@simplifyinAI: RAG is officially cooked. IBM researchers built a retriever that produced 65× fewer invalid section references than a f…
Summary
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.
View Cached Full Text
Cached at: 11/01/26, 01:55 PM
RAG is officially cooked.
IBM researchers built a retriever that produced 65× fewer invalid section references than a fine-tuned baseline.
RAG gives an AI relevant documents to read before answering. But finding the right information is still a problem.
Many systems split documents into chunks, losing the chapter headings and structure that help explain where information belongs.
IBM’s approach, STAIR, uses that structure.
Think of opening a book’s contents page and finding the right section before reading.
STAIR uses the table of contents to guide retrieval towards the section most likely to contain the answer.
On SearchTome, a benchmark covering 18 books across six domains, it retrieved the correct section first 82.6% of the time. The fine-tuned DSI baseline scored 76.9%.
Invalid section references fell from 3.25% to 0.05%. That’s where the 65× figure comes from. It measures retrieval references, not hallucinations in the final answer.
For businesses searching long manuals, policies and technical documents, this is a promising improvement.
It still needs testing on real company data, but preserving document structure could help agents find better evidence before they answer.
Similar Articles
@omarsar0: Great RAG paper from IBM. There are some really good ideas on how to solve common RAG issues. It's well known that retr…
IBM researchers introduce STAIR, a generative retriever that uses table of contents to preserve document structure, achieving 82.6% Recall@1 and reducing hallucination in RAG systems.
@oliviscusAI: The entire RAG industry is about to get cooked. Researchers developed a new RAG approach that bypasses almost everythin…
Researchers introduced PageIndex, a novel RAG approach that eliminates reliance on vector databases, embeddings, and chunking, achieving 98.7% accuracy on FinanceBench and outperforming existing methods while being free and open source.
@pritipatelor: The entire RAG industry is about to be turned upside down Researchers have come up with a brand-new RAG solution that n…
研究人员推出名为 PageIndex 的全新 RAG 方案，无需向量数据库、嵌入或分块，而是构建树状索引让大模型像人类读书一样检索，在 FinanceBench 上以 98.7% 的准确率超越所有向量 RAG 方法，且完全免费开源。
@vintcessun: Feeding too many documents into RAG causes retrieval quality to drop from 75% to 40%? Vector search is diluted by a large amount of irrelevant content, causing a sharp drop in hit rate in real deployment. Root cause: heterogeneous documents are retrieved together, noise drowns out signal. Multi-agent orchestration seems intelligent but actually introduces a precision-fidelity paradox—poor configuration leads to failure in both aspects. The paper proposes MA…
This paper identifies 'vector search dilution' in RAG systems when scaling to large heterogeneous document collections, where accuracy dropped from 75% to 40% in a real-world deployment. The proposed MASDR-RAG method uses domain scoping via organizational metadata before retrieval, improving P@10 from 0.77 to 0.86 with low cost and easy deployment.
@akshay_pachaar: RAG vs. retriever-free RAG, clearly explained! (powered by decision models like Jev) In a standard RAG setup, documents…
介绍康奈尔大学与 ASAPP 的 DecisionRAG 论文：用决策模型（如 Jev 1.13）在文档树上导航取代向量相似度检索，实现 retriever-free RAG，在 Natural Questions 上 top-1 命中率达 65.4%，优于 embedding 检索加 8B reranker 的 48.0% 和 GPT-5 reranking 的 60.8%。