arXiv cs.CL· Jiaqi Wei, Wenxuan Huang, Chengdong Xu, Xiang Zhang, Wanli Ouyang·· 2 天前AI 评分36
超越检索失败：LLM 为何在多跳 RAG 中漏掉证据
Beyond Retrieval Failure: Understanding Why LLMs Miss Evidence in Multi-Hop RAG
AI 导读
研究提出无需训练的 ExecutionRAG 需求登记机制，记录未解决的子问题及其支撑证据，以区分检索器无法访问与智能体未发起查询两类失败原因。在 1,000 道 MuSiQue 问题上，all-MiniLM-L6-v2 的缺失证据错误答案从 382 降至 234，Qwen3-Embedding-0.6B 从 383 降至 274。
正文 · 原文
Abstract:Multi-hop retrieval-augmented generation (RAG) requires an agent to retrieve evidence for a sequence of dependent sub-questions. When a wrong answer lacks a required passage, the passage may be inaccessible to the retriever, or the agent may have failed to issue a query that would retrieve it. Standard retrieval evaluations conflate these two causes. We study how much apparent retrieval failure is instead caused by search drift: the search no longer tracks an unanswered sub-question. Holding the retriever, corpus, and generation model fixed, we introduce ExecutionRAG, a training-free requirement register that records open sub-questions and their supporting evidence. The register changes search guidance but not evidence access, so each gold passage it recovers certifies that the passage was reachable by the original retriever. On 1,000 MuSiQue questions, the number of wrong answers with unretrieved required evidence falls from 382 to 234 with all-MiniLM-L6-v2 and from 383 to 274 with Qwen3-Embedding-0.6B. These counts correspond to 69% to 44% and 66% to 48% of each system's wrong answers, respectively. Crucially, evidence recovery does not produce a proportional reduction in total error: premature resolutions and abstentions increase as missing-evidence failures decrease. The central finding is therefore a shift in the remaining bottleneck, from guiding search toward required evidence to verifying that the retrieved evidence is sufficient.
| Comments: | Accepted by DAI 2026 Oral |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.21184 [cs.CL] |
| (or arXiv:2610.21184v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21184 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jiaqi Wei [view email]
[v1]
Thu, 15 Oct 2026 03:39:25 UTC (780 KB)
来源：arXiv cs.CL · arxiv.org