A Question Answering System for Aerospace Large Language Models Based on Knowledge Graph and RAG Collaboration

At present, the data in the aerospace science and technology is massive, heterogeneous, distributed and discrete, and lacks unified management, which seriously affects the quality and efficiency of retrieval. To solve this problem, a construction method of question answering system for aerospace large language models based on knowledge graph and retrieval enhanced generation (RAG) collaboration is proposed. Firstly, the aerospace science and technology information representation model is constructed, and then the aerospace science and technology information resource database is constructed based on the model. On this basis, the large language model in the aerospace field is trained by using the efficient fine-tuning technology; Create knowledge graph and vector database in the neo4j, and use entity attributes and vector indexes to integrate them; Finally, the langchain framework is used to link the knowledge graph, RAG and the large language model of the aerospace field to develop and construct the aerospace intelligent question answering system. The experimental results show that the system covers a wide range of knowledge in the aerospace science and technology field, and can effectively improve the accuracy of the answer and reduce the hallucination rate.