关键词: 大语言模型, 检索增强生成, 图检索增强生成, 知识图谱

Abstract: Aiming at the problem of  the information overload caused by the explosion of big data teaching resources and the insufficient accuracy of traditional retrieval-augmented generation (RAG) in multi-source information fusion, we proposed a big data knowledge learning method based on GraphRAG. Firstly, we designed a Chinese prompt template to drive GraphRAG to automatically extract course entities and relationships, constructed an initial knowledge graph, and persisted it to Neo4j graph database. Secondly, through entity alignment and relationship completion, manually organized knowledge points were integrated with the automatically constructed graph to form a unified and evolving knowledge graph database. Finally, the community summaries pre generated by GraphRAG were utilized to achieve global semantic search, while relying on the Neo4j graph database to achieve precise local retrieval of knowledge points. The experimental results show that the proposed method is significantly better than traditional RAG in terms of question answering accuracy, response correlation, and smoothness of multi\|source information integration.

Key words: large language model, retrieval-augmented generation, graph retrieval-augmented generation, knowledge graph