Extraction Method of Disturbance Elements of Urban Emergencies Based on Large Language Model

The general knowledge extraction model lacks the knowledge learning of such as urban emergencies, and the precision of knowledge extraction is insufficient. In this paper, a large language model (LLM) based method for urban disturbance elements extraction (UDEE-LLM) is proposed. The method deploys the local large language model through LangChain and integrates the domain knowledge of urban disturbance, fine-tuning the large language model based on the prompting learning, designing the heterogeneous entity alignment strategy, and optimizing the entity extraction results. The results show that UDEE-LLM can extract and classify the information of urban disturbance elements. The coupling relationship of urban disturbance elements is analyzed, and the potential relationship between mode, object and influence of urban disturbance is dug out, which has important reference for improving urban resilience.