arXiv cs.AI· Vineet Kumar, Meghanadh Pulivarthi, vishwajeet kumar, Jaydeep Sen, Riyaz Ahmad Bhat, Sachindra Joshi·· 3 天前AI 评分31
STAIR：利用文档目录结构增强检索的 LLM 检索器与 SearchTome 基准
STAIR (STructure Aware Information Retriever): A novel dataset and LLM based retriever for document structure augmentation
AI 导读
研究者提出基于 LLM 的检索系统 STAIR，利用语料库的全局结构（如目录 ToC）来存储和检索信息，在自建基准 SearchTome 上取得 82.6% 的 Recall@1，高于 DSI 的 76.9%、BM25 的 59.5%、DPR 的 68.7% 和开箱即用的 Mistral 的 13.8%。
正文 · 原文
Abstract:Retrieval Augmented Generation (RAG) is a key component for generating accurate and hallucination free answers using Large Language Models (LLMs). LLMs are improving at handling long context, but still suffer from "lost in the middle" problem. Thus, precise and accurate retrieval is important. Current retrievers chunk long context into length-based manageable chunks - in the process throwing away rich and informative semantic global structure in the corpus. We introduce a novel retrieval system STAIR that empowers an LLM to exploit global structure in a corpus such as a Table of Contents (ToC) to efficiently store and retrieve information from its model parameters. Our thorough and careful ablation studies with a finetuned Differentiable Search Index (DSI) system show that ToC helps build a low hallucination (less than 0.05%) generative Information Retrieval (IR) system and can generalize to examples where very few training samples are available. To further research in this novel direction of ToC based retrieval we release SearchTome - a diverse benchmark created from 18 books across 6 diverse domains to further research in this novel direction. STAIR achieves a high Recall@1 score of 82.6% on SearchTome as compared to DSI (76.9%), where the difference is found to be statistically significant. STAIR easily beats other strong baselines such as BM25 (59.5%), DPR (68.7%) and out-of-the-box Mistral (13.8%).
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.03874 [cs.AI] |
| (or arXiv:2609.03874v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.03874 arXiv-issued DOI via DataCite |
Submission history
From: Meghanadh Pulivarthi [view email]
[v1]
Thu, 3 Sep 2026 14:02:53 UTC (3,408 KB)
[v2]
Tue, 13 Oct 2026 05:54:31 UTC (3,408 KB)
来源：arXiv cs.AI · arxiv.org