton tactical analysis\.InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining,pp\. 5126–5136\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.47.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.36.1)\.
- Wanget al\.\(2025c\)W\. Wang, Z\. He, W\. Hong, Y\. Cheng, X\. Zhang, J\. Qi, M\. Ding, X\. Gu, S\. Huang, B\. Xu,et al\.Lvbench: an extreme long video understanding benchmark\.InProceedings of the IEEE/CVF International Conference on Computer Vision,pp\. 22958–22967\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.27.1)\.
- Wanget al\.\(2025d\)X\. Wang, D\. Song, S\. Chen, J\. Chen, Z\. Cai, C\. Zhang, L\. Sun, and B\. WangLongLLaVA: scaling multi\-modal LLMs to 1000 images efficiently via a hybrid architecture\.InFindings of the Association for Computational Linguistics: EMNLP 2025,C\. Christodoulopoulos, T\. Chakraborty, C\. Rose, and V\. Peng \(Eds\.\),Suzhou, China,pp\. 21419–21436\.External Links:[Link](https://aclanthology.org/2025.findings-emnlp.1168/),[Document](https://dx.doi.org/10.18653/v1/2025.findings-emnlp.1168),ISBN 979\-8\-89176\-335\-7Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Wanget al\.\(2024b\)Y\. Wang, Y\. He, Y\. Li, K\. Li, J\. Yu, X\. Ma, X\. Li, G\. Chen, X\. Chen, Y\. Wang,et al\.InternVid: a large\-scale video\-text dataset for multimodal understanding and generation\.InThe Twelfth International Conference on Learning Representations,Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.20.1),[§3\.1](https://arxiv.org/html/2608.14377#S3.SS1.p2.1)\.
- Wanget al\.\(2026\)Y\. Wang, Z\. Chen, R\. Chen, S\. Gu, W\. Hu, J\. Liu, Y\. Dong, H\. Su, J\. Zhu, M\. Wang,et al\.Benchmarking trustworthiness in multimodal llms for video understanding\.InProceedings of the AAAI Conference on Artificial Intelligence,Vol\.40,pp\. 37979–37987\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.51.1)\.
- Wang and Wang \(2024\)Y\. Wang and X\. WangArtificial intelligence in physical education: comprehensive review and future teacher training strategies\.Frontiers in public health12,pp\. 1484848\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1)\.
- Wanget al\.\(2024c\)Z\. Wang, Y\. Hu, J\. Liu, and L\. HuImpact of chatgpt technology on sports industry\.Journal of New Media and Economics1\(4\),pp\. 29–37\.Cited by:[§1](https://arxiv.org/html/2608.14377#S1.p3.1)\.
- Wanget al\.\(2024d\)Z\. Wang, P\. Veličković, D\. Hennes, N\. Tomašev, L\. Prince, M\. Kaisers, Y\. Bachrach, R\. Elie, L\. K\. Wenliang, F\. Piccinini,et al\.TacticAI: an ai assistant for football tactics\.Nature communications15\(1\),pp\. 1906\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p5.1)\.
- Wang and Yoshinaga \(2024\)Z\. Wang and N\. YoshinagaCommentary generation from data records of multiplayer strategy esports game\.InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies \(Volume 4: Student Research Workshop\),pp\. 263–271\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.5.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.5.4),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1)\.
- Washifet al\.\(2024\)J\. Washif, J\. Pagaduan, C\. James, I\. Dergaa, and C\. BeavenArtificial intelligence in sport: exploring the potential of using chatgpt in resistance training prescription\.Biology of sport41\(2\),pp\. 209–220\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Wickramasinghe \(2025\)I\. WickramasingheAssessing the accuracy of large language models in extracting latest cricket information\.Scientific Journal of Sport and Performance4\(2\),pp\. 268–284\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p6.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.50.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.50.4)\.
- Wisemanet al\.\(2017\)S\. Wiseman, S\. M\. Shieber, and A\. M\. RushChallenges in data\-to\-document generation\.InProceedings of the 2017 Conference on Empirical Methods in Natural Language Processing,pp\. 2253–2263\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.38.1)\.
- Wohlin \(2014\)C\. WohlinGuidelines for snowballing in systematic literature studies and a replication in software engineering\.InProceedings of the 18th international conference on evaluation and assessment in software engineering,pp\. 1–10\.Cited by:[Appendix A](https://arxiv.org/html/2608.14377#A1.p1.1),[§1](https://arxiv.org/html/2608.14377#S1.p4.1)\.
- Xiet al\.\(2025a\)Z\. Xi, G\. Shi, X\. Li, J\. Yan, Z\. Li, L\. Wu, Z\. Liu, and L\. WangA simple yet effective knowledge guided method for entity\-aware video captioning on a basketball benchmark\.Neurocomputing619,pp\. 129177\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.21.1)\.
- Xiet al\.\(2025b\)Z\. Xi, H\. Sun, Y\. Wu, J\. Yan, H\. Zhang, L\. Wu, L\. Wang, and C\. ChenPlayer\-centric multimodal prompt generation for large language model based identity\-aware basketball video captioning\.InProceedings of the IEEE/CVF International Conference on Computer Vision,pp\. 24330–24339\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.20.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.20.4.1),[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p2.1)\.
- Xiaet al\.\(2026\)H\. Xia, H\. Ge, J\. Zou, H\. W\. Choi, X\. Zhang, D\. Suradja, B\. Rui, E\. Tran, W\. Jin, Z\. Ye, X\. Lin, C\. Lai, S\. Zhang, J\. Miao, S\. Chen, R\. Tracy, V\. Ordonez, W\. Shen, and H\. ChenSportR: a benchmark for multimodal large language model reasoning in sports\.InThe Fourteenth International Conference on Learning Representations,External Links:[Link](https://openreview.net/forum?id=cPCGB402ff)Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p4.1)\.
- Xiaet al\.\(2024a\)H\. Xia, Z\. Yang, Y\. Wang, R\. Tracy, Y\. Zhao, D\. Huang, Z\. Chen, Y\. Zhu, Y\. Wang, and W\. ShenSportQA: a benchmark for sports understanding in large language models\.InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies \(Volume 1: Long Papers\),pp\. 5061–5081\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.5.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.6.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.7.1),[§3\.1](https://arxiv.org/html/2608.14377#S3.SS1.p2.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.3.1)\.
- Xiaet al\.\(2024b\)H\. Xia, Z\. Yang, Y\. Zhao, Y\. Wang, J\. Li, R\. Tracy, Z\. Zhu, Y\. Wang, H\. Chen, and W\. ShenLanguage and multimodal models in sports: a survey of datasets and applications\.arXiv preprint arXiv:2406\.12252\.Cited by:[§A\.1](https://arxiv.org/html/2608.14377#A1.SS1.p1.1),[§1](https://arxiv.org/html/2608.14377#S1.p3.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p4.1)\.
- Xiaet al\.\(2025\)H\. Xia, Z\. Yang, J\. Zou, R\. Tracy, Y\. Wang, C\. Lu, C\. Lai, Y\. He, X\. Shao, Z\. Xie,et al\.SPORTU: a comprehensive sports understanding benchmark for multimodal large language models\.InThe Thirteenth International Conference on Learning Representations,Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p2.1),[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.63.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.8.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.9.1),[Table 2](https://arxiv.org/html/2608.14377#S3.T2.2.4.1)\.
- Xiaet al\.\(2023\)J\. Xia, M\. Zhuge, T\. Geng, S\. Fan, Y\. Wei, Z\. He, and F\. ZhengSkating\-mixer: long\-term sport audio\-visual modeling with mlps\.InProceedings of the AAAI Conference on Artificial Intelligence,Vol\.37,pp\. 2901–2909\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.33.1)\.
- Xieet al\.\(2025\)L\. Xie, G\. Z\. Wei, A\. Kuthiala, C\. Zheng, A\. Bal, M\. Dabhi, L\. Wen, T\. Rustagi, E\. Lai, S\. Khyalia,et al\.MAVERIX: multimodal audio\-visual evaluation reasoning index\.arXiv preprint arXiv:2503\.21699\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.42.1),[§3\.2](https://arxiv.org/html/2608.14377#S3.SS2.p3.1)\.
- Xinget al\.\(2025\)Q\. Xing, X\. Xing, P\. Guo, Z\. Tang, and Y\. ShenLLM\-fms: a fine\-grained dataset for functional movement screen action quality assessment\.PloS one20\(3\),pp\. e0313707\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.37.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.37.4)\.
- Xuet al\.\(2019\)C\. Xu, Y\. Fu, B\. Zhang, Z\. Chen, Y\. Jiang, and X\. XueLearning to score figure skating sport videos\.IEEE transactions on circuits and systems for video technology30\(12\),pp\. 4578–4590\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.32.2)\.
- Xuet al\.\(2022\)J\. Xu, Y\. Rao, X\. Yu, G\. Chen, J\. Zhou, and J\. LuFinediving: a fine\-grained dataset for procedure\-aware action quality assessment\.InProceedings of the IEEE/CVF conference on computer vision and pattern recognition,pp\. 2949–2958\.Cited by:[§C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1.p3.1)\.
- Xuet al\.\(2026\)R\. Xu, G\. Xiao, Y\. Chen, L\. He, K\. Peng, Y\. Lu, and S\. HanStreamingVLM: real\-time understanding for infinite video streams\.InThe Fourteenth International Conference on Learning Representations,External Links:[Link](https://openreview.net/forum?id=gVbPWbA97s)Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Xuet al\.\(2024a\)X\. Xu, M\. Li, C\. Tao, T\. Shen, R\. Cheng, J\. Li, C\. Xu, D\. Tao, and T\. ZhouA survey on knowledge distillation of large language models\.arXiv preprint arXiv:2402\.13116\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p5.1)\.
- Xuet al\.\(2024b\)Y\. Xu, Q\. Liu, J\. Pang, C\. Zeng, X\. Ma, P\. Li, L\. Ma, J\. Huang, and H\. XieAssessment of personalized exercise prescriptions issued by chatgpt 4\.0 and intelligent health promotion systems for patients with hypertension comorbidities based on the transtheoretical model: a comparative analysis\.Journal of Multidisciplinary Healthcare,pp\. 5063–5078\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Xueet al\.\(2021\)L\. Xue, N\. Constant, A\. Roberts, M\. Kale, R\. Al\-Rfou, A\. Siddhant, A\. Barua, and C\. RaffelMT5: a massively multilingual pre\-trained text\-to\-text transformer\.InProceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,pp\. 483–498\.Cited by:[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.33.5.1)\.
- Xunet al\.\(2025\)S\. Xun, S\. Tao, J\. Li, Y\. Shi, Z\. Lin, Z\. Zhu, Y\. Yan, H\. Li, L\. Zhang, S\. Wang,et al\.RTV\-bench: benchmarking mllm continuous perception, understanding and reasoning through real\-time video\.InThe Thirty\-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track,Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.43.1)\.
- Yanget al\.\(2025a\)S\. Yang, W\. Yu, W\. Yang, X\. Liu, H\. Tan, L\. Lan, and N\. XiaoWildVideo: benchmarking lmms for understanding video\-language interaction\.IEEE Transactions on Pattern Analysis and Machine Intelligence\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.49.1)\.
- Yanget al\.\(2025b\)Z\. Yang, Y\. Shu, Z\. Yang, Y\. Zhang, Y\. Li, K\. Lu, G\. Zeng, S\. Liu, Y\. Zhou, and N\. SebeVidText: towards comprehensive evaluation for video text understanding\.arXiv preprint arXiv:2505\.22810\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.44.1)\.
- Yehet al\.\(2023\)W\. Yeh, P\. H\. Lin, Y\. Su, W\. H\. Cheng, and L\. KuMAAIG: motion analysis and instruction generation\.InProceedings of the 5th ACM International Conference on Multimedia in Asia Workshops,pp\. 1–5\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.3.5),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.3.2)\.
- Yehet al\.\(2025\)W\. Yeh, Y\. Su, C\. Chen, Y\. Lin, C\. Ku, W\. Chiu, M\. Hu, and L\. KuCoachMe: decoding sport elements with a reference\-based coaching instruction generation model\.InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),pp\. 29126–29151\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.11.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.11.4.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.12.1)\.
- Yenisoy and Silik \(2025\)C\. Yenisoy and C\. E\. SilikInvestigating esports tourism research using artificial intelligence applications: chatgpt versus zekai\.Tourism and Recreation7\(1\),pp\. 54–68\.Cited by:[§B\.6](https://arxiv.org/html/2608.14377#A2.SS6.p3.1),[§2\.6](https://arxiv.org/html/2608.14377#S2.SS6.p3.1)\.
- Yiet al\.\(2025\)H\. Yi, Y\. Pan, F\. He, X\. Liu, B\. Zhang, O\. Oguntola, and G\. BertasiusExAct: a video\-language benchmark for expert action analysis\.InThe Thirty\-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track,Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.46.1)\.
- Youet al\.\(2025\)L\. You, W\. Huang, X\. Xie, X\. Wei, B\. Li, S\. Lin, Y\. Li, and C\. WangTimesoccer: an end\-to\-end multimodal large language model for soccer commentary generation\.InProceedings of the 33rd ACM International Conference on Multimedia,pp\. 3418–3427\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.13.4),[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p3.1),[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p3.1)\.
- Yuet al\.\(2025\)J\. Yu, Y\. Wu, M\. Chu, Z\. Ren, Z\. Huang, P\. Chu, R\. Zhang, Y\. He, Q\. Li, S\. Li,et al\.Vrbench: a benchmark for multi\-step reasoning in long narrative videos\.InProceedings of the IEEE/CVF International Conference on Computer Vision,pp\. 21655–21666\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.47.1)\.
- Yu and Chai \(2025\)K\. P\. Yu and J\. ChaiTemporally\-grounded language generation: a benchmark for real\-time vision\-language models\.arXiv preprint arXiv:2505\.11326\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.24.4)\.
- Yuet al\.\(2024\)T\. Yu, Y\. Yao, H\. Zhang, T\. He, Y\. Han, G\. Cui, J\. Hu, Z\. Liu, H\. Zheng, M\. Sun,et al\.Rlhf\-v: towards trustworthy mllms via behavior alignment from fine\-grained correctional human feedback\.InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp\. 13807–13816\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p2.1)\.
- Yuanet al\.\(2025\)H\. Yuan, Z\. Liu, Y\. Wang, J\. Zhou, Z\. Liang, B\. Zhao, Z\. Cao, J\. Wen, Z\. Dou,et al\.MomentSeeker: a task\-oriented benchmark for long\-video moment retrieval\.InThe Thirty\-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track,Cited by:[§C\.2\.3](https://arxiv.org/html/2608.14377#A3.SS2.SSS3.p1.1)\.
- Zandycke \(2024\)G\. V\. ZandyckeDeepSport dataset\.Note:[https://www\.kaggle\.com/datasets/gabrielvanzandycke/deepsport\-dataset](https://www.kaggle.com/datasets/gabrielvanzandycke/deepsport-dataset)Accessed: October 5, 2025Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.45.2)\.
- Zhanget al\.\(2024a\)B\. Zhang, J\. Gao, and Y\. YuanA descriptive basketball highlight dataset for automatic commentary generation\.InProceedings of the 32nd ACM international conference on multimedia,pp\. 10316–10325\.Cited by:[§B\.4](https://arxiv.org/html/2608.14377#A2.SS4.p1.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.12.1),[Table 4](https://arxiv.org/html/2608.14377#A3.T4.2.12.4)\.
- Zhanget al\.\(2023\)H\. Zhang, X\. Li, and L\. BingVideo\-llama: an instruction\-tuned audio\-visual language model for video understanding\.InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing: System Demonstrations,pp\. 543–553\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.31.6),[§2\.4](https://arxiv.org/html/2608.14377#S2.SS4.p1.1),[Table 1](https://arxiv.org/html/2608.14377#S2.T1.2.1.17.1)\.
- Zhanget al\.\(2021\)H\. Zhang, C\. Sciutto, M\. Agrawala, and K\. FatahalianVid2player: controllable video sprites that behave and appear like professional tennis players\.ACM Transactions on Graphics \(TOG\)40\(3\),pp\. 1–16\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.25.1)\.
- Zhanget al\.\(2025a\)J\. Zhang, D\. Han, S\. Han, H\. Li, W\. Lam, and M\. ZhangChatMatch: exploring the potential of hybrid vision–language deep learning approach for the intelligent analysis and inference of racket sports\.Computer Speech & Language89,pp\. 101694\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p3.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.44.2),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.44.5)\.
- Zhanget al\.\(2025b\)K\. Zhang, Y\. Qin, and B\. QinRAG\-llm based evaluation pathway and technological exploration for the scientific validity of mass fitness\.InOf Papers Presented at 2025 6th Asia Sport Science Conference \(ASSC\),Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p1.1)\.
- Zhanget al\.\(2018\)S\. Zhang, X\. Zhang, H\. Wang, L\. Guo, and S\. LiuMulti\-scale attentive interaction networks for chinese medical question answer selection\.IEEE Access6,pp\. 74061–74071\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.13.2)\.
- Zhang and Liu \(2024\)Y\. Zhang and X\. LiuUsing chatgpt to promote college students’ participation in physical activities and its effect on mental health\.World Journal of Psychiatry14\(2\),pp\. 330\.Cited by:[§B\.2](https://arxiv.org/html/2608.14377#A2.SS2.p5.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p5.1)\.
- Zhanget al\.\(2024b\)Y\. Zhang, S\. Lei, R\. Qiao, Z\. GongQue, X\. Song, G\. Dong, Q\. Tan, Z\. Wei, P\. Yang, Y\. Tian,et al\.Multi\-dimensional insights: benchmarking real\-world personalization in large multimodal models\.arXiv preprint arXiv:2412\.12606\.Cited by:[§C\.2\.3](https://arxiv.org/html/2608.14377#A3.SS2.SSS3.p1.1)\.
- Zhanget al\.\(2025c\)Y\. Zhang, J\. Wu, W\. Li, B\. Li, Z\. Ma, Z\. Liu, and C\. LiLLaVA\-video: video instruction tuning with synthetic data\.Transactions on Machine Learning Research\.Cited by:[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.35.5)\.
- Zhaoet al\.\(2024\)J\. Zhao, Z\. Zhang, B\. Chen, Z\. Wang, A\. Anandkumar, and Y\. TianGaLore: memory\-efficient llm training by gradient low\-rank projection\.InInternational Conference on Machine Learning,pp\. 61121–61143\.Cited by:[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p2.1)\.
- Zhaoet al\.\(2025\)Z\. Zhao, W\. Chai, S\. Hao, W\. Hu, G\. Wang, S\. Cao, M\. Song, J\. Hwang, and G\. WangA survey of deep learning in sports applications: perception, comprehension, and decision\.IEEE Transactions on Visualization and Computer Graphics\.Cited by:[§A\.1](https://arxiv.org/html/2608.14377#A1.SS1.p1.1),[§1](https://arxiv.org/html/2608.14377#S1.p3.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p1.1),[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1)\.
- Zhouet al\.\(2025a\)D\. Zhou, J\. W\. Keogh, Y\. Ma, R\. K\. Tong, A\. R\. Khan, and N\. R\. JenningsArtificial intelligence in sport: a narrative review of applications, challenges and future trends\.Journal of Sports Sciences,pp\. 1–16\.Cited by:[§A\.1](https://arxiv.org/html/2608.14377#A1.SS1.p1.1),[§1](https://arxiv.org/html/2608.14377#S1.p1.1),[§1](https://arxiv.org/html/2608.14377#S1.p3.1)\.
- Zhouet al\.\(2025b\)J\. Zhou, Y\. Shu, B\. Zhao, B\. Wu, Z\. Liang, S\. Xiao, M\. Qin, X\. Yang, Y\. Xiong, B\. Zhang,et al\.Mlvu: benchmarking multi\-task long video understanding\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 13691–13701\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p2.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.30.1)\.
- Zhouet al\.\(2024\)K\. Zhou, R\. Cai, L\. Wang, H\. P\. Shum, and X\. LiangA comprehensive survey of action quality assessment: method and benchmark\.arXiv preprint arXiv:2412\.11149\.Cited by:[§2\.2](https://arxiv.org/html/2608.14377#S2.SS2.p2.1)\.
- Zhouet al\.\(2025c\)Z\. Zhou, K\. Mei, Y\. Lu, T\. Wang, and F\. RaoHarmonyset: a comprehensive dataset for understanding video\-music semantic alignment and temporal synchronization\.InProceedings of the Computer Vision and Pattern Recognition Conference,pp\. 3152–3162\.Cited by:[§C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2.p3.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.36.1),[Table 5](https://arxiv.org/html/2608.14377#A3.T5.2.37.1)\.
- Zhuet al\.\(2024a\)W\. Zhu, W\. Geng, L\. Huang, X\. Qin, Z\. Chen, and H\. YanWho could and should give exercise prescription: physicians, exercise and health scientists, fitness trainers, or chatgpt?\.Journal of Sport and Health Science13\(3\),pp\. 368–372\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p1.1)\.
- Zhuet al\.\(2025\)X\. Zhu, Z\. Gao, and X\. A\. WangFull\-parameter fine\-tuning method of llms for sports injury prevention and treatment\.International Journal of Mobile Computing and Multimedia Communications \(IJMCMC\)16\(1\),pp\. 1–14\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p2.1),[Table 3](https://arxiv.org/html/2608.14377#A3.T3.2.13.5),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p2.1)\.
- Zhuet al\.\(2024b\)X\. Zhu, J\. Li, Y\. Liu, C\. Ma, and W\. WangA survey on model compression for large language models\.Transactions of the Association for Computational Linguistics12,pp\. 1556–1577\.Cited by:[§4\.2](https://arxiv.org/html/2608.14377#S4.SS2.p5.1)\.
- Zouet al\.\(2024\)H\. Zou, T\. Luo, G\. Xie, F\. Lv, G\. Wang, J\. Chen, Z\. Wang, H\. Zhang, H\. Zhang,et al\.From seconds to hours: reviewing multimodal large language models on comprehensive long video understanding\.arXiv preprint arXiv:2409\.18938\.Cited by:[§4\.1](https://arxiv.org/html/2608.14377#S4.SS1.p3.1)\.
- Zuccolotto \(2025\)P\. Zuccolotto11 th mathsport international conference 4\-6 june 2025\.Cited by:[§B\.1](https://arxiv.org/html/2608.14377#A2.SS1.p3.1),[§2\.1](https://arxiv.org/html/2608.14377#S2.SS1.p3.1)\.

Figure 4:Papers on large models in sports over the years \(data for 2025 is up to July\)\.## Appendix AMethodology for Literature Selection

In this section, we detail the systematic methodology employed for literature identification, screening, and selection\. To capture the fragmented and rapidly evolving landscape of large models in sports, we adopted asystematic snowballing methodology[281](https://arxiv.org/html/2608.14377#bib.bib304), adhering to the reporting standards of thePreferred Reporting Items for Systematic Reviews and Meta\-Analysis\(PRISMA\) statement[188](https://arxiv.org/html/2608.14377#bib.bib305)\. This dual\-direction strategy \(leveraging both reference lists and citation networks\) is particularly effective for interdisciplinary fields such as large models in sports, ensuring high relevance by tracing semantic connections rather than relying solely on keyword indexing\.

### A\.1Construction of the Start Set

The effectiveness of snowballing relies heavily on the quality of the initial start set\. Instead of a broad, potentially noisy keyword search, we established our foundation by identifying3highly relevant and comprehensive survey papers based on domain expertise\([286](https://arxiv.org/html/2608.14377#bib.bib12);[321](https://arxiv.org/html/2608.14377#bib.bib4);[320](https://arxiv.org/html/2608.14377#bib.bib11)\)\. These papers serve as our "seeds" for initiating the iterative snowballing process\. These papers were chosen for their:

Comprehensiveness\.Collectively, they cover the entire spectrum from traditional deep learning to modern large models\.

Recency\.All selected seeds were published in 2024–2025, ensuring the survey is anchored in the most current research landscape\.

Academic Standing\.The set combines rigorous articles from premier journals with pioneering preprints that address the rapid evolution of large models before formal publication cycles\.

Connectivity\.They serve as central hubs in the citation network, linking to a wide range of task\-specific studies\.

### A\.2Iterative Snowballing Procedure

Starting from these seed papers, we performed iterative forward and backward snowballing to expand our corpus\. To manage the scale of the literature and ensure precision, we applied a specific Boolean query as a filtering mechanism during the forward pass\.

Backward Snowballing\.We scrutinized the reference lists of the included papers to uncover relevant prior studies and foundational works\.

Forward Snowballing\.We leveraged Google Scholar’s “Cited by” feature to access the citation list of each paper\. To efficiently filter out out\-of\-domain works from the large volume of citations, we enabled the “Search within citing articles” option and applied the following Boolean search string:

"Sports" AND \("Large Language Model" OR "LLM" OR "GPT" OR "BERT" OR "T5"\)

This step allowed us to strictly identify studies that integrate large models within sports contexts, capturing the latest research developments up to July 2025\.

Iteration & Saturation\.Newly identified papers that met the inclusion criteria were added to the set and treated as new seeds\. This cycle was repeated until theoretical saturation was reached \(i\.e\., the filtered search yielded no new relevant papers\)\.

### A\.3Inclusion and Exclusion Criteria

To isolate relevant studies from the retrieved pool, we applied the following rigorous filters across four dimensions:

Research Topic\.We included studies that target tasks within the sports domain or involve sports data analysis, provided that they utilize largemodels\(e\.g\., LLMs, MLLMs\) as a core methodological component\. To maintain the survey’s specific focus on the era of large models, we excluded studies that rely solely on traditional deep learning architectures \(e\.g\., CNNs, LSTMs\) without the integration of large models\.

Publication Type\.To ensure technical depth, scientific rigor, and mitigate the risk of low\-quality evidence, we restricted our selection to full\-length academic contributions\. Included works comprise peer\-reviewed conference and journal papers, as well as cutting\-edge preprints that represent the latest advancements in the field\. We excluded non\-technical documents such as editorials, posters, extended abstracts, opinion pieces, and short papers that lacked sufficient implementation details or experimental validation\.

Time Window\.We defined a specific temporal scope to align with the emergence and proliferation of large models\. The search and inclusion window was strictly defined from January 1, 2020, to July 31, 2025\. Although the final search and screening process was executed on October 4, 2025, we enforced this cutoff date to ensure a consistent timeframe for data analysis\.

Language\.To ensure accessibility and consistent analysis, we included only articles written in English\. Studies published in other languages were excluded\.

### A\.4Selection Results

Throughout the iterative forward and backward snowballing process, we examined a cumulative total of approximately2,200candidate records\. After rigorously applying the inclusion and exclusion criteria to these candidates, a final set of241core academic papers was selected for this survey\. The rapid growth trend and temporal distribution of these included works are illustrated in Figure[4](https://arxiv.org/html/2608.14377#A0.F4)\.

## Appendix BMore Details on Large Model Applications in Sports

This section serves as a comprehensive supplement to Section[2](https://arxiv.org/html/2608.14377#S2), offering a detailed literature review of specific studies and methodologies\. Given that task definitions, analyses of large model–related technologies, and common evaluation metrics have been elaborated in the main text, this section will focus onsystematically listingthe specific contributions and relevant content of each research work\. The organization follows the taxonomy illustrated in Figure[2](https://arxiv.org/html/2608.14377#S1.F2), detailing applications across the6 stakeholder groupsand19 specific tasks\.

### B\.1Applications for Athletes and Trainers

Exercise and Training Plans\.Recent AI coaches powered by LLMs have significantly streamlined the generation of effective training plans\. Many works have used LLMs to generate exercise prescriptions for various health conditions and fitness goals\([27](https://arxiv.org/html/2608.14377#bib.bib19);[48](https://arxiv.org/html/2608.14377#bib.bib51);[199](https://arxiv.org/html/2608.14377#bib.bib18);[190](https://arxiv.org/html/2608.14377#bib.bib50);[161](https://arxiv.org/html/2608.14377#bib.bib301);[126](https://arxiv.org/html/2608.14377#bib.bib303)\), including weight management\([221](https://arxiv.org/html/2608.14377#bib.bib24)\), resistance and jump training\([278](https://arxiv.org/html/2608.14377#bib.bib20);[86](https://arxiv.org/html/2608.14377#bib.bib23);[189](https://arxiv.org/html/2608.14377#bib.bib300)\), upper body and core training\([25](https://arxiv.org/html/2608.14377#bib.bib28);[63](https://arxiv.org/html/2608.14377#bib.bib37)\), and nutritional strategies for ultra\-endurance sports\([200](https://arxiv.org/html/2608.14377#bib.bib40);[235](https://arxiv.org/html/2608.14377#bib.bib41)\)\. LLMs can also help trainers develop fitness programs for specific patient populations, including obese people\([133](https://arxiv.org/html/2608.14377#bib.bib31);[197](https://arxiv.org/html/2608.14377#bib.bib302)\), those with chronic diseases\([295](https://arxiv.org/html/2608.14377#bib.bib298);[184](https://arxiv.org/html/2608.14377#bib.bib53);[4](https://arxiv.org/html/2608.14377#bib.bib26)\), and those with epilepsy\([215](https://arxiv.org/html/2608.14377#bib.bib52);[214](https://arxiv.org/html/2608.14377#bib.bib42)\)\. In terms of methods, some studies employ digital twins with multimodal outputs\([256](https://arxiv.org/html/2608.14377#bib.bib30)\), behavioral science theories\([90](https://arxiv.org/html/2608.14377#bib.bib43);[113](https://arxiv.org/html/2608.14377#bib.bib27);[59](https://arxiv.org/html/2608.14377#bib.bib44)\), and RAG technology\([314](https://arxiv.org/html/2608.14377#bib.bib39);[119](https://arxiv.org/html/2608.14377#bib.bib29)\); in terms of effectiveness, some emphasize the importance of personalization and contextual understanding\([55](https://arxiv.org/html/2608.14377#bib.bib22);[325](https://arxiv.org/html/2608.14377#bib.bib36);[84](https://arxiv.org/html/2608.14377#bib.bib45)\), focus on acceptance, trust, and quality\([62](https://arxiv.org/html/2608.14377#bib.bib38);[262](https://arxiv.org/html/2608.14377#bib.bib25)\), and foster user self\-reflection\([140](https://arxiv.org/html/2608.14377#bib.bib21)\)\. In addition, some studies have designed AI coaches tailored to the specific requirements of individual sports, such as boxing\([23](https://arxiv.org/html/2608.14377#bib.bib47)\)and table tennis\([156](https://arxiv.org/html/2608.14377#bib.bib32);[157](https://arxiv.org/html/2608.14377#bib.bib33)\)\. Within these sport\-specific domains, two prominent tasks have emerged to provide professional\-grade feedback\. One isexpert commentary generation, which utilizes MLLMs to provide evaluative insights and skill\-level\-aware feedback for basketball\([229](https://arxiv.org/html/2608.14377#bib.bib49)\)and soccer\([12](https://arxiv.org/html/2608.14377#bib.bib48)\)based on video demonstrations\. The other ismotion instruction generation, where frameworks like MAAIG\([300](https://arxiv.org/html/2608.14377#bib.bib34)\)and the reference\-based CoachMe\([301](https://arxiv.org/html/2608.14377#bib.bib35)\)automatically derive technical corrective guidance from 3D skeletal data to assist athletes in figure skating and boxing\.

Sports Injury and Rehabilitation\.Diagnosing and treating sports injuries necessitates extensive interdisciplinary knowledge, and LLMs have demonstrated a broad understanding of this domain\([85](https://arxiv.org/html/2608.14377#bib.bib55);[153](https://arxiv.org/html/2608.14377#bib.bib60)\), encompassing orthopedics\([64](https://arxiv.org/html/2608.14377#bib.bib58)\)and sports rehabilitation\([164](https://arxiv.org/html/2608.14377#bib.bib56);[165](https://arxiv.org/html/2608.14377#bib.bib59)\)\. Specifically, these models assist in providing preventive advice\([326](https://arxiv.org/html/2608.14377#bib.bib54)\), identifying and labeling medical information\([21](https://arxiv.org/html/2608.14377#bib.bib61)\), and supporting diagnostic imaging\([153](https://arxiv.org/html/2608.14377#bib.bib60)\)and data processing\([175](https://arxiv.org/html/2608.14377#bib.bib62)\)\. Furthermore, they play a crucial role in clinical decision\-making\([216](https://arxiv.org/html/2608.14377#bib.bib63)\), surgical treatment planning\([39](https://arxiv.org/html/2608.14377#bib.bib57)\), and enabling patient outcome prediction\([3](https://arxiv.org/html/2608.14377#bib.bib64)\)and medical oversight\.

Sports Psychology and Behavior\.LLMs have demonstrated initial potential in this field, capable of answering sports\-related questions\([258](https://arxiv.org/html/2608.14377#bib.bib68)\), assessing cognitive abilities\([329](https://arxiv.org/html/2608.14377#bib.bib69)\), and summarizing psychological theories\([183](https://arxiv.org/html/2608.14377#bib.bib66)\)\. A significant line of research integrates these models with wearable technology for real\-time monitoring and behavioral modeling\([67](https://arxiv.org/html/2608.14377#bib.bib71);[105](https://arxiv.org/html/2608.14377#bib.bib73);[108](https://arxiv.org/html/2608.14377#bib.bib74);[168](https://arxiv.org/html/2608.14377#bib.bib70)\)\. Furthermore, LLMs provide assistance in specialized areas such as managing exercise addiction\([244](https://arxiv.org/html/2608.14377#bib.bib65)\)and facilitate behavioral interventions by enhancing motivation for sports participation\([236](https://arxiv.org/html/2608.14377#bib.bib72)\)and delivering sleep education\([162](https://arxiv.org/html/2608.14377#bib.bib67)\)\.

### B\.2Applications for Coaches and Educators

Action Spotting and Recognition\.In this task, the majority of approaches employ MLLMs to facilitate direct action spotting and recognition\. Specific methodologies include keyframe sampling\([120](https://arxiv.org/html/2608.14377#bib.bib146)\), contrastive pretraining\([233](https://arxiv.org/html/2608.14377#bib.bib148)\), and domain adaptation\([112](https://arxiv.org/html/2608.14377#bib.bib93)\), primarily focusing on soccer\. Applications extend to other sports, with studies fine\-tuning MLLMs for rally\-sequence recognition in tennis\([251](https://arxiv.org/html/2608.14377#bib.bib147)\), utilizing high\-frame\-rate modeling for gymnastics and diving\([138](https://arxiv.org/html/2608.14377#bib.bib152)\), and performing scene\-level classification in rugby\([182](https://arxiv.org/html/2608.14377#bib.bib149)\)\. Beyond visual\-centric approaches, textual signals such as commentary have also been leveraged for spotting tasks\([28](https://arxiv.org/html/2608.14377#bib.bib153)\)\. Additionally, benchmarks like ActionAtlas\([217](https://arxiv.org/html/2608.14377#bib.bib150)\)and F³Set\([150](https://arxiv.org/html/2608.14377#bib.bib151)\)provide platforms for evaluating fine\-grained recognition capabilities\.

Sports Action Quality Assessment\.Recent endeavors in this domain focus on fine\-tuning large multimodal models to facilitate personalized fitness evaluation\([58](https://arxiv.org/html/2608.14377#bib.bib170)\)\. Researchers have also proposed unified agent frameworks tailored for open\-set and user\-specific assessments\([247](https://arxiv.org/html/2608.14377#bib.bib171)\), and established fine\-grained datasets incorporating LLM\-based evaluations for functional movement screening\([290](https://arxiv.org/html/2608.14377#bib.bib172)\)\. In specific sports such as figure skating, MLLMs have been used to quantify technical and program scores, providing critical support for both athlete training and referee judging\([264](https://arxiv.org/html/2608.14377#bib.bib169)\)\.

Sports Tactics and Strategies\.Early work in this field was largely text\-centric, converting structured data into natural language for tactical modeling, such as fine\-tuning LLMs on event sequences\([26](https://arxiv.org/html/2608.14377#bib.bib154)\), transforming cycling commentary into graph representations\([106](https://arxiv.org/html/2608.14377#bib.bib160)\), or parsing play\-by\-play logs into spatial spray charts\([171](https://arxiv.org/html/2608.14377#bib.bib157)\)\. Subsequent studies have examined the analytical reasoning of LLMs, including computing team scores from play\-by\-play data\([98](https://arxiv.org/html/2608.14377#bib.bib162)\)and aggregating narratives for score inference\([99](https://arxiv.org/html/2608.14377#bib.bib161)\)\. More recent approaches integrate diverse structural and spatial information for richer tactical reasoning, utilizing sketch\-based LLM agents for interactive tactic design\([151](https://arxiv.org/html/2608.14377#bib.bib155)\), graph LLMs for zero\-shot generalization\([145](https://arxiv.org/html/2608.14377#bib.bib156)\), and multi\-agent systems that combine video detection with statistical inference\([313](https://arxiv.org/html/2608.14377#bib.bib158)\)\.

Game and Player Performance Prediction\.Early research for this task predominantly utilized BERT\-based models\([57](https://arxiv.org/html/2608.14377#bib.bib237)\)to predict specific player actions or traits, such as forecasting badminton strokes from skeleton poses\([104](https://arxiv.org/html/2608.14377#bib.bib163)\)and analyzing NBA players’ performance deviations based on pre\-game interview transcripts\([187](https://arxiv.org/html/2608.14377#bib.bib164)\)\. More recent advancements leverage LLMs to synthesize diverse data sources for broader game outcome predictions\. Specific applications include predicting basketball results via in\-context learning on social media data\([240](https://arxiv.org/html/2608.14377#bib.bib166)\), explaining handball match outcomes through feature attribution summarization\([66](https://arxiv.org/html/2608.14377#bib.bib165)\), and fusing features from multimodal pre\-match reports to enhance cricket score predictions\([18](https://arxiv.org/html/2608.14377#bib.bib167)\)\.

Sports Education\.In this domain, LLMs are extensively applied to assist educators with lesson planning\([75](https://arxiv.org/html/2608.14377#bib.bib175);[274](https://arxiv.org/html/2608.14377#bib.bib299)\), designing interactive activities\([50](https://arxiv.org/html/2608.14377#bib.bib177)\), providing formative feedback\([117](https://arxiv.org/html/2608.14377#bib.bib174)\), and automating assignment creation\([116](https://arxiv.org/html/2608.14377#bib.bib173)\)\. Research also emphasizes curricular support through automated visualization, synthetic dataset creation\([65](https://arxiv.org/html/2608.14377#bib.bib176)\), and sensing\-driven feedback mechanisms\([71](https://arxiv.org/html/2608.14377#bib.bib180)\)\. Furthermore, student\-centered applications focus on personalized exercise planning and mental health support\([316](https://arxiv.org/html/2608.14377#bib.bib179)\), while other studies investigate the inclusion, trust, and acceptance of tools like ChatGPT in educational practice\([30](https://arxiv.org/html/2608.14377#bib.bib178)\)\.

### B\.3Applications for Referees

Sports Refereeing\.In this task, large models are leveraged to enhance fairness and transparency in decision\-making\. A prominent example is X\-VARS\([93](https://arxiv.org/html/2608.14377#bib.bib181)\), which introduces an explainable Video Assistant Referee system\. By fine\-tuning MLLMs on expert\-annotated foul data, this system provides textual rationales alongside decisions, thereby demonstrating clear benefits in improving decision accuracy, consistency, and trust among referees\([91](https://arxiv.org/html/2608.14377#bib.bib182)\)\.

### B\.4Applications for Fans and Social Media

Sports Commentary Generation\.Recent studies leverage LLMs to automatically produce commentary, offering fans an enhanced viewing experience\. Most approaches adopt an agentic framework, where LLMs are prompted with extracted match information such as detected key events\([194](https://arxiv.org/html/2608.14377#bib.bib105);[7](https://arxiv.org/html/2608.14377#bib.bib96);[6](https://arxiv.org/html/2608.14377#bib.bib98);[219](https://arxiv.org/html/2608.14377#bib.bib106)\), player and ball tracking data\([7](https://arxiv.org/html/2608.14377#bib.bib96);[6](https://arxiv.org/html/2608.14377#bib.bib98);[261](https://arxiv.org/html/2608.14377#bib.bib95)\), player background information\([174](https://arxiv.org/html/2608.14377#bib.bib94);[283](https://arxiv.org/html/2608.14377#bib.bib103)\), audio signals\([72](https://arxiv.org/html/2608.14377#bib.bib99)\), and external knowledge\([136](https://arxiv.org/html/2608.14377#bib.bib101)\)\. Beyond agentic frameworks, recent work explores end\-to\-end training to improve quality, either by fine\-tuning MLLMs\([277](https://arxiv.org/html/2608.14377#bib.bib91);[47](https://arxiv.org/html/2608.14377#bib.bib102);[17](https://arxiv.org/html/2608.14377#bib.bib104);[112](https://arxiv.org/html/2608.14377#bib.bib93)\)or designing novel architectures for better temporal alignment\([310](https://arxiv.org/html/2608.14377#bib.bib109);[209](https://arxiv.org/html/2608.14377#bib.bib90);[304](https://arxiv.org/html/2608.14377#bib.bib89)\)\. Additionally, efforts target complementary directions like constructing benchmarks\([34](https://arxiv.org/html/2608.14377#bib.bib100);[74](https://arxiv.org/html/2608.14377#bib.bib97)\), enabling real\-time streaming\([34](https://arxiv.org/html/2608.14377#bib.bib100);[60](https://arxiv.org/html/2608.14377#bib.bib314);[306](https://arxiv.org/html/2608.14377#bib.bib282)\), supporting multilingual commentary\([219](https://arxiv.org/html/2608.14377#bib.bib106)\), generating personalized narratives\([6](https://arxiv.org/html/2608.14377#bib.bib98)\), and advancing commercial applications\([17](https://arxiv.org/html/2608.14377#bib.bib104)\)\.

Sports Highlight Generation\.Most research in this domain employs MLLMs to facilitate highlight generation via key event detection, incorporating techniques such as action spotting\([16](https://arxiv.org/html/2608.14377#bib.bib110)\)and multimodal fusion with textual encoding\([51](https://arxiv.org/html/2608.14377#bib.bib115)\)\. Some approaches explicitly leverage commentary transcripts or role\-play prompting to enhance event classification accuracy\([226](https://arxiv.org/html/2608.14377#bib.bib112);[114](https://arxiv.org/html/2608.14377#bib.bib113)\)\. Other works use MLLMs for summary and caption generation to support social media highlights\([172](https://arxiv.org/html/2608.14377#bib.bib111)\), or for personalized highlight generation via simulated watch histories and preference descriptions\([128](https://arxiv.org/html/2608.14377#bib.bib114)\)\.

Sports News Generation\.Early work primarily focused on summarizing unstructured text commentary, utilizing LLMs to select and rewrite key segments\([266](https://arxiv.org/html/2608.14377#bib.bib121);[267](https://arxiv.org/html/2608.14377#bib.bib116)\)or to extract salient events\([223](https://arxiv.org/html/2608.14377#bib.bib118)\)\. Recent advancements have extended these capabilities to process structured data, employing chain\-of\-thought prompting to interpret statistical tables\([44](https://arxiv.org/html/2608.14377#bib.bib119)\)or leveraging CSV inputs to generate comprehensive game reports\([43](https://arxiv.org/html/2608.14377#bib.bib120)\)\. Beyond summarization,[40](https://arxiv.org/html/2608.14377#bib.bib117)propose an insight\-driven approach where high\-level user queries guide LLMs to construct narrative episodes enriched with data visualizations\.

Sports Narratives and Storytelling\.Recent research in this field focuses on generating factually consistent highlight narrations through advanced prompt engineering techniques\([222](https://arxiv.org/html/2608.14377#bib.bib123)\)\. Significant advancements have also been made in leveraging multimodal embedded visualizations and personalized narratives to facilitate tactical understanding for general audiences\([127](https://arxiv.org/html/2608.14377#bib.bib122);[143](https://arxiv.org/html/2608.14377#bib.bib124)\)\. Furthermore, other works adapt narrative generation pipelines to platform\-specific contexts, enabling the production of personalized reports and posts designed for large\-scale fan interaction\([224](https://arxiv.org/html/2608.14377#bib.bib125);[17](https://arxiv.org/html/2608.14377#bib.bib104)\)\.

Public Opinion Analysis in Sports\.In this field, LLMs have been applied to identify key discussion themes within large\-scale social media datasets\([203](https://arxiv.org/html/2608.14377#bib.bib126)\)\. Significant progress has been made in performing fine\-grained sentiment and stance detection via aspect\-based analysis\([202](https://arxiv.org/html/2608.14377#bib.bib128)\)and utilizing few\-shot prompting to analyze controversial topics\([211](https://arxiv.org/html/2608.14377#bib.bib129)\)\. Additionally, researchers employ social science frameworks combined with LLMs to examine user acceptance and perceptions of emerging AI tools within the sports community\([11](https://arxiv.org/html/2608.14377#bib.bib127)\)\.

Sports Models and Systems\.LLMs have been extensively applied to build sports chatbots for interactive dialogue\([198](https://arxiv.org/html/2608.14377#bib.bib130)\)or co\-viewing experiences\([118](https://arxiv.org/html/2608.14377#bib.bib131)\), often incorporating dialogue state tracking for sports\-specific contexts\([237](https://arxiv.org/html/2608.14377#bib.bib134)\)\. General sports models, particularly for soccer, are developed using diverse techniques including fine\-tuning\([255](https://arxiv.org/html/2608.14377#bib.bib144);[73](https://arxiv.org/html/2608.14377#bib.bib141);[208](https://arxiv.org/html/2608.14377#bib.bib143)\), knowledge graph integration\([31](https://arxiv.org/html/2608.14377#bib.bib188)\), and multi\-agent LLM architectures\([207](https://arxiv.org/html/2608.14377#bib.bib142)\)\. In the area of search and retrieval, interactive agents combine LLMs with offline query understanding and online decision\-making\([115](https://arxiv.org/html/2608.14377#bib.bib132)\)\. Furthermore, RAG systems are utilized to query sports knowledge from natural language sources\([227](https://arxiv.org/html/2608.14377#bib.bib133);[242](https://arxiv.org/html/2608.14377#bib.bib135);[243](https://arxiv.org/html/2608.14377#bib.bib136);[230](https://arxiv.org/html/2608.14377#bib.bib137);[231](https://arxiv.org/html/2608.14377#bib.bib138);[265](https://arxiv.org/html/2608.14377#bib.bib140)\)\. Extended applications also include online information retrieval for cricket\([279](https://arxiv.org/html/2608.14377#bib.bib139)\)and fine\-grained video retrieval for sports such as gymnastics and diving\([80](https://arxiv.org/html/2608.14377#bib.bib145)\)\.

### B\.5Applications for Researchers

Sports Academic Writing\.In fields such as sports science and medicine, LLMs like ChatGPT are increasingly utilized to generate outlines, draft abstracts, and provide grammar and style suggestions\([125](https://arxiv.org/html/2608.14377#bib.bib78);[83](https://arxiv.org/html/2608.14377#bib.bib79)\)\. However, the literature emphasizes the need for caution due to inherent risks in content accuracy\([54](https://arxiv.org/html/2608.14377#bib.bib77)\), the reliability of generated references\([5](https://arxiv.org/html/2608.14377#bib.bib76)\), calculation precision\([170](https://arxiv.org/html/2608.14377#bib.bib75)\), and originality\.

### B\.6Applications for the Sports Industry

Sports Management\.Large models are increasingly applied to streamline diverse functions within sports organizations\. In financial management, research demonstrates their ability to conduct interviews, extract key themes, and develop tailored organizational strategies\([81](https://arxiv.org/html/2608.14377#bib.bib81)\)\. For database management, LLMs are used to structure and analyze complex club data to enhance operational efficiency\([167](https://arxiv.org/html/2608.14377#bib.bib82)\)\. In facility management, these models support human\-computer dialogue to facilitate site selection and knowledge acquisition\([218](https://arxiv.org/html/2608.14377#bib.bib83)\)\. Furthermore, they are employed to simulate future industry scenarios and provide robust support for data\-driven strategic decisions\([82](https://arxiv.org/html/2608.14377#bib.bib80)\)\.

Sports Talent Scouting\.Large models enhance this domain by introducing more objective and data\-driven methodologies for athlete evaluation\([163](https://arxiv.org/html/2608.14377#bib.bib88)\)\. Recent research has deployed LLMs to analyze complex player datasets\([210](https://arxiv.org/html/2608.14377#bib.bib87)\)and convert unstructured scouting reports into searchable, structured knowledge formats\. Furthermore, some work combines large models with RAG strategies to optimize the integration of diverse information sources\([160](https://arxiv.org/html/2608.14377#bib.bib86)\)\.

Sports Tourism\.Large models are increasingly leveraged in this domain to enhance intelligence and personalization, offering solutions for virtual guides, information assistants, and community building\([166](https://arxiv.org/html/2608.14377#bib.bib85)\)\. Research also highlights the role of these models in improving operational efficiency and fan engagement during major events\. Notably, LLMs also show strong potential in the specialized sector of esports tourism\([302](https://arxiv.org/html/2608.14377#bib.bib84)\)\.

## Appendix CMore Details on Datasets for Large Models in Sports

In this section, we further provide a comprehensive introduction to the datasets for large models in sports, as an extension of the main discussion in Section[3](https://arxiv.org/html/2608.14377#S3)\.

### C\.1Task\-Specific Datasets

This subsection echoes Section[3\.1](https://arxiv.org/html/2608.14377#S3.SS1)of the paper and provides a further overview of these task\-specific datasets\. Table[3](https://arxiv.org/html/2608.14377#A3.T3)and Table[4](https://arxiv.org/html/2608.14377#A3.T4)present the datasets related to specific tasks for 5 sports stakeholder groups: athletes and trainers, coaches and educators, referees, fans and social media, and the sports industry\. The information covers dataset names, involved sports types, data modalities, methods used in the papers, corresponding large models, best achieved performance results with their respective evaluation metrics, and the availability of open\-source links, which can be directly accessed by clicking in the table\.

These datasets are unevenly distributed across tasks\.From the perspective of target users, datasets for coaches and fans are relatively abundant, while those for referees, researchers, and the sports industry are relatively scarce\. In terms of task types, early tasks in traditional computer vision and natural language processing, such as action spotting and recognition and sports commentary generation, have received more research attention and have richer datasets, whereas tasks like public opinion analysis in sports and sports talent scouting lack open\-source data, reflecting an imbalance in scholarly focus across different tasks\.

These datasets are unevenly distributed across sports\.Popular sports such as soccer, basketball, and badminton receive more attention and have richer datasets, whereas niche sports like track and field, aquatics, and even esports are severely underrepresented\. Nevertheless, these underexplored areas hold research value and warrant further expansion and investigation\.

These datasets are unevenly distributed across modalities\.Video and text are the most common modalities in sports datasets, while audio, sensor data \(e\.g\., IMU\), and skeletal data are relatively scarce\. This reflects the current research focus on video in the sports domain and also highlights the untapped potential of other modalities\.

These datasets are generally used with pre\-existing models rather than being used to train or fine\-tune models\.Most researchers tend to rely on the inherent capabilities of large models, which explains the widespread use of powerful closed\-source models such as GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)\. This trend reflects both the scarcity of sports data and the significant value of constructing dedicated sports datasets and models, emphasizing the need for more attention to the field of large models in sports\.

TaskDatasetSportsModalMethodRelated Large ModelPerformanceLinkAthletes and TrainersPLAYourSkatingCoach\([36](https://arxiv.org/html/2608.14377#bib.bib233)\)Figure SkatingV, TMAAIG\([300](https://arxiv.org/html/2608.14377#bib.bib34)\)T5\([206](https://arxiv.org/html/2608.14377#bib.bib239)\)22\.08 \(METEOR\)✗PACE\([259](https://arxiv.org/html/2608.14377#bib.bib234)\)FitnessT[90](https://arxiv.org/html/2608.14377#bib.bib43)LaMDA\([252](https://arxiv.org/html/2608.14377#bib.bib235)\)3\.78 ± 1\.00 / 5\.00 \(Likert\)[✓](https://github.com/fitllm/classifiers)NSCA\-CSCS\([179](https://arxiv.org/html/2608.14377#bib.bib232)\)FitnessTPH\-LLM\([48](https://arxiv.org/html/2608.14377#bib.bib51)\)Gemini Ultra 1\.0\([248](https://arxiv.org/html/2608.14377#bib.bib1)\)88\.00 \(Acc\)✗T3Set\([156](https://arxiv.org/html/2608.14377#bib.bib32)\)Table TennisV, M, TSenseCoach\([156](https://arxiv.org/html/2608.14377#bib.bib32)\)Llama 3\.3\-70B\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)51\.64 \(P@6\-S1L\)[✓](https://github.com/jima-cs/T3Set)SCD\([84](https://arxiv.org/html/2608.14377#bib.bib45)\)SoccerT[84](https://arxiv.org/html/2608.14377#bib.bib45)BERT\([57](https://arxiv.org/html/2608.14377#bib.bib237)\)85\.64 \(BERTScore\)✗Custom Dataset\([157](https://arxiv.org/html/2608.14377#bib.bib33)\)Table TennisV, I, S, T[157](https://arxiv.org/html/2608.14377#bib.bib33)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)67\.40 \(Acc\)[✓](https://github.com/mwlsus/ttcs_by_llm)Ego\-Exo4D\([79](https://arxiv.org/html/2608.14377#bib.bib197)\)SC, BK, CLV, TExpertAF\([12](https://arxiv.org/html/2608.14377#bib.bib48)\)Llama 3\-8B\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)49\.60 \(METEOR\)[✓](https://github.com/thechargedneutron/ExpertAF)Ego\-Exo4D\([79](https://arxiv.org/html/2608.14377#bib.bib197)\)BasketballV, T[229](https://arxiv.org/html/2608.14377#bib.bib49)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)25\.60 \(METEOR\)✗FS\([301](https://arxiv.org/html/2608.14377#bib.bib35)\)Figure SkatingS, TCoachMe\([301](https://arxiv.org/html/2608.14377#bib.bib35)\)T5\([206](https://arxiv.org/html/2608.14377#bib.bib239)\)26\.5 \(BERTScore\)[✓](https://motionxperts.github.io/)BX\([301](https://arxiv.org/html/2608.14377#bib.bib35)\)Boxing36\.9 \(BERTScore\)INJcMedQA2\([315](https://arxiv.org/html/2608.14377#bib.bib240)\)\-T[326](https://arxiv.org/html/2608.14377#bib.bib54)Qwen2\-0\.5B\([250](https://arxiv.org/html/2608.14377#bib.bib241)\)30\.56 \(BLEU\-4\)✗Custom Dataset\([216](https://arxiv.org/html/2608.14377#bib.bib63)\)\-T[216](https://arxiv.org/html/2608.14377#bib.bib63)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)47\.80 \(Cronbach’sα\\alpha\)✗Custom Dataset\([21](https://arxiv.org/html/2608.14377#bib.bib61)\)\-T[21](https://arxiv.org/html/2608.14377#bib.bib61)phi\-3\-mini\([1](https://arxiv.org/html/2608.14377#bib.bib242)\)34\.13 \(Spearman’sρ\\rho\)✗PSYCustom Dataset\([168](https://arxiv.org/html/2608.14377#bib.bib70)\)FitnessTPHIA\([168](https://arxiv.org/html/2608.14377#bib.bib70)\)Gemini 1\.0 Ultra\([248](https://arxiv.org/html/2608.14377#bib.bib1)\)84\.20 \(Acc\)[✓](https://github.com/yahskapar/personal-health-insights-agent)Capture24\([29](https://arxiv.org/html/2608.14377#bib.bib243)\)FitnessMHARGPT\([108](https://arxiv.org/html/2608.14377#bib.bib74)\)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)79\.50 \(F1\)[✓](https://github.com/aiot-lab/HARGPT)In\-the\-Wild\([105](https://arxiv.org/html/2608.14377#bib.bib73)\)FitnessM, TLLaSA\([105](https://arxiv.org/html/2608.14377#bib.bib73)\)Vicuna\-7B\([45](https://arxiv.org/html/2608.14377#bib.bib244)\)79\.95 \(Acc\)[✓](https://github.com/BASHLab/LLaSA)Coaches and EducatorsACTCustom Dataset\([182](https://arxiv.org/html/2608.14377#bib.bib149)\)RugbyI, T[182](https://arxiv.org/html/2608.14377#bib.bib149)LLaVA\-7B\([147](https://arxiv.org/html/2608.14377#bib.bib250)\)63\.10 ± 2\.20 \(F1\)✗ActionAtlas v1\.0\([217](https://arxiv.org/html/2608.14377#bib.bib150)\)56 SportsV[217](https://arxiv.org/html/2608.14377#bib.bib150)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)42\.95 ± 2\.91 \(Acc\)[✓](https://github.com/mrsalehi/action-atlas?tab=readme-ov-file)NSVA Subset\([239](https://arxiv.org/html/2608.14377#bib.bib245)\)BK, AFV, TSV3\.3B\([120](https://arxiv.org/html/2608.14377#bib.bib146)\)Llama 3\.2\-3B\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)85\.60 ± 5\.20 \(BERT F1\)[✓](https://huggingface.co/sportsvision/SV3.3B)FineTennis\([251](https://arxiv.org/html/2608.14377#bib.bib147)\)TennisV[251](https://arxiv.org/html/2608.14377#bib.bib147)Video\-LLaMA2\-7B\([41](https://arxiv.org/html/2608.14377#bib.bib246)\)76\.00 \(Edit Score\)[✓](https://github.com/bigcrushes/videollama2_tennis)SoccerNet\-v2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)SoccerVSoccer\-CLIP\([233](https://arxiv.org/html/2608.14377#bib.bib148)\)ViT\-B/32\([61](https://arxiv.org/html/2608.14377#bib.bib248)\)75\.70 \(t\-AmAP\)✗Tennis7\([312](https://arxiv.org/html/2608.14377#bib.bib249)\)TennisV93\.80 \(Acc\)F³Set\([150](https://arxiv.org/html/2608.14377#bib.bib151)\)TN, BM, TTV, TF³ED\([150](https://arxiv.org/html/2608.14377#bib.bib151)\)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)75\.20 \(F1elm\)[✓](https://github.com/F3Set/F3Set)Video\-MME\([68](https://arxiv.org/html/2608.14377#bib.bib194)\)SC,BK,GY,DVV, TF\-16\([138](https://arxiv.org/html/2608.14377#bib.bib152)\)LLaVA\-OV\([130](https://arxiv.org/html/2608.14377#bib.bib251)\)65\.00 \(Acc\)[✓](https://github.com/bytedance/F-16)SoccerNet\-v2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)SoccerT[28](https://arxiv.org/html/2608.14377#bib.bib153)Llama 3\.1\-8B\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)64\.50 \(mAP\)✗SoccerNet\-v2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)SoccerV, T[112](https://arxiv.org/html/2608.14377#bib.bib93)LLaVA\-NeXT\-Video\([132](https://arxiv.org/html/2608.14377#bib.bib252)\)63\.50 \(Acc\)✗UCI\-HAR\([8](https://arxiv.org/html/2608.14377#bib.bib272)\)FitnessM[71](https://arxiv.org/html/2608.14377#bib.bib180)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)92\.30 \(Acc\)✗SoccerNet\([76](https://arxiv.org/html/2608.14377#bib.bib281)\)SoccerV, A, T[16](https://arxiv.org/html/2608.14377#bib.bib110)Video\-LLaMA\([311](https://arxiv.org/html/2608.14377#bib.bib284)\)87\.00 \(F1\)✗AQAFis\-V\([291](https://arxiv.org/html/2608.14377#bib.bib265)\)Figure SkatingV, A, T[264](https://arxiv.org/html/2608.14377#bib.bib169)InternVL2\([38](https://arxiv.org/html/2608.14377#bib.bib268)\)84\.00 \(Spearman’sρ\\rho\)[✓](https://github.com/ycwfs/FigureSkating-Quality-Assessment)FS1000\([288](https://arxiv.org/html/2608.14377#bib.bib266)\)90\.00 \(Spearman’sρ\\rho\)FineFS\([109](https://arxiv.org/html/2608.14377#bib.bib267)\)76\.00 \(Spearman’sρ\\rho\)Fitness\-AQA\([192](https://arxiv.org/html/2608.14377#bib.bib269)\)FitnessV, T[58](https://arxiv.org/html/2608.14377#bib.bib170)LLaVA\-Video\-7B\([318](https://arxiv.org/html/2608.14377#bib.bib270)\)22\.82 \(mAP\)[✓](https://github.com/GaetanoDibenedetto/UMAP25)FMS\([247](https://arxiv.org/html/2608.14377#bib.bib171)\)FitnessV, TFitnessAgent\([247](https://arxiv.org/html/2608.14377#bib.bib171)\)ChatGLM4\([77](https://arxiv.org/html/2608.14377#bib.bib271)\)39\.34 \(Acc\)✗LLM\-FMS\([290](https://arxiv.org/html/2608.14377#bib.bib172)\)FitnessV, T[290](https://arxiv.org/html/2608.14377#bib.bib172)\-91\.00 \(Acc\)Custom Dataset\([26](https://arxiv.org/html/2608.14377#bib.bib154)\)SoccerTTacticalGPT\([26](https://arxiv.org/html/2608.14377#bib.bib154)\)GPT\-NeoX\-20B\([20](https://arxiv.org/html/2608.14377#bib.bib257)\)50\.00 \(Acc\)✗STATS SportVU[152](https://arxiv.org/html/2608.14377#bib.bib258)BasketballI, TSmartboard\([151](https://arxiv.org/html/2608.14377#bib.bib155)\)GPT\-4V\([185](https://arxiv.org/html/2608.14377#bib.bib259)\)\-✗Custom Dataset\([195](https://arxiv.org/html/2608.14377#bib.bib255)\)BaseballT[171](https://arxiv.org/html/2608.14377#bib.bib157)Curie\([22](https://arxiv.org/html/2608.14377#bib.bib256)\)97\.00 \(Acc\)[✓](https://github.com/tony-baseball/Hitting-Spray-Charts-with-ggplot-and-ggplotly)SportsMetrics\([98](https://arxiv.org/html/2608.14377#bib.bib162)\)BK, AFT[97](https://arxiv.org/html/2608.14377#bib.bib159);[98](https://arxiv.org/html/2608.14377#bib.bib162)Gemini\-Pro[248](https://arxiv.org/html/2608.14377#bib.bib1)32\.30 \(Δ\\DeltaGScore\)[✓](https://github.com/YebowenHu/SportsMetrics)Custom Dataset\([106](https://arxiv.org/html/2608.14377#bib.bib160)\)CyclingT[106](https://arxiv.org/html/2608.14377#bib.bib160)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)\-✗Custom Dataset\([99](https://arxiv.org/html/2608.14377#bib.bib161)\)BasketballTSportsGen\([99](https://arxiv.org/html/2608.14377#bib.bib161)\)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)98\.41 \(DnC\-10\)[✓](https://github.com/YebowenHu/SportsGen)Custom Dataset\([313](https://arxiv.org/html/2608.14377#bib.bib158)\)BadmintonV, TChatMatch\([313](https://arxiv.org/html/2608.14377#bib.bib158)\)GPT\-3\.5\-turbo\([186](https://arxiv.org/html/2608.14377#bib.bib260)\)98\.84 \(Acc\)✗TACBasketball\-Instants[309](https://arxiv.org/html/2608.14377#bib.bib254)BasketballI, TTacticExpert\([145](https://arxiv.org/html/2608.14377#bib.bib156)\)Vicuna\-7B\-v1\.5\([45](https://arxiv.org/html/2608.14377#bib.bib244)\)83\.33 \(Macro F1\)✗PRDCustom Dataset\([187](https://arxiv.org/html/2608.14377#bib.bib164)\)BasketballT[187](https://arxiv.org/html/2608.14377#bib.bib164)BERT\([57](https://arxiv.org/html/2608.14377#bib.bib237)\)58\.50 \(Acc\)✗ShuttleSet\([269](https://arxiv.org/html/2608.14377#bib.bib262)\)BadmintonV, TRallyTemPose\([104](https://arxiv.org/html/2608.14377#bib.bib163)\)BERT\([57](https://arxiv.org/html/2608.14377#bib.bib237)\)54\.30 \(Acc\)[✓](https://github.com/MagnusPetersenTbh/RallyTempPose)BadmintonDB\([14](https://arxiv.org/html/2608.14377#bib.bib263)\)62\.80 \(Acc\)Custom Dataset\([240](https://arxiv.org/html/2608.14377#bib.bib166)\)BasketballT[240](https://arxiv.org/html/2608.14377#bib.bib166)GPT\-3\.5\-turbo\([186](https://arxiv.org/html/2608.14377#bib.bib260)\)64\.90 \(Acc\)[✓](https://github.com/gsprint23/DivisionIBasketballTwitter)SportDevs\([238](https://arxiv.org/html/2608.14377#bib.bib261)\)HandballT[66](https://arxiv.org/html/2608.14377#bib.bib165)Mistral\-7B\([110](https://arxiv.org/html/2608.14377#bib.bib264)\)5\.20 \(RMSE\)✗Custom Dataset\([18](https://arxiv.org/html/2608.14377#bib.bib167)\)CricketV, T[18](https://arxiv.org/html/2608.14377#bib.bib167)GPT\-4o mini\([102](https://arxiv.org/html/2608.14377#bib.bib238)\), etc\.86\.30 \(F1\)[✓](https://bit.ly/iplscorepredictions)RefereesREFSoccerNet\-XFoul\([93](https://arxiv.org/html/2608.14377#bib.bib181)\)SoccerV, TX\-VARS\([93](https://arxiv.org/html/2608.14377#bib.bib181);[91](https://arxiv.org/html/2608.14377#bib.bib182)\)Video\-ChatGPT\([158](https://arxiv.org/html/2608.14377#bib.bib273)\)3\.80 / 5\.00 \(Likert\)[✓](https://github.com/heldJan/X-VARS)Table 3:Summary of task\-specific sports datasets related to large models, including athletes and trainers, coaches and educators, and referees\. Task: PLA: exercise and training plans, INJ: sports injury and rehabilitation, PSY: sports psychology and behavior, ACT: action spotting and recognition, AQA: sports action quality assessment, TAC: sports tactics and strategies, PRD: game and player performance prediction, REF: sports refereeing\. Sports: SC: soccer, BK: basketball, CL: sports climbing, AF: American football, TN: tennis, BM: badminton, TT: table tennis, GY: gymnastics, DV: diving\. Modal: V: video, I: image, A: audio, S: skeleton data, M: IMU data, T: text\.TaskDatasetSportsModalMethodRelated Large ModelPerformanceLinkFans and Social MediaCMTCustom Dataset\([72](https://arxiv.org/html/2608.14377#bib.bib99)\)SoccerV, A, T[72](https://arxiv.org/html/2608.14377#bib.bib99)GPT\-3\([22](https://arxiv.org/html/2608.14377#bib.bib256)\)0\.31 \(ROUGE\-L\)[✓](https://github.com/simula/soccer-summarization)SN\-Caption\-test\-align\([209](https://arxiv.org/html/2608.14377#bib.bib90)\)SoccerV, TMatchVoice\([209](https://arxiv.org/html/2608.14377#bib.bib90)\)Llama 3\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)42\.00 \(CIDEr\)[✓](https://haoningwu3639.github.io/MatchTime/)LoL19\([277](https://arxiv.org/html/2608.14377#bib.bib91)\)EsportsT[277](https://arxiv.org/html/2608.14377#bib.bib91)Llama 2 13B\([254](https://arxiv.org/html/2608.14377#bib.bib274)\)\-4\.61 \(BARTScore\)[✓](https://github.com/ArnoZWang/esports-data-to-text)Custom Dataset\([7](https://arxiv.org/html/2608.14377#bib.bib96)\)SoccerV, TAiCommentator\([7](https://arxiv.org/html/2608.14377#bib.bib96)\)GPT\-3\.5\-turbo\([186](https://arxiv.org/html/2608.14377#bib.bib260)\)0\.56 \(Cohen’s d\)✗CommentarySet\([74](https://arxiv.org/html/2608.14377#bib.bib97)\)TF,SC,BK,GY,TT,TNV, T[74](https://arxiv.org/html/2608.14377#bib.bib97)InternVL\-Chat\-2\([38](https://arxiv.org/html/2608.14377#bib.bib268)\)5\.44 \(SCORES\)✗Custom Dataset\([154](https://arxiv.org/html/2608.14377#bib.bib277);[228](https://arxiv.org/html/2608.14377#bib.bib278)\)SoccerTLLM\-Commentator\([47](https://arxiv.org/html/2608.14377#bib.bib102)\)LLaMA 7B\([253](https://arxiv.org/html/2608.14377#bib.bib279)\)92\.00 \(F1\)[✓](https://github.com/paddelcourt/llm-sport-commentator)Custom Dataset\([17](https://arxiv.org/html/2608.14377#bib.bib104)\)GolfV, T[17](https://arxiv.org/html/2608.14377#bib.bib104)Llama 2 7B\([254](https://arxiv.org/html/2608.14377#bib.bib274)\)99\.12 \(ROUGE\-L\)✗TennisSandstone 3B\([206](https://arxiv.org/html/2608.14377#bib.bib239)\)86\.80 \(ROUGE\-L\)American FootballLlama 2 7B\([254](https://arxiv.org/html/2608.14377#bib.bib274)\)86\.80 \(ROUGE\-L\)BH\-Commentary\([310](https://arxiv.org/html/2608.14377#bib.bib109)\)BasketballV, T[310](https://arxiv.org/html/2608.14377#bib.bib109)BERT\([57](https://arxiv.org/html/2608.14377#bib.bib237)\)12\.19 \(CIDEr\)[✓](https://anonymous.4open.science/r/dataset-DC8E)SoccerNet\-Caption\([173](https://arxiv.org/html/2608.14377#bib.bib107)\)SoccerV, TTimeSoccer\([304](https://arxiv.org/html/2608.14377#bib.bib89)\)Llama 2 7B\([254](https://arxiv.org/html/2608.14377#bib.bib274)\)8\.30 \(CIDEr\)[✓](https://vpx-ecnu.github.io/TimeSoccer-Website/)SoccerNet\-V2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)SoccerV, T[112](https://arxiv.org/html/2608.14377#bib.bib93)Claude 3\.5 Sonnet\([10](https://arxiv.org/html/2608.14377#bib.bib275)\)2\.59 / 5\.00 \(Likert\)✗WyScout\([101](https://arxiv.org/html/2608.14377#bib.bib253)\)2\.96 / 5\.00 \(Likert\)LFCBI\([174](https://arxiv.org/html/2608.14377#bib.bib94)\)SoccerV, T[174](https://arxiv.org/html/2608.14377#bib.bib94)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)15\.50 \(MSE\)[✓](https://drive.google.com/drive/folders/1_EqBtLr9YCnRDlB4IS9p69PhnZffTmPx)SoccerTrack\-Commentary\([261](https://arxiv.org/html/2608.14377#bib.bib95)\)SoccerV, I, T[261](https://arxiv.org/html/2608.14377#bib.bib95)GPT\-3\([22](https://arxiv.org/html/2608.14377#bib.bib256)\)33\.84 \(CIDEr\)✗LiveSports\-3K\-CC\([34](https://arxiv.org/html/2608.14377#bib.bib100)\)49 SportsV, A, TLiveCC\([34](https://arxiv.org/html/2608.14377#bib.bib100)\)Qwen2\-VL\-7B\([268](https://arxiv.org/html/2608.14377#bib.bib276)\)40\.08 \(Win Rate\)[✓](https://showlab.github.io/livecc/)SoccerNet\-v2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)SoccerV, A, TSoccerComment\([136](https://arxiv.org/html/2608.14377#bib.bib101)\)Vicuna\-7B\-v1\.5\([45](https://arxiv.org/html/2608.14377#bib.bib244)\)36\.58 \(CIDEr\)✗NBA\-Identity\([283](https://arxiv.org/html/2608.14377#bib.bib103)\)BasketballV, TLLM\-IAVC\([283](https://arxiv.org/html/2608.14377#bib.bib103)\)Llama 3\.2\-3B\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)105\.30 \(CIDEr\)[✓](https://github.com/Zeyu1226-mt/LLM-IAVC)VC\-NBA\-2022\([282](https://arxiv.org/html/2608.14377#bib.bib280)\)150\.70 \(CIDEr\)Custom Dataset\([219](https://arxiv.org/html/2608.14377#bib.bib106)\)CricketV, T[219](https://arxiv.org/html/2608.14377#bib.bib106)GPT\-4o mini\([102](https://arxiv.org/html/2608.14377#bib.bib238)\), etc\.83\.00 \(BERT F1\)✗SoccerNet\-Caption\([173](https://arxiv.org/html/2608.14377#bib.bib107)\)SoccerV, TStreamMind\([60](https://arxiv.org/html/2608.14377#bib.bib314)\)Video\-LLaMA2\-7B\([41](https://arxiv.org/html/2608.14377#bib.bib246)\)82\.04 \(ROUGE\-L\)[✓](https://aka.ms/StreamMind)SoccerNet\([76](https://arxiv.org/html/2608.14377#bib.bib281)\)SoccerV, TVLM\-TSI\([306](https://arxiv.org/html/2608.14377#bib.bib282)\)VideoLLM\-Online\([33](https://arxiv.org/html/2608.14377#bib.bib283)\)39\.10 \(TRACE\)[✓](https://github.com/yukw777/tglg)HLGCricPulse\([226](https://arxiv.org/html/2608.14377#bib.bib112)\)CricketV, T[226](https://arxiv.org/html/2608.14377#bib.bib112)BERT\([57](https://arxiv.org/html/2608.14377#bib.bib237)\)97\.00 \(F1\)✗Custom Dataset\([172](https://arxiv.org/html/2608.14377#bib.bib111)\)SoccerV, ASmartCrop\([172](https://arxiv.org/html/2608.14377#bib.bib111)\)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)\-✗Custom Dataset\([114](https://arxiv.org/html/2608.14377#bib.bib113)\)BaseballTDIAMOND\([114](https://arxiv.org/html/2608.14377#bib.bib113)\)Mistral\-Large\([249](https://arxiv.org/html/2608.14377#bib.bib286)\)76\.50 \(F1\)✗HIPPO\-Video\([128](https://arxiv.org/html/2608.14377#bib.bib114)\)\-V, THiPHer\([128](https://arxiv.org/html/2608.14377#bib.bib114)\)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)76\.60 \(mAP\)[✓](https://github.com/jeongeunnn-e/HIPPO-Video)Custom Dataset\([51](https://arxiv.org/html/2608.14377#bib.bib115)\)CricketV, A, TSportSummarizer\([51](https://arxiv.org/html/2608.14377#bib.bib115)\)DistilBERT\([220](https://arxiv.org/html/2608.14377#bib.bib285)\)0\.93 \(HD\)✗SoccerNet\([76](https://arxiv.org/html/2608.14377#bib.bib281)\)Soccer0\.92 \(HD\)NSGSportsSum2\.0\([266](https://arxiv.org/html/2608.14377#bib.bib121)\)SoccerT[266](https://arxiv.org/html/2608.14377#bib.bib121)RoBERTa\([149](https://arxiv.org/html/2608.14377#bib.bib287)\), etc\.47\.78 \(ROUGE\-L\)[✓](https://github.com/krystalan/SportsSum2.0)SportsSum\([100](https://arxiv.org/html/2608.14377#bib.bib13)\)Soccer47\.49 \(ROUGE\-L\)K\-SportsSum\([267](https://arxiv.org/html/2608.14377#bib.bib116)\)SoccerTKES\([267](https://arxiv.org/html/2608.14377#bib.bib116)\)mT5\([296](https://arxiv.org/html/2608.14377#bib.bib288)\)47\.17 \(ROUGE\-L\)[✓](https://github.com/krystalan/K-SportsSum)SportsSum\([100](https://arxiv.org/html/2608.14377#bib.bib13)\)Soccer47\.79 \(ROUGE\-L\)NBA API\([193](https://arxiv.org/html/2608.14377#bib.bib289)\)BasketballTSNIL\([40](https://arxiv.org/html/2608.14377#bib.bib117)\)GPT\-3\.5\([186](https://arxiv.org/html/2608.14377#bib.bib260)\)63\.00 \(Acc\)[✓](https://github.com/13Lychee/SNIL)ShuttleSet\([269](https://arxiv.org/html/2608.14377#bib.bib262)\)BadmintonTBADGE\([43](https://arxiv.org/html/2608.14377#bib.bib120)\)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)8\.63 / 10\.00 \(LLM\)[✓](https://github.com/AndyChiangSH/BADGE)Custom Dataset\([223](https://arxiv.org/html/2608.14377#bib.bib118)\)CricketT[223](https://arxiv.org/html/2608.14377#bib.bib118)Google Gemini\([248](https://arxiv.org/html/2608.14377#bib.bib1)\)9\.20 / 10\.00 \(ACS\)✗RotoWire\([280](https://arxiv.org/html/2608.14377#bib.bib290)\)BasketballTTree\-of\-Report\([44](https://arxiv.org/html/2608.14377#bib.bib119)\)GPT\-4o mini\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)54\.92 \(CS F1\)✗MLB\([201](https://arxiv.org/html/2608.14377#bib.bib291)\)Baseball62\.99 \(CS F1\)ShuttleSet\+\([44](https://arxiv.org/html/2608.14377#bib.bib119)\)Badminton93\.94 \(CS F1\)NARCustom Dataset\([222](https://arxiv.org/html/2608.14377#bib.bib123)\)SoccerT[222](https://arxiv.org/html/2608.14377#bib.bib123)T5\-large\([206](https://arxiv.org/html/2608.14377#bib.bib239)\)49\.04 \(ROUGE\-L\)✗SportsVU\([127](https://arxiv.org/html/2608.14377#bib.bib122)\)BasketballV, TSportify\([127](https://arxiv.org/html/2608.14377#bib.bib122)\)\-72\.22 \(Acc\)[✓](https://chungyi347.github.io/Sportify/)SoccerSum\([224](https://arxiv.org/html/2608.14377#bib.bib125);[225](https://arxiv.org/html/2608.14377#bib.bib108)\)SoccerV, ASoccerSum\([224](https://arxiv.org/html/2608.14377#bib.bib125)\)GPT\-4 Turbo\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)\-[✓](https://github.com/simula/soccersum)Custom Dataset\([143](https://arxiv.org/html/2608.14377#bib.bib124)\)BasketballVSportsBuddy\([143](https://arxiv.org/html/2608.14377#bib.bib124)\)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)90\.80 \(Acc\)✗OPICustom Dataset\([211](https://arxiv.org/html/2608.14377#bib.bib129)\)SoccerT[211](https://arxiv.org/html/2608.14377#bib.bib129)GPT\-4\-turbo\([2](https://arxiv.org/html/2608.14377#bib.bib2)\)70\.30 \(F1\)✗Custom Dataset\([202](https://arxiv.org/html/2608.14377#bib.bib128)\)SoccerTABSA\([202](https://arxiv.org/html/2608.14377#bib.bib128)\)RoBERTa\([149](https://arxiv.org/html/2608.14377#bib.bib287)\)80\.00 \(F1\)[✓](https://github.com/TyrealQ/Experience-is-all-you-need)MODCustom Dataset\([227](https://arxiv.org/html/2608.14377#bib.bib133)\)SoccerT[227](https://arxiv.org/html/2608.14377#bib.bib133)GPT\-3\.5\([186](https://arxiv.org/html/2608.14377#bib.bib260)\)71\.40 \(Acc\)✗SoccerNet\([76](https://arxiv.org/html/2608.14377#bib.bib281)\)SoccerV, A, ISoccerRAG\([242](https://arxiv.org/html/2608.14377#bib.bib135);[243](https://arxiv.org/html/2608.14377#bib.bib136)\)GPT\-4\([2](https://arxiv.org/html/2608.14377#bib.bib2)\), etc\.80\.00 \(Acc\)[✓](https://github.com/simula/soccer-rag)Custom Dataset\([115](https://arxiv.org/html/2608.14377#bib.bib132)\)SoccerT[115](https://arxiv.org/html/2608.14377#bib.bib132)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)88\.25 \(Precision\)✗Custom Dataset\([279](https://arxiv.org/html/2608.14377#bib.bib139)\)CricketT[279](https://arxiv.org/html/2608.14377#bib.bib139)Copilot\([35](https://arxiv.org/html/2608.14377#bib.bib292)\)100\.00 \(Acc\)✗TF\-CoVR\([80](https://arxiv.org/html/2608.14377#bib.bib145)\)GY, DVV, TTF\-CoVR\-Base\([80](https://arxiv.org/html/2608.14377#bib.bib145)\)BLIP\([135](https://arxiv.org/html/2608.14377#bib.bib293)\)23\.02 \(mAP@10\)[✓](https://github.com/UCF-CRCV/TF-CoVR)KICK\([237](https://arxiv.org/html/2608.14377#bib.bib134)\)SoccerT[237](https://arxiv.org/html/2608.14377#bib.bib134)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)15\.86 \(JGA\)[✓](https://github.com/ezzy4me/KICK)SoccerNet\-XFoul\([93](https://arxiv.org/html/2608.14377#bib.bib181)\)SoccerV, TSoccerChat\([73](https://arxiv.org/html/2608.14377#bib.bib141)\)Qwen2\-VL\-7B\([268](https://arxiv.org/html/2608.14377#bib.bib276)\)6\.81 / 10\.00 \(LLM\)[✓](https://github.com/simula/SoccerChat)SoccerNet\-v2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)6\.42 / 10\.00 \(LLM\)SoccerBench\([207](https://arxiv.org/html/2608.14377#bib.bib142)\)SoccerV, A, TSoccerAgent\([207](https://arxiv.org/html/2608.14377#bib.bib142)\)DeepSeek\-v3\([146](https://arxiv.org/html/2608.14377#bib.bib294)\)60\.90 \(Acc\)[✓](https://github.com/jyrao/SoccerAgent)SoccerNet\-v2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)SoccerV, TMatchVision\([208](https://arxiv.org/html/2608.14377#bib.bib143)\)Llama 3\-8B\([78](https://arxiv.org/html/2608.14377#bib.bib236)\)80\.10 \(Acc\)[✓](https://jyrao.github.io/UniSoccer/)SN\-Caption\-test\-align\([209](https://arxiv.org/html/2608.14377#bib.bib90)\)44\.18 \(CIDEr\)MVFoul\([92](https://arxiv.org/html/2608.14377#bib.bib295)\)44\.00 \(Acc\)SoccerNet\-V2\([52](https://arxiv.org/html/2608.14377#bib.bib247)\)SoccerV, T[112](https://arxiv.org/html/2608.14377#bib.bib93)LLaVA\-NeXT\-Video\([132](https://arxiv.org/html/2608.14377#bib.bib252)\)83\.76 \(Acc\)✗WyScout\([101](https://arxiv.org/html/2608.14377#bib.bib253)\)81\.83 \(Acc\)Gym\-QA\([31](https://arxiv.org/html/2608.14377#bib.bib188)\)GymnasticsV, TFineQuest\([31](https://arxiv.org/html/2608.14377#bib.bib188)\)Video\-LLaVA\([141](https://arxiv.org/html/2608.14377#bib.bib296)\), etc\.57\.00 \(Acc\)✗Diving\-QA\([31](https://arxiv.org/html/2608.14377#bib.bib188)\)DivingV, TSPORTU\([287](https://arxiv.org/html/2608.14377#bib.bib184)\)7 SportsV, T73\.20 \(Acc\)The Sports IndustryMANCustom Dataset\([167](https://arxiv.org/html/2608.14377#bib.bib82)\)\-T[167](https://arxiv.org/html/2608.14377#bib.bib82)Claude 3 Opus\([9](https://arxiv.org/html/2608.14377#bib.bib297)\)90\.28 \(Acc\)✗TSCCustom Dataset\([160](https://arxiv.org/html/2608.14377#bib.bib86)\)SoccerT[160](https://arxiv.org/html/2608.14377#bib.bib86)GPT\-4o\([102](https://arxiv.org/html/2608.14377#bib.bib238)\)3\.80 / 5\.00 \(Likert\)✗Custom Dataset\([210](https://arxiv.org/html/2608.14377#bib.bib87)\)SoccerI, TFootyintel\([210](https://arxiv.org/html/2608.14377#bib.bib87)\)\-\-✗Table 4:Summary of task\-specific sports datasets related to large models, including fans and social media, and the sports industry\. Task: CMT: sports commentary generation, HLG: sports highlight generation, NSG: sports news generation, NAR: sports narratives and storytelling, OPI: public opinion analysis in sports, MOD: sports models and systems, MAN: sports management, TSC: sports talent scouting\. Sports: TF: track and field, SC: soccer, BK: basketball, TN: tennis, TT: table tennis, GY: gymnastics, DV: diving\. Modal: V: video, I: image, A: audio, T: text\.
### C\.2Sports Understanding Datasets

This subsection echoes Section[3\.1](https://arxiv.org/html/2608.14377#S3.SS1)and provides a more detailed overview of datasets related to sports understanding tasks for large models\. We cover datasets specifically designed for sports understanding with large models \(§[C\.2\.1](https://arxiv.org/html/2608.14377#A3.SS2.SSS1)\), general video understanding datasets that include sports content \(§[C\.2\.2](https://arxiv.org/html/2608.14377#A3.SS2.SSS2)\), and other general\-purpose datasets containing sports\-related data \(§[C\.2\.3](https://arxiv.org/html/2608.14377#A3.SS2.SSS3)\)\. Table[5](https://arxiv.org/html/2608.14377#A3.T5)presents a comprehensive summary of the first two categories of datasets from multiple perspectives, including dataset names, covered sports types, data sources, annotation methods, benchmark availability, input modalities, the number and average duration of videos, the number of QA pairs, and open\-source links, which are directly accessible by clicking\.

#### C\.2\.1Specialized Sports Understanding Datasets

Recently, numerous datasets have been developed to evaluate and enhance the general sports understanding capabilities of large models\.

For LLMs, QASports\([107](https://arxiv.org/html/2608.14377#bib.bib186)\)introduced the first large\-scale sports question\-answering dataset with rich contextual information and diverse questions for model training and evaluation\. The sports understanding subtask in BIG\-bench\([241](https://arxiv.org/html/2608.14377#bib.bib190)\)includes 986 binary\-choice questions, primarily testing models’ general understanding of sports activities\. SportQA\([285](https://arxiv.org/html/2608.14377#bib.bib185)\)comprises over 70,000 multiple\-choice questions across three difficulty levels, enabling a comprehensive evaluation of LLMs’ performance in sports understanding\. SPORTU\-text\([287](https://arxiv.org/html/2608.14377#bib.bib184)\)and FSBench\-Text\([70](https://arxiv.org/html/2608.14377#bib.bib183)\)assess models’ understanding of rules, events, and scenarios in 5 major sports and figure skating, respectively\.

For MLLMs, Sports\-QA\([134](https://arxiv.org/html/2608.14377#bib.bib189)\)is the first dataset specifically designed for sports video question answering, advancing the evaluation of multimodal models in sports video understanding\. SPORTU\-video\([287](https://arxiv.org/html/2608.14377#bib.bib184)\)covers 7 sports and provides systematic video understanding tasks across three difficulty levels, while Sports\-3K\-QA\([34](https://arxiv.org/html/2608.14377#bib.bib100)\)includes a broader range of 49 different sports\. FSAnno\([70](https://arxiv.org/html/2608.14377#bib.bib183)\)constructs a large\-scale, multi\-task, multimodal figure skating dataset, while FSBench\-Motion\([70](https://arxiv.org/html/2608.14377#bib.bib183)\)extends it by adding motion data and QA pairs, supporting tasks ranging from single\-action analysis to full\-performance commentary\. FineBadminton\([88](https://arxiv.org/html/2608.14377#bib.bib187)\)is a large\-scale badminton video dataset with fine\-grained annotations, on which FBBench\([88](https://arxiv.org/html/2608.14377#bib.bib187)\)evaluates models’ fine\-grained sports video understanding\. Gym\-QA and Diving\-QA\([31](https://arxiv.org/html/2608.14377#bib.bib188)\), built upon FineGym\([232](https://arxiv.org/html/2608.14377#bib.bib191)\)and FineDiving\([292](https://arxiv.org/html/2608.14377#bib.bib192)\), respectively, offer new benchmarks for sports video question answering in gymnastics and diving\.

#### C\.2\.2General Video Understanding Datasets Featuring Sports

In addition to datasets specifically designed for sports understanding, many general video understanding datasets also include sports content, in which sports constitute an important component\.

In addition to large\-scale, multi\-task, and comprehensive video understanding datasets\([68](https://arxiv.org/html/2608.14377#bib.bib194);[87](https://arxiv.org/html/2608.14377#bib.bib215);[298](https://arxiv.org/html/2608.14377#bib.bib217)\), some focus on specific capabilities\. For example, InternVid\([272](https://arxiv.org/html/2608.14377#bib.bib195)\), FIOVA\([96](https://arxiv.org/html/2608.14377#bib.bib200)\), and VidText\([299](https://arxiv.org/html/2608.14377#bib.bib210)\)are primarily used for video description or subtitle generation, while Ego\-Exo4D\([79](https://arxiv.org/html/2608.14377#bib.bib197)\)and EgoExoBench\([89](https://arxiv.org/html/2608.14377#bib.bib211)\)focus on video understanding from different viewpoints\. LVBench\([270](https://arxiv.org/html/2608.14377#bib.bib193)\), MLVU\([322](https://arxiv.org/html/2608.14377#bib.bib198)\), Neptune\([177](https://arxiv.org/html/2608.14377#bib.bib201)\), LongVILA\_sft\([37](https://arxiv.org/html/2608.14377#bib.bib219)\), and VRBench\([305](https://arxiv.org/html/2608.14377#bib.bib220)\)are dedicated to long video understanding, while E\.T\. Bench\([148](https://arxiv.org/html/2608.14377#bib.bib196)\), MotionBench\([95](https://arxiv.org/html/2608.14377#bib.bib202)\), and ExAct\([303](https://arxiv.org/html/2608.14377#bib.bib212)\)are used for fine\-grained action, skill, or motion understanding\. TUNA\([122](https://arxiv.org/html/2608.14377#bib.bib213)\)and VideoA11y\-40K\([131](https://arxiv.org/html/2608.14377#bib.bib214)\)emphasize temporal information and dynamic video understanding, while VideoVista\([139](https://arxiv.org/html/2608.14377#bib.bib218)\), V\-STaR\([42](https://arxiv.org/html/2608.14377#bib.bib207)\), MINERVA\([176](https://arxiv.org/html/2608.14377#bib.bib208)\), VRBench\([305](https://arxiv.org/html/2608.14377#bib.bib220)\), and CausalStep\([137](https://arxiv.org/html/2608.14377#bib.bib222)\)target video reasoning tasks such as temporal\-spatial, multi\-step, and causal reasoning\.

Furthermore, video\-SALMONN\-2\([246](https://arxiv.org/html/2608.14377#bib.bib199);[245](https://arxiv.org/html/2608.14377#bib.bib223)\), WorldSense\([94](https://arxiv.org/html/2608.14377#bib.bib205)\), HarmonySet\([324](https://arxiv.org/html/2608.14377#bib.bib206)\), and MAVERIX\([289](https://arxiv.org/html/2608.14377#bib.bib216)\)focus on joint understanding of audio and video, demonstrating multimodal capabilities; while OVO\-Bench\([181](https://arxiv.org/html/2608.14377#bib.bib203)\)and RTV\-Bench\([297](https://arxiv.org/html/2608.14377#bib.bib209)\)examine the real\-time processing capabilities of models\. In terms of model capability evaluation, Trust\-videoLLMs\([273](https://arxiv.org/html/2608.14377#bib.bib221)\)is used to evaluate the credibility of video understanding, while SIV\-Bench\([123](https://arxiv.org/html/2608.14377#bib.bib224)\)studies the understanding of social interaction behaviors in videos\.

#### C\.2\.3Other General Datasets Featuring Sports

Moreover, some other types of general datasets also contain sports content\. To evaluate the image understanding capabilities of large models, MDI\-Benchmark\([317](https://arxiv.org/html/2608.14377#bib.bib227)\)collected 514 real images and 1,298 question\-answer pairs to test basic perception and complex reasoning, and designed sports\-related questions for different age groups\. MIP\-GAF\([159](https://arxiv.org/html/2608.14377#bib.bib228)\)constructed a large dataset to examine the understanding of key figures in images, which also includes sports scenes\. Furthermore, to assess the ability of large models as multimodal search engines, MMSearch\([111](https://arxiv.org/html/2608.14377#bib.bib225)\)collected 300 unimodal and multimodal samples, and MomentSeeker\([308](https://arxiv.org/html/2608.14377#bib.bib226)\)constructed a dataset consisting of 268 long videos with an average length of over 1,200 seconds, all of which focus on sports scenes\.

DatasetSportsSourceAnnotationBenchmarkModal\# VideoAvg\. Length\# QALinkSpecialized Sports Understanding DatasetsQASports\([107](https://arxiv.org/html/2608.14377#bib.bib186)\)SC, BK, AFFandomauto✗T\-\-∼\\sim1500K[✓](https://osf.io/n7r23/)BIG\-bench\-SU\([241](https://arxiv.org/html/2608.14377#bib.bib190)\)SC, BK, AF, BB, IHprogramcrowd✓T\-\-986[✓](https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/sports_understanding)SportQA\-Level\-1\([285](https://arxiv.org/html/2608.14377#bib.bib185)\)\-existing \(dataset\)manual✓T\-\-21385[✓](https://github.com/haotianxia/SportQA)SportQA\-Level\-2\([285](https://arxiv.org/html/2608.14377#bib.bib185)\)35 SportsWikipediaexpert✓T\-\-45685SportQA\-Level\-3\([285](https://arxiv.org/html/2608.14377#bib.bib185)\)SC, BK, TN, AF, TT, VBexpertiseexpert✓T\-\-3522SPORTU\-text\([287](https://arxiv.org/html/2608.14377#bib.bib184)\)SC, BK, TN, AF, VBexistingexpert✓T\-\-900[✓](https://github.com/chili-lab/SPORTU)SPORTU\-video\([287](https://arxiv.org/html/2608.14377#bib.bib184)\)SC,BK,BM,TN,BB,VB,IHcompetitionexpert✓V, T1701\-12048Sports\-3K\-QA\([34](https://arxiv.org/html/2608.14377#bib.bib100)\)49 SportsYouTubemanual✓V, T412\-1174[✓](https://showlab.github.io/livecc/)FSAnno\([70](https://arxiv.org/html/2608.14377#bib.bib183)\)FScompetitionexpert✗V, A, T783∼\\sim3\.5m\-✗FSBench\-Text\([70](https://arxiv.org/html/2608.14377#bib.bib183)\)✓T\-500✗FSBench\-Motion\([70](https://arxiv.org/html/2608.14377#bib.bib183)\)✓V, T7833500FineBadminton\([88](https://arxiv.org/html/2608.14377#bib.bib187)\)BMYouTubemanual✗V, T321512\.4s\-[✓](https://github.com/FineBadminton/FineBadminton)FBBench\([88](https://arxiv.org/html/2608.14377#bib.bib187)\)✓V, T2563\-2563Gym\-QA\([31](https://arxiv.org/html/2608.14377#bib.bib188)\)GYexistingmanual✓V, T6031\-27469✗Diving\-QA\([31](https://arxiv.org/html/2608.14377#bib.bib188)\)DVexistingmanual✓V, T∼\\sim100\-1055✗Sports\-QA\([134](https://arxiv.org/html/2608.14377#bib.bib189)\)SC, BK, GY, VBexistingmanual✓V, T596720\.9s94073[✓](https://github.com/HopLee6/Sports-QA)General Video Understanding DatasetsInternVid\([272](https://arxiv.org/html/2608.14377#bib.bib195)\)\-YouTubeauto✗V, A, T7\.1M6\.4mN/A[✓](https://github.com/OpenGVLab/InternVideo/tree/main/Data/InternVid)Ego\-Exo4D\([79](https://arxiv.org/html/2608.14377#bib.bib197)\)SC, BK, CLfieldexpert✗V, A50352\.6mN/A[✓](https://ego-exo4d-data.org/)E\.T\. Bench\([148](https://arxiv.org/html/2608.14377#bib.bib196)\)SC, BK, TN, CR, etc\.existingmanual✓V, T7002129s7289[✓](https://polyu-chenlab.github.io/etbench/)VideoVista\([139](https://arxiv.org/html/2608.14377#bib.bib218)\)SC, etc\.existingauto✓V, A, T894131s24906[✓](https://github.com/HITsz-TMG/VideoVista)FIOVA\([96](https://arxiv.org/html/2608.14377#bib.bib200)\)BB, etc\.\-manual✓V300233\.6sN/A[✓](https://huggingface.co/datasets/huuuuusy/FIOVA)Neptune\([177](https://arxiv.org/html/2608.14377#bib.bib201)\)BK, etc\.existingmanual✓V, A, T24052\.5m3268[✓](https://github.com/google-deepmind/neptune)video\-SALMONN2\([245](https://arxiv.org/html/2608.14377#bib.bib223)\)BB, etc\.\-manual✓V, A, T48351sN/A[✓](https://huggingface.co/datasets/videoSALMONN2/video-SALMONN_2_testset)LVBench\([270](https://arxiv.org/html/2608.14377#bib.bib193)\)BK, etc\.YouTubemanual✓V, T1034101s1549[✓](https://lvbench.github.io/)MMWorld\([87](https://arxiv.org/html/2608.14377#bib.bib215)\)SC, BK, GY, VBexistingmanual✓V, A, T1910∼\\sim105s6627[✓](https://huggingface.co/datasets/Xuehai/MMWorld)LongVILA\_sft\([37](https://arxiv.org/html/2608.14377#bib.bib219)\)\-existingmanual✗V, T15292\-15292[✓](https://huggingface.co/datasets/LongVILA/longvila_sft_dataset)MLVU\([322](https://arxiv.org/html/2608.14377#bib.bib198)\)SC, BK, BM, TT, VB\-manual✓V, T3102930s3102[✓](https://github.com/JUNJIE99/MLVU)Video\-MME\([68](https://arxiv.org/html/2608.14377#bib.bib194)\)SC, BK, etc\.YouTubemanual✓V, A, T9001017\.9s2700[✓](https://video-mme.github.io/)MotionBench\([95](https://arxiv.org/html/2608.14377#bib.bib202)\)BB, etc\.existing, syn\., webmanual✓V, T5385<10s8052[✓](https://huggingface.co/datasets/zai-org/MotionBench)OVO\-Bench\([181](https://arxiv.org/html/2608.14377#bib.bib203)\)\-existing, YouTubemanual✓V, T644428\.89s2814[✓](https://huggingface.co/datasets/JoeLeelyf/OVO-Bench)VISTA\-400K\([213](https://arxiv.org/html/2608.14377#bib.bib204)\)\-existingmanual✗V, T40399448\.6s∼\\sim381K[✓](https://huggingface.co/datasets/TIGER-Lab/VISTA-400K)HRVideoBench\([213](https://arxiv.org/html/2608.14377#bib.bib204)\)\-onlinemanual✓V, T2005\.4s200[✓](https://huggingface.co/datasets/TIGER-Lab/HRVideoBench)HarmonySet\-train\([324](https://arxiv.org/html/2608.14377#bib.bib206)\)\-YouTubemanual✗V, A, T4447031\.5s44470[✓](https://huggingface.co/datasets/Zzitang/HarmonySet/tree/main)HarmonySet\-MC\([324](https://arxiv.org/html/2608.14377#bib.bib206)\)✓38583858VideoA11y\-40K\([131](https://arxiv.org/html/2608.14377#bib.bib214)\)\-onlineauto✗V, A40000\-N/A[✓](https://huggingface.co/datasets/chaoyuli/VideoA11y-40K)TUNA\([122](https://arxiv.org/html/2608.14377#bib.bib213)\)SC, BK, etc\.existingmanual✓V, T100014\.5s2000[✓](https://huggingface.co/datasets/friedrichor/TUNA-Bench)V\-STaR\([42](https://arxiv.org/html/2608.14377#bib.bib207)\)\-existing, YouTubemanual✓V, T2094110\.23s\-[✓](https://huggingface.co/datasets/V-STaR-Bench/V-STaR)MINERVA\([176](https://arxiv.org/html/2608.14377#bib.bib208)\)BK, TN, etc\.YouTubemanual✓V, T22312m1515[✓](https://github.com/google-deepmind/neptune)MAVERIX\([289](https://arxiv.org/html/2608.14377#bib.bib216)\)SC, BK, etc\.existingmanual✓V, A, T7005\.7m2556✗RTV\-Bench\([297](https://arxiv.org/html/2608.14377#bib.bib209)\)SC, BK, etc\.existing, onlinemanual✓V, T55218\.2m4631[✓](https://github.com/LJungang/RTV-Bench)VidText\([299](https://arxiv.org/html/2608.14377#bib.bib210)\)SC, BK, BM, TT, SWexisting, YouTubemanual✓V, A, T939108\.2s2857[✓](https://huggingface.co/datasets/sy1998/VidText)SIV\-Bench\([123](https://arxiv.org/html/2608.14377#bib.bib224)\)SC, etc\.YouTube, TikTokmanual✓V, A, T279232\.49s8728[✓](https://huggingface.co/datasets/Fancylalala/SIV-Bench)ExAct\([303](https://arxiv.org/html/2608.14377#bib.bib212)\)SC, BK, CLexistingexpert✓V, T3521105s3521[✓](https://huggingface.co/datasets/Alexhimself/ExAct)VRBench\([305](https://arxiv.org/html/2608.14377#bib.bib220)\)SC, BK, VB, etc\.YouTubemanual✓V, A, T9601\.6h8243[✓](https://huggingface.co/datasets/OpenGVLab/VRBench)EgoExoBench\([89](https://arxiv.org/html/2608.14377#bib.bib211)\)BK, etc\.existingmanual✓V, T\-\-7350[✓](https://github.com/ayiyayi/EgoExoBench)WildVideo\([298](https://arxiv.org/html/2608.14377#bib.bib217)\)\-existingmanual✓V, T1318∼\\sim30s17625✗WorldSense\([94](https://arxiv.org/html/2608.14377#bib.bib205)\)\-existingmanual✓V, A, T1662141\.1s3172[✓](https://huggingface.co/datasets/honglyhly/WorldSense)Trust\-videoLLMs\([273](https://arxiv.org/html/2608.14377#bib.bib221)\)\-existing, syn\., YouTubemanual✓V, A, T6955\-\-[✓](https://github.com/wangyouze/Trust-videoLLMs)CausalStep\([137](https://arxiv.org/html/2608.14377#bib.bib222)\)\-existingmanual✓V, T100430\.5s1852✗Table 5:Summary of large\-model\-related datasets specifically for sports understanding and general video understanding with sports content\. Sports: SC: soccer, BK: basketball, BM: badminton, TN: tennis, GY: gymnastics, FS: figure skating, AF: American football, BB: baseball, CR: cricket, TT: table tennis, VB: volleyball, DV: diving, IH: ice hockey, CL: sports climbing, SW: swimming\. Source: syn\.: synthesis\. Modal: V: video, A: audio, T: text\.Abbr\.SportsAbbr\.TasksAFAmerican FootballACTAction Spotting and RecognitionBBBaseballAQASports Action Quality AssessmentBKBasketballCMTSports Commentary GenerationBMBadmintonEDUSports EducationBXBoxingHLGSports Highlight GenerationCLSports ClimbingINJSports Injury and RehabilitationCRCricketMANSports ManagementCYCyclingMODSports Models and SystemsDVDivingNARSports Narratives and StorytellingFSFigure SkatingNSGSports News GenerationGYGymnasticsOPIPublic Opinion Analysis in SportsHBHandballPLAExercise and Training PlansIHIce HockeyPRDGame and Player Performance PredictionRGRugbyPSYSports Psychology and BehaviorSCSoccerREFSports RefereeingTFTrack and FieldTACSports Tactics and StrategiesTNTennisTOUSports TourismTTTable TennisTSCSports Talent ScoutingVBVolleyballWRISports Academic WritingTable 6:Abbreviations for sports and tasks mentioned in this paper \(sorted alphabetically by abbreviation\)\.