STAIR (STructure Aware Information Retriever): A novel dataset and LLM based retriever for document structure augmentation
Vineet Kumar and colleagues at IBM introduce STAIR, a generative retriever that stores and addresses a corpus through its table of contents rather than through length-based chunks.
Ask this paper
Chunking discards global structure. Standard retrievers split long documents by length, which throws away the hierarchy a table of contents already encodes. STAIR keeps that hierarchy as the addressing scheme.
Recall@1 of 82.6% on SearchTome against 76.9% for a fine-tuned Differentiable Search Index, a statistically significant difference, and well above BM25 at 59.5% and DPR at 68.7%.
Hallucination below 0.05% in the generative IR setting, which is the main risk of generative retrieval and the reason ToC grounding is worth the added structure.
Generalizes with very few training samples, according to the ablations, which matters because generative retrievers usually need dense per-document supervision.
SearchTome is released: 18 books across 6 domains, built to evaluate structure-aware retrieval specifically.
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%).