Tag
ReCoPII is a relation-grounded framework for building contextual PII masking datasets without relying on author-based assumptions, producing the ReCoPII-Ko Korean corpus with MASK/KEEP annotations that significantly outperform SPY-style augmentation in human agreement (88.42% vs 26.27%).
SLICE-Mem 提出一种轻量级 LLM Agent 长期记忆框架，通过显著性引导的记忆准入与基于谓词的语义抽取实现免 LLM 调用的在线记忆构建，在保持接近 MemOS 基线精度的同时将 token 消耗降低 92.4%–99.5%。
This paper proposes a multi-granularity interaction network for relational triple extraction that jointly leverages word-word, entity-relation, and head-tail interactions to better handle unseen entities, achieving F1 improvements of 3.1%/20.5% on the NYT-R and WebNLG-R datasets in the Unseen metric.
中国电信发布 Xing4.0-29B Intent，一个基于 MoE 架构的 29B/4B 专业化模型，面向意图理解与结构化信息抽取，支持意图分类、命名实体识别、关系抽取与事件抽取四项任务，采用 Apache 2.0 开源许可。
论文提出了一种混合生成-标注解码架构，将源文本跨度标注与开放生成相结合，用于电商产品的属性-值抽取。该方法在 vLLM 推理中将属性覆盖率提升 10%、正确属性-值对提取量增加 63%，同时延迟降低 44% 且不损失精确率。
ReviewPilot is a conversational multi-agent system that automates the PRISMA systematic review workflow from paper screening to information extraction, using project memory to retain configurations and providing source-backed decisions. It outperforms single-prompt LLM and AgentSLR baselines across three LLMs, achieving screening macro F1 up to 0.776 and extraction F1 up to 0.788, with case studies showing transferability to biomedicine and HCI.
CLUE提出了一种对比学习框架，用于无监督的事件类型归纳、命名与标注，通过联合嵌入表示捕捉动词语义与多义性，并用对比损失引导的聚类替代人工事件本体构建。
发布 CalOn，一个 0.6B 参数的校准类型化决策模型，用于实体、关系与事件抽取，速度达每秒 120 句，是 GLiNER2 的 5.5 倍，并在 40 个基准上超越基线。权重、推理与训练代码均以 Apache 2.0 开源。
This paper systematically evaluates how LLMs generalize across four dimensions—domain shift, adversarial perturbations, compositionality, and length increase—in time and event expression extraction, finding that inductive prompting generalizes most consistently while gains from scale, architecture, and other prompting strategies are uneven and dimension-specific.
A self-run experiment compares a six-model 1-2B local swarm (modeled on a 1930s Soviet design institute's discipline-based org chart) against a single model on a document extraction task, finding the real win came from splitting inputs into blocks rather than agent collaboration (+0.05 from the swarm, with strict emergence at zero).
A practitioner details a pipeline for distilling an LLM's document-extraction labels (entities, actions, values) into two 287M-parameter encoders — a GLiNER-style span tagger plus a multiple-choice model for entity resolution — to process ~5M court decisions, and asks for feedback on why it lags the teacher model.
研究人员提出了一种结合深度学习与图像预处理技术的OCR模型，用于自动提取韩文发票中的关键信息，在自建数据集上达到了87%的F1分数，且处理耗时极低。
This paper evaluates the use of large language models as AI respondents to generate structured survey responses from policy documents, demonstrating high agreement in structured indicators and potential for hybrid human-AI workflows in policy monitoring.
EAGER is a reinforcement learning framework that improves generative event extraction through fine-grained verifiable rewards and schema-contrastive advantage estimation, outperforming prompting, fine-tuning, and prior RL methods on seven benchmark datasets.
This paper presents WaterBERT, a domain-adapted BERT model for water treatment literature mining, enhancing semantic representation and enabling large-scale structured information extraction and knowledge graph construction.
Schematize is an open-source multi-agent system that interactively generates and refines information-extraction schemas for legal research, achieving top performance in human evaluations.
GLiNER2.5 is a multilingual boundary checkpoint for unified schema-based information extraction, supporting entities, classification, structured records, relations, and span attributes in one model.
This paper introduces Self-Meta-Evolve, a hierarchical framework that personalizes prompts for each user in enterprise information extraction tasks, improving performance through continuous refinement based on interaction feedback.
The paper introduces a weakly supervised framework using large language models to extract dataset mentions in forced displacement and FCV documents, achieving high accuracy with limited labeled data.
CMNIE is a new information extraction benchmark for Chinese military news that jointly annotates events, entities, and relations to evaluate models on schema adherence and exact span matching.