NGM-RAG: Neural Graph Matching based Retrieval-Augmented Generation

2026-07-13

AI Summary

This work addresses the limitations of traditional retrieval-augmented generation (RAG) approaches, which rely solely on textual retrieval and struggle with multi-hop reasoning tasks. The authors propose a unified framework that integrates graph-structured knowledge into the RAG pipeline, jointly optimizing graph construction, graph-text alignment, and answer generation to enhance complex reasoning capabilities. Central to their approach is a neural graph matching mechanism that combines graph neural networks with semantic text matching, complemented by an adaptive weighting strategy to effectively fuse multi-source matching signals. Experimental results demonstrate that the proposed method significantly outperforms baseline models—including NaiveRAG, GraphRAG, and LightRAG—on both multi-hop question answering and long-context summarization benchmarks.

Abstract

Retrieval-Augmented Generation (RAG) significantly enhances the ability of Large Language Models (LLMs) to provide accurate and contextually relevant answers by dynamically integrating external databases. However, traditional RAG methods are primarily constrained by their reliance on text-based retrieval strategies, which often struggle with complex questions requiring multi-hop reasoning. To address this limitation, we introduce Neural Graph Matching based Retrieval-Augmented Generation (NGM-RAG), a novel framework that leverages graph structures to effectively capture and utilize relational knowledge for improved retrieval and answer generation. NGM-RAG explicitly incorporates graph construction, graph matching, and answer generation into a unified process. Within this framework, we propose a neural graph matching approach that combines text-based matching with Graph Neural Networks (GNNs). By employing an adaptive weighting strategy, NGM-RAG efficiently integrates multiple matching methods to select the most relevant contextual node information for answer generation. Experimental results on multi-hop question answering and long-context summarization tasks demonstrate that our NGM-RAG model achieves superior performance compared to both traditional NaiveRAG methods and state-of-the-art graph-enhanced approaches such as GraphRAG and LightRAG.

Problem

• Retrieval-Augmented Generation
• multi-hop reasoning
• graph-based retrieval
• relational knowledge
• Large Language Models

Innovation

• Neural Graph Matching
• Retrieval-Augmented Generation
• Graph Neural Networks
• Multi-hop Reasoning
• Adaptive Weighting