Iterative LLM Prompting for Intelligent Cockpit Knowledge Graph Construction
Haomin Dong, Wenbin Wang, Zhenjiang Sun, Ziyi Kang, Xiaojun Ge · 2024
With the development of digital and intelligent cockpits, the data within these environments exhibit complexity and diversity, leading to issues such as inefficient data processing and difficulty in information extraction. Consequently, effectively capturing and representing the hidden associative knowledge in cockpits is crucial. Against this backdrop, knowledge graphs serve as an effective tool capable of retrieving and organizing a vast amount of information within a connected and interpretable structure. By systematically representing complex data relationships in the cockpit, they help enhance the prediction of precise interaction intentions and provide richer and more relevant knowledge support for personalized recommendations. However, rapidly and flexibly generating domain-specific knowledge graphs still poses certain challenges. This paper introduces an innovative method of knowledge graph construction using generative large language models, implementing a novel iterative zero-shot and domain-agnostic strategy. We propose an innovative strategy of iteratively prompting large language models to extract relevant triples for constructing knowledge graphs, effectively addressing key challenges such as entity recognition ambiguity and relationship extraction complexity in cockpit data. Experiments conducted in a domain-specific dataset demonstrate the feasibility of this method.