Research on Intelligent Monitoring of E-Government Using LLM, RAG, and Hybrid Retrieval Technologies

Large language models are one of the core research contents of natural language processing and have been widely applied in many fields including government affairs. The problems that need to be solved in the research of large language models for government affairs are proposed, namely, data multimodality, correctly facing the trend of "model as a service", emphasizing high data security, and clarifying the boundaries of responsibility. In addition, the technical path for the research of large language models for government affairs has also been proposed. Then, the feasibility of the large language model in the process of industrial collaborative emergency preparedness is analyzed. The P-Tuning method is adopted for model fine-tuning. A local knowledge base is constructed based on the question-answering data related to the intelligent monitoring events of e-government. The content generated by the model is standardized, and the LangChain architecture is utilized to build the industrial collaborative emergency question-answering system. Based on the large model knowledge base, a hybrid retrieval dual-tower model was proposed. The model integrates the multi-path recall strategy to ensure the comprehensive retrieval results. The multi-level ranking of the retrieval results was achieved through the adoption of hybrid retrieval. The correlation of the retrieval results has been significantly improved. Further, the optimized information and the original query are input into the large language model to generate accurate answers. Experiments show that this method significantly improves the accuracy of retrieval.