Knowledge Generation from Large Language Models in the Automotive Field
Haomin Dong, Wenbin Wang, Yibo Hu, Ziyi Kang, Tao Yu, Xiaokang Liu · 2024
Knowledge construction based on text and symbols generally requires expensive human labor or complex text mining models. In recent years, large models in the text domain, such as ChatGPT, have demonstrated that they can implicitly encode large amounts of knowledge that can be constructed and queried through properly designed hints. However, compared to knowledge graphs, implicit knowledge in language models is often difficult to access or edit. This paper uses large language models to obtain symbolic knowledge, and proposes a new framework for automatic knowledge graph construction supported by the flexibility and scalability of large models. Compared with methods that usually rely on large amounts of human annotated data or existing large knowledge graphs as training data, our method only requires a small number of relationship definitions as input, and is therefore suitable for extracting and enriching knowledge systems containing new relationships. Our framework can automatically generate different hints and perform efficient knowledge search within a given large model to obtain consistent output.