ConRAG: Lightweight inference of multi-hop relations

2026-09-28

AI Summary

This study addresses the challenges of recovering bridge entities and evidence chains, as well as the high indexing costs, in cross-document multi-hop relational reasoning. To this end, it proposes a lightweight, endpoint-constrained path retrieval mechanism. Instead of relying on expensive full-scale knowledge graph extraction, the method constructs an entity-document graph through entity co-occurrence analysis and large language model-assisted filtering. It then performs connectivity retrieval and semantic path ranking to infer relationships between endpoints. Experimental results demonstrate that the proposed approach significantly improves the recovery accuracy of bridge entities and reasoning chains on benchmarks such as MuSiQue. Furthermore, it reduces token consumption for index construction by approximately 1.5 orders of magnitude, achieving efficient and cost-effective cross-document relationship discovery.

Abstract

Understanding how two entities are connected often requires tracing multi-hop relations across documents to identify intermediate entities and supporting evidence that explain a connection. This is a task that appears frequently in scientific research and other knowledge-intensive analyses. We formalise this setting as multi-hop relation inference: given two known endpoint entities, we aim to recover the bridge entities and evidence-grounded reasoning chains that connect them across a document corpus, and to generate an explanation grounded in the retrieved evidence. Existing multi-hop RAG systems typically seek an unknown answer entity rather than explicitly recovering the connection between two known endpoints and graph-based approaches often rely on costly LLM-extracted knowledge graphs that limit scalability to large document collections. We introduce ConRAG, which builds a lightweight entity-document graph from entity co-occurrence and LLM-based entity filtering. Its connective retrieval infers and semantically ranks paths between two endpoints. On MuSiQue and 2WikiMultiHopQA, ConRAG consistently improves bridge entity and reasoning chain recovery over strong RAG baselines, while reducing graph-indexing token cost by up to roughly 1.5 orders of magnitude. Our results show that endpoint-constrained path retrieval provides an effective and index-efficient approach to evidence-grounded relation discovery.

Problem

• multi-hop relation inference
• bridge entity recovery
• reasoning chain
• retrieval-augmented generation
• knowledge graph scalability

Innovation

• Multi-hop relation inference
• Lightweight entity-document graph
• Connective retrieval
• Retrieval-Augmented Generation
• Reasoning chain recovery