关键词: 大语言模型, 知识图谱构建, 提示学习

Abstract: The construction of traditional knowledge graphs mainly relies on technical means such as data preprocessing, entity recognition, relation extraction, and entity alignment. Such methods are usually accompanied by high computational and time costs. To address this issue, a domain knowledge graph construction meth-od ERC-KG (Extraction Retrieval and Error Correction Knowledge Graph) is proposed, which combines large language models to optimize the efficiency and accuracy of knowledge graph generation. By combining feature word extraction with domain expert knowledge, the entity set of the knowledge graph is precisely determined. An entity corpus retriever is con-structed to select the context sentences most relevant to the target entity as the input of the large language model. A prompt template and verification feedback mechanism are designed to achieve high-quality triple extraction and complete the construction of the knowledge graph in the field of national defense science and technology. Experimental results show that the accuracy of the graph construction reaches 94.32%, verifying the accuracy and rationality of the method and also contributing new research ideas for the rapid construction of domain knowledge graphs.

Key words: large language model, knowledge graph construction, prompt learning