Knowledge Recommendation Approach for Emergency Response of Gas Tunnel Driven by Knowledge Graph-Large Language Model

The safety risks of gas tunnels are becoming increasingly prominent, but emergency response knowledge is often scattered in a large number of unstructured documents, resulting in low information acquisition efficiency and affecting the speed and quality of emergency response. A new emergency response knowledge recommendation approach that integrates knowledge graph and large language model is proposed. This method automatically extracts unstructured emergency related documents into structured knowledge graph through large language model prompt optimization, and uses semantic vector retrieval and reordering algorithm to obtain the most relevant subgraph as external enhanced knowledge of the Large Language Model, in order to solve the problem of inaccurate generated content due to the lack of professional knowledge in the large language model. The experimental results show that on a specific dataset, the cosine similarity between the results generated by this method and the semantic vectors of real decision reaches 0.83. The qualitative evaluation performs well in accuracy, completeness, and sentence coherence, significantly surpassing other baseline models and methods. The method proposed in the study not only improves the speed and quality of emergency response for gas tunnels, but also has broad promotion value, providing new ideas and methodological foundations for the design of emergency management systems in other high-risk environments.