关键词: 政务问答系统, 检索增强生成, 大型语言模型, 语义检索, 知识增强

Abstract: A government affairs hotline Question Answering(QA) system based on knowledge-enhanced large language model architecture(ChatGovt) was proposed to address the issues in current systems, such as low manual response efficiency, inaccurate query screening mechanisms, and insufficient intention difference recognition in traditional Retrieval-Augmented Generation (RAG) systems. First, to improve response efficiency, a system architecture integrating intelligent question triage and structured feedback was designed, enabling classified processing of consultation, complaint-suggestion, and other types of questions through intention recognition. Then, to improve knowledge retrieval quality, a multi-stage semantic-enhanced retrieval method was proposed, which includes three stages: historical dialogue summary retrieval, semantic re-ranking, and self-reflective decision-making. Finally, cross-domain knowledge was supplemented through online queries to form a service closed-loop for government consultation. Experimental results show that in terms of retrieval quality, compared with the traditional RAG system, query-knowledge relevance, real answer-knowledge relevance, and knowledge support are improved by 15%, 7.4%, and 24.6% respectively. For overall system performance, the answer recall is increased by 55.4% compared with the fine-tuned GLM(General Language Model)4-9b-chat model, and manual evaluation shows a 27.3% improvement over Doubao. This system provides a referenceable architecture and methodology for the technical optimization of government question answering systems, can effectively improve the response efficiency and service accuracy of government hotlines, and promotes the intelligent transformation of government services.

Key words: government Question Answering(QA), system, Retrieval-Augmented Generation (RAG), Large Language Model(LLM), semantic retrieval, knowledge enhancement