講演情報
[6D-03]A survey on GraphRAG: Methods, Challenges and Future Directions
*Li Guangcan1、Onizuka Makoto1 (1. Graduate School of Information Science and Technology, Osaka University)
発表者区分：学生
論文種別：ロングペーパー
インタラクティブ発表：あり
論文種別：ロングペーパー
インタラクティブ発表：あり
キーワード：
GraphRAG、RAG、LLM、Database Management、Knowledge Graph
Recently, Graph-based Retrieval-Augmented Generation (GraphRAG) has received a lot of attentions, which enhances RAG using graph-structured data in order to improve the ability of LLMs to handle the relationships among entities. Traditional RAG systems can optimize the output of LLMs by referencing an external knowledge base and perform remarkably well when answering retrieval-focused specific queries. When addressing query-focused summarization (QFS) tasks, such as global questions requiring the synthesis of relationships between numerous entities, traditional RAG models often focus only on local key information, making it difficult to perform comprehensive analysis and summarization of the entire text. These limitations hinder the performance of RAG in complex reasoning and large-scale text processing tasks. GraphRAG introduces a graph-based indexing mechanism that constructs an entity knowledge graph and generates community-based summaries to solve the QFS tasks. This survey provides a comprehensive overview of GraphRAG methodologies, including Graph-Based Indexing, Graph-Guided Retrieval, and Graph-Enhanced Generation. Additionally, we identify current challenges, including how to integrate LLMs with graph databases, and propose future research directions to enhance the scalability, accuracy, and versatility of GraphRAG. This survey aims to serve as a foundational reference for researchers and practitioners interested in integrating graph-based techniques into RAG systems.