IndexRAG: Bridging Facts for Cross-Document Reasoning at Index Time

2026-03-17

AI Summary

Existing retrieval-augmented generation (RAG) approaches for multi-hop question answering typically rely on online graph construction or multi-step iterative reasoning, resulting in high computational complexity and inefficiency. This work proposes shifting cross-document reasoning to the offline indexing phase by identifying bridging entities across documents and pre-generating standalone bridge facts that encapsulate multi-hop relationships. This enables multi-hop inference through a single retrieval operation and one call to a large language model, eliminating the need for additional training, fine-tuning, or online graph processing. The method significantly simplifies the pipeline while achieving an average F1 improvement of 4.6 points across three standard benchmarks—HotpotQA, 2WikiMultiHopQA, and MuSiQue—and, when combined with IRCoT, outperforms all existing graph-based baselines.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Natural Language Processing: Question Answering
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Multi-hop question answering (QA) requires reasoning across multiple documents, yet existing retrieval-augmented generation (RAG) approaches address this either through graph-based methods requiring additional online processing or iterative multi-step reasoning. We present IndexRAG, a novel approach that shifts cross-document reasoning from online inference to offline indexing. IndexRAG identifies bridge entities shared across documents and generates bridging facts as independently retrievable units, requiring no additional training or fine-tuning. Experiments on three widely-used multi-hop QA benchmarks (HotpotQA, 2WikiMultiHopQA, MuSiQue) show that IndexRAG improves F1 over Naive RAG by 4.6 points on average, while requiring only single-pass retrieval and a single LLM call at inference time. When combined with IRCoT, IndexRAG outperforms all graph-based baselines on average, including HippoRAG and FastGraphRAG, while relying solely on flat retrieval. Our code will be released upon acceptance.

Problem

• multi-hop question answering
• retrieval-augmented generation
• cross-document reasoning
• RAG
• bridge entities

Innovation

• IndexRAG
• cross-document reasoning
• bridge entities
• retrieval-augmented generation
• offline indexing