arXiv：cs.CL· Nabaraj Subedi, Ahmed Abdelaty, Shivanand Venkanna Sheshappanavar·· 16 小时前
更多文档为何损害 RAG：用领域限定检索衡量并缓解向量搜索稀释
When More Documents Hurt RAG: Measuring and Reducing Vector Search Dilution with Domain-Scoped Retrieval
AI 导读
RAG 在大规模异构文档集上会出现"向量搜索稀释"，稠密相似度判别力下降，使 top-k 检索返回语义相似但上下文错误的片段。在 WYDOT 部署系统中，语料从 54 篇增至 1,128 篇使 precision@10 下降 63%、答案文档召回率下降 23%。
正文
Abstract:Retrieval-augmented generation (RAG) degrades when applied to large, heterogeneous document collections. In RAG systems, dense similarity becomes less discriminative, causing top-$k$ retrieval to return semantically similar but contextually incorrect chunks. We refer to this failure mode as vector search dilution. Even under hybrid dense+sparse retrieval, we first observed it in a deployed Wyoming Department of Transportation (WYDOT) system; in a controlled experiment, growing the corpus from 54 to 1,128 documents cut precision@10 by 63% and recall of the answer-bearing document by 23%. Similar patterns were observed in six other DOTs. To diagnose this issue, we introduce MASDR-RAG (Metadata-Scoped Domain Retrieval for RAG) as a controlled testbed over 200 human-validated queries, five LLM backbones, six public benchmark corpora, and two index stacks. In MASDR-RAG, domain scoping is the load-bearing component. We found hard partitioning to be unsafe, whereas training-free soft scoping is safe when the scope structure is real. Furthermore, we can discover these scopes without hand-curated metadata. However, multi-agent orchestration triggers a precision-faithfulness paradox that survives format controls, prompt controls, and blind human evaluation. Based on these varied outcomes, our practical recommendation is simple: soft-scope first, then perform a single synthesis call.
| Comments: | 19 pages, 5 figures, 20 tables. Preprint under review at ACL ARR October |
| Subjects: | Computation and Language (cs.CL); Information Retrieval (cs.IR) |
| Cite as: | arXiv:2606.11350 [cs.CL] |
| (or arXiv:2606.11350v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.11350 arXiv-issued DOI via DataCite |
Submission history
From: Nabaraj Subedi [view email]
[v1]
Tue, 9 Jun 2026 18:26:24 UTC (844 KB)
[v2]
Wed, 14 Oct 2026 00:58:36 UTC (787 KB)
来源：arXiv：cs.CL · arxiv.org