Query-Centric Graph Retrieval Augmented Generation

2025-09-25

AI Summary

Existing graph-augmented RAG approaches face a granularity dilemma: entity-level graphs incur high token overhead and suffer from context loss, while document-level graphs lack the resolution to model fine-grained semantic relationships. To address this, we propose a query-centric graph indexing and multi-hop retrieval framework that pioneers a paradigm where the user query serves as the anchor for constructing controllable-granularity knowledge graphs. Our method integrates Doc2Query and an enhanced variant—Doc2Query--—to generate high-quality, interpretable graph structures. We further design a customized multi-hop retrieval mechanism for precise text chunk localization. Evaluated on LiHuaWorld and MultiHop-RAG, our approach significantly outperforms state-of-the-art chunk-level and graph-level RAG baselines, achieving substantial gains in question-answering accuracy. It establishes a new, high-fidelity, and interpretable baseline for multi-hop reasoning.

Technology Category

• Natural Language Processing: Question Answering
• Data Mining & Knowledge Management: Intelligent Query Processing
• Machine Learning: Graph-based Machine Learning

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

Graph-based retrieval-augmented generation (RAG) enriches large language models (LLMs) with external knowledge for long-context understanding and multi-hop reasoning, but existing methods face a granularity dilemma: fine-grained entity-level graphs incur high token costs and lose context, while coarse document-level graphs fail to capture nuanced relations. We introduce QCG-RAG, a query-centric graph RAG framework that enables query-granular indexing and multi-hop chunk retrieval. Our query-centric approach leverages Doc2Query and Doc2Query{-}{-} to construct query-centric graphs with controllable granularity, improving graph quality and interpretability. A tailored multi-hop retrieval mechanism then selects relevant chunks via the generated queries. Experiments on LiHuaWorld and MultiHop-RAG show that QCG-RAG consistently outperforms prior chunk-based and graph-based RAG methods in question answering accuracy, establishing a new paradigm for multi-hop reasoning.

Problem

• Resolving granularity dilemma in graph-based RAG methods
• Enabling query-granular indexing for multi-hop chunk retrieval
• Improving question answering accuracy for multi-hop reasoning

Innovation

• Query-centric graph RAG framework
• Query-granular indexing for controllable granularity
• Tailored multi-hop retrieval mechanism