Efficient Retrieval-Augmented Generation via Token Co-occurrence Graphs

2026-06-29

AI Summary

Standard retrieval-augmented generation (RAG) exhibits limited effectiveness on multi-hop reasoning tasks, while existing graph-based RAG approaches rely heavily on large language models (LLMs) to construct knowledge graphs, resulting in high computational overhead and error propagation. To address these limitations, this work proposes TIGRAG, a novel framework that eliminates the need for LLMs by constructing a token co-occurrence knowledge graph via sliding-window statistics. TIGRAG integrates graph-based semantic expansion, neural re-ranking, and a bridge-entity-driven iterative retrieval strategy to efficiently expand queries and retrieve multi-hop evidence during inference. Evaluated on three multi-hop question answering benchmarks, TIGRAG significantly outperforms dense retrieval and state-of-the-art graph RAG methods while substantially reducing indexing time, inference latency, and prompt length.

Abstract

Retrieval-Augmented Generation (RAG) mitigates hallucinations in Large Language Models (LLMs) by grounding the generation process on external knowledge. However, standard RAG approaches struggle with multi-hop reasoning. While recent graph-based RAG methods improve the retrieval of interconnected chunks, they often rely on computationally expensive and error-prone LLM-based extraction pipelines. To address these issues, we propose TIGRAG (Token-Induced GraphRAG), an efficient graph-augmented RAG framework based on a token co-occurrence Knowledge Graph. TIGRAG directly models topological relationships between tokens using sliding-window co-occurrence statistics, thus enabling scalable graph construction. During inference, it combines graph-based semantic expansion and neural reranking to retrieve interconnected evidence for multi-hop reasoning. Specifically, it introduces an iterative entity-driven retrieval strategy that progressively expands the query using bridging entities extracted from previously retrieved contexts. We evaluated TIGRAG on three widely adopted multi-hop Question Answering (QA) benchmarks. Experimental results demonstrated that our framework consistently outperforms dense retrieval and graph-based RAG methods in both retrieval and downstream QA tasks, while substantially reducing indexing time, inference latency, and prompt footprint.

Problem

• Retrieval-Augmented Generation
• multi-hop reasoning
• Knowledge Graph
• token co-occurrence
• hallucination mitigation

Innovation

• Token Co-occurrence Graph
• Retrieval-Augmented Generation
• Multi-hop Reasoning
• Graph-based RAG
• Entity-driven Retrieval