Qinggang Zhang, Shengyuan Chen, Yuan-Qi Bei, Zheng Yuan, Huachi Zhou, Zijin Hong, Junnan Dong, Hao Chen et al.
Graph-based Retrieval-Augmented Generation (GraphRAG) is a new paradigm that utilizes knowledge graphs to overcome the limitations of traditional flat-text retrieval-based RAG.
When applying large language models (LLMs) to specialized domains, existing RAG systems face limitations in complex query understanding, integration of distributed knowledge, and system efficiency at scale.
GraphRAG addresses these issues through graph-structured knowledge representation, efficient graph-based retrieval (including multi-hop reasoning), and structure-aware knowledge integration algorithms.
This survey systematically analyzes the technical foundations of GraphRAG, reviews implementation cases across various professional domains, and identifies key technical challenges and promising research directions. It also provides a GitHub repository collecting related resources (papers, open-source data, projects).