Tag
This paper shows that using incomplete relevance judgments as a shortcut for scoring abstention in conversational RAG benchmarks systematically distorts evaluation: stronger retrievers yield more 'false miss' contexts that still contain answers, badly deflating answerability AUROC. The authors release code, runs, and labels and recommend scoring refusals against content-based labels instead of gold-document presence.
IBM researchers introduced STAIR, a retriever that leverages document tables of contents for structured retrieval, achieving 82.6% top-1 section accuracy on the SearchTome benchmark and 65× fewer invalid section references than a fine-tuned DSI baseline.
Researchers from Los Alamos National Laboratory and NVIDIA present an expert-curated benchmark of 250 questions on inertial confinement fusion to evaluate LLMs' scientific fluency, physical reasoning, and quantitative analysis. They compare frontier open/closed models with Prospero, a domain-specialized model, finding that specialization improves scores but critical physics errors persist, and recommend assessing existing capabilities and retrieval benefits before investing in further specialization.
DeFT-RAG introduces a RAG framework that decomposes multi-hop questions into self-contained sub-questions with reusable cached answers, offloading reasoning to rule-based updates and a lightweight decision model. It achieves the best accuracy among thirteen baselines on MuSiQue, 2WikiMultiHopQA and HotpotQA while reducing LLM calls.
Glance integrates reranking and generation into a single 9B backbone using lightweight decision adapters and pointer heads that score passages from a shared prefill, reducing memory overhead while beating bge-reranker-v2-m3 and cross-encoders on MS MARCO benchmarks and HotpotQA answer F1.
A researcher-curated reading list of 3 essential papers for getting started with agentic RL: Search-R1 for fundamentals, Agent Lightning v1.0 for practical frontier, and a survey on the landscape of agentic RL for LLMs. The post includes the Search-R1 paper, which trains LLMs to autonomously generate search queries during step-by-step reasoning with real-time retrieval.
该论文提出知识冲突的解决不应依赖固定的启发式规则，而应引入对上下文可靠性的事先概率先验；在 GRACE 与 ClashEval 基准上，基于先验偏移的双声部选择器在各种上下文混合比例下显著优于现有基线。
The paper proposes TaLenT, an end-to-end LLM-based retriever-generator framework for textual graph question answering that uses gold answer signals as retriever supervision via a weighted pseudo-label loss to improve evidence coverage and compactness across three GraphQA datasets.
提出 MARCA 四智能体框架，利用 Gemini 2.5 Pro 和 DeepSeek-V3.2 结合检索增强生成（RAG）对 FAERS 个体案例安全报告进行基于 Naranjo 量表的因果关系评估，最终因果分类与人类专家的一致率达 79.4%（κ=0.587）。
该论文提出 Parametric Session State (PSS)，一种将长对话交互编译为固定形状的 LoRA 会话记忆层，使 RAG 系统在无需重放历史的情况下恢复未明确表述的查询中的检索意图，并在受控与公开评测中验证其跨模型与检索后端的有效性。
GEMS-RAG 是一个面向工程标准文档的图驱动多模态检索增强生成框架，通过层级解析、视觉实体抽取、多模态知识图谱构建与图驱动混合检索，在 MUTCD 数据集上相较文本、多模态和图基 RAG 基线提升了检索与问答效果。
提出 silver-evidence demonstrations 方法，用生成的、目标条件化的证据摘要替换 ICL 示例中的原始检索段落，以减少噪声检索带来的干扰。该方法在 EI 2.0、SciFact、合成策略检索和 HICRIC/IMR 数据集上优于 gold-only ICL 和 InstructRAG 式 ICL，并在所有标注数据集上提升了证据 F1。
This paper analyzes semantic synchronization of operator manuals over ISO 11783 CAN buses to enable offline, multilingual LLM assistance on agricultural machinery, benchmarking Markdown, JSON, and XML transfer and evaluating open-weight LLMs with retrieval-augmented generation across multiple languages.
论文提出 Compositional Concept Unlearning (C2U) 方法，在固定编码器与文档索引的 RAG 系统中通过查询侧表示编辑来抑制检索到不需要的概念，跨五个数据集和四个编码器验证了其在不损害控制查询效用的情况下降低目标概念的检索暴露。
This paper compares 11 retrieval configurations for long-document RAG—including lexical, dense, hybrid reranking, and full-document scanning—evaluating evidence coverage, answer quality, and cost on QASPER and textbook questions, finding that hybrid approaches often outperform full scans.
论文提出联合约束聚合（JCA）方法，通过将问题要求表示为共享约束并逐一核对候选答案，缓解检索偏向导致的限定词敏感问答失败，在三种评测设置和多个指令微调模型上提升准确率。
This paper introduces evidence-based generation (EBG) with a joint Ans+Evid metric and reformed benchmarks (MultiHop-RAG, QASPER-MP) to show that RAG's similarity-driven chunked retrieval—not reading—is the key bottleneck for long-context multi-hop question answering across frontier and open-weight LLMs.
AutoTerm-SST 是一个交互式流式语音翻译系统，通过自适应术语记忆和话题切片路由，实现零会话设置的术语感知翻译，无需用户上传术语表或选择领域即可匹配已知领域基线的术语准确率与 BLEU 表现。
提出 Self-Assessment 强化学习框架（SARL），让检索增强的 LLM 搜索智能体在训练中学会感知自身搜索状态、识别缺失信息，从而提升搜索效率而无需额外推理时评估，在多跳推理与深度研究基准上取得显著相对提升。
论文提出开源 RAG 受控评测框架 LASAGNA，在八个基准和 5K-200K 文档语料上测试 450 种配置，发现单独有效的 RAG 技术组合后收益大幅缩水（预期 +47% 仅得 +17%），检索质量与答案忠实度的设计取舍互相冲突，且最优组合随语料规模变化。