question-answering

This paper compares chunking and embedding strategies for RAG systems in Turkish, evaluating their impact on document question answering through a systematic study with multiple configurations.

The paper describes a submission to the WMT 2026 MIST shared task, using the Tiny Aya Global model with task-specialized QLoRA adapters for multilingual summarization and question answering.

This paper introduces Conformalized Agentic Search (CAS), a framework that uses Conformal Prediction to enhance the reliability of search agents by adaptively retrieving documents and weighting policies during reinforcement learning, thereby improving accuracy and reducing redundant tool calls.

SABET-QA introduces an iterative framework for temporal knowledge graph question answering that enhances multi-hop reasoning through bidirectional entity-temporal scoring and contextualization, showing consistent improvements over baselines on benchmarks like CronQuestions and TimeQuestions.

IAR is a three-stage post-training framework that internalizes structured document knowledge into language models for retrieval-free question answering, enhancing both domain-specific accuracy and general model performance.

This paper proposes a Multi-Modal Generative Fuzzy System (MMGFS) to enhance multimodal question answering by addressing modality bias and uncertainty through fuzzy inference and multi-hop reasoning, demonstrating improved performance on multiple benchmarks.

The paper proposes A-CRC-QA, a post-hoc calibration framework for selective question answering that controls error rates among accepted answers via asymptotic risk calibration, demonstrating improved reliability-retention trade-offs on CoQA and MedMCQA.

This paper introduces Lodestar, a method that uses reinforcement learning to train a short polarizer prompt string that helps a frozen LLM avoid being misled by misleading retrieved passages in RAG question answering. It improves F1 and exact match scores across five QA benchmarks compared to existing entropy-based selection rules.

This paper introduces BEST-KAG, a multimodal knowledge-driven framework for question answering on building engineering standards, leveraging a multimodal knowledge graph and graph-retrieval-based generation to improve clause-grounded traceability and outperform multiple LLMs.

This paper studies how instruction tuning affects model confidence and lexical diversity in question answering, finding that it alters confidence and reduces rationale diversity without improving calibration.

This paper evaluates the reliability of BERT-based QA models (RoBERTa, ALBERT, DistilBERT) under Monte Carlo Dropout and input paraphrasing, finding RoBERTa more consistent and validating MCD as a reliability metric.

This paper presents a self-improving 'researcher agent' for Text-to-SPARQL question answering over knowledge graphs, which iteratively refines its own prompts and tools. Evaluated on DBpedia, it achieves 0.22 accuracy and identifies predicate selection as the main bottleneck.

ENTLORE is a graph-grounded benchmark framework for enterprise question answering that evaluates latent organizational reasoning, showing that even with gold document sources, many implicit relation questions remain unanswered.

This paper studies false-presupposition QA (FPQA) methods and shows that methods performing well on false-presupposition questions often degrade accuracy on normal true-presupposition questions, due to weak fact-checking modules that reject valid presuppositions. The authors argue that FPQA benchmarks over-represent false presuppositions and that evaluation should consider real-world QA performance.

This paper proposes Archive, a framework for ambiguity detection in open-domain QA that distinguishes ambiguity from answer diversity using logical conflict, and introduces QuireQA, a 4,703-query benchmark. Experiments show Archive improves F1 by up to 21.6% while being 16x faster than competitors.

Scrimba Explain lets users ask any question and receive an instant video answer, making learning and explanations more accessible.

RAGOCR is a novel framework that compresses retrieved documents into compact visual representations conditioned on the input query, using query-aware dynamic resolution to balance compression rate and information fidelity. Experiments show it surpasses naive RAG by over 15% accuracy while using only one-eighth the input tokens.

The paper proposes QQ, a framework that leverages the intrinsic duality between multi-hop question generation and question answering via bidirectional alignment constraints and contrastive learning, improving question quality on HotpotQA and MuSiQue.

Zero-Mem introduces a zero-token memory operation system for LLM agents, eliminating LLM calls and token consumption during memory retrieval by preserving original interaction traces and using deterministic entity-context and temporal structures. It achieves competitive performance on long-memory and long-context QA benchmarks while reducing memory-operation time cost by 57.6%.

This paper introduces SESA, a self-evolving skill-augmented search agent that co-evolves task generation and skill memory via tool-augmented search self-play. It improves accuracy across seven QA benchmarks over baselines while supporting memory-free deployment.