To reduce development time, we are creating a chatbot system using Retrieval-Augmented Generation (RAG) to handle specialized inquiries. Challenges include the unfamiliarity of Large Language Model (LLM) with specific terms and the handling of ambiguous queries. We propose using a knowledge graph to supplement terms with explanations, synonyms, and hierarchical concepts. We developed a method using LLM to construct knowledge graphs for RAG more efficiently, maintaining accuracy while significantly reducing processing time compared to traditional methods. This paper details the methods for automating synonym and hierarchical relationship estimation using LLM.
