Knowledge Graph Construction for Intelligent Vehicle Cyber-Physical Systems Based on Large Language Models and LoRA

The Intelligent Vehicle Cyber-Physical System (IVCPS) involves multi-domain and multi-format data. By defining a unified semantic model, knowledge graphs can transform this heterogeneous data into structured knowledge, thereby supporting the design and development of IVCPS. In view of the high annotation costs and insufficient domain adaptability of traditional knowledge graph construction methods, this paper proposes an IVCPS knowledge graph construction method that integrates a large language model and Low-Rank Adaptation (LoRA) fine-tuning. By designing appropriate prompt templates and constructing an instruction dataset, the knowledge extraction effects of four open-source models from both domestic and international sources were compared. The Qwen3-8B model, which demonstrated the best overall performance, was selected for LoRA fine-tuning. After fine-tuning, the precision of knowledge extraction reached 89.8%. Ultimately, the fine-tuned model was utilized to construct the IVCPS knowledge graph, and its visualization and preliminary application validation were completed. This research provides an effective solution for efficiently constructing domain-specific knowledge graphs in low-resource scenarios.