Graph method improves multi-hop retrieval in question answering tasks
Corpus-Guided Dual-Path Propagation for Graph Retrieval-Augmented Generation
Computation and Language
Summary
Finding answers that require connecting several pieces of information is challenging. The authors designed NexusRAG, a method that organizes information like a map of related facts to better find useful clues even if they don’t seem directly related to the question. It uses two ways to spread importance across the graph so that it picks out both closely relevant facts and structurally connected ones. This approach outperforms previous methods on multiple question-answering tests, showing it finds better supporting evidence for complex questions.
What this means in practice
- •For natural language processing engineers: Improve multi-step question answering systems by integrating graph-based retrieval that better identifies connected evidence beyond surface text similarity.
- •For medical data teams: Enhance retrieval of linked biomedical facts in domain-specific knowledge graphs to support complex clinical question answering and research.
Authors
Baoxian Liu, Tong Wei
Abstract
Graph-based retrieval-augmented generation supports multi-hop retrieval by organizing corpus information into graphs. However, existing relation-free graph retrieval methods rely primarily on query-sentence similarity to search for evidence. This can exclude useful bridging evidence with low query similarity and activate incidental entities unrelated to the reasoning chain. In this paper, we propose a simple and effective approach called NexusRAG, which augments the relation-free Tri-Graph with a corpus-level entity neighborhood structure derived from joint entity co-occurrence and semantic similarity. NexusRAG employs this structure to guide two complementary propagation paths: neighborhood-constrained semantic propagation through sentences identifies the query-relevant entity frontier, while direct structural propagation between neighboring entities expands that frontier to structurally related entities. The propagated entity weights also inform neighborhood-aware passage initialization for Personalized PageRank. Experiments on three multi-hop QA benchmarks and a domain-specific subset of GraphRAG-Bench show that NexusRAG consistently outperforms existing approaches. On the GraphRAG-Bench subset, NexusRAG achieves the highest evidence recall in all question categories, exceeding baselines by 4.2-8.1 points. The implementation code is available at https://github.com/Jacob-biu/NexusRAG.