GRAG: Graph Retrieval-Augmented Generation
A method for graph retrieval-augmented generation that tackles networked documents.
GRAG addresses limitations of naive RAG by retrieving textual subgraphs and integrating joint textual and topological information into LLMs. It proposes a divide-and-conquer strategy for efficient retrieval and incorporates textual graphs into LLMs through two views. Experiments demonstrate GRAG's effectiveness in multi-hop reasoning on textual graphs.
Based on: GRAG: Graph Retrieval-Augmented Generation