Automatic Government Response Reference Generation System Based on Large Language Models and Multi-Agent

Abstract: In digital governance era, governmental platforms need to respond to citizens’ governmental inquires in a timely and effective manner. However, the existing government Q&A system mainly relies on manual responses, with limited assistance from automatic algorithms, which makes it difficult to efficiently handle the large volume of citizens’ government consultation needs in big data era. Therefore, in the era of digital governance, government platforms need to establish more effective and intelligent Q&A systems to respond to citizens’ government consultations. Nowadays, large language models (LLMs) are expected to help government platforms handle citizens’ government consultations in an automated and effective manner. LLMs can improve the efficiency of government platforms’ interactions with citizens and provide natural language responses to various types of citizen consultations. However, existing general LLMs have limited understanding of specific expressions in the government field and are temporarily unable to make effective responses like platform staff. This study builds a response reference system specifically for citizens’ governmental inquires based on LLMs and a historical vector database of government consultation questions and answers (GovLLM) by Muti-Agent systems. After inputting new citizens’ inquires, the system is able to generate practical and effective example answers for platform staff to refer to when handling citizens’ inquires. The system shows better text generation performance than the baseline model, which is conducive to improving the efficiency and effectiveness of government platforms in responding to citizens’ inquires.