摘要
集成检索增强生成技术的大语言模型能较好地解决大语言模型的知识更新与“幻觉”问题,适用于教务查询领域。考虑到教务查询领域的通用问题、特定领域问题,根据大语言模型的不同需求和用户反馈持续优化系统。首先,利用基于LLM的Agent将文本分类模型与RAG组成处理流程;其次,根据问题分类结果融合Agent的记忆内容;再次,调用大语言模型生成系统应答;最后,在框架中引入基于用户反馈的强化学习机制,以优化Agent的记忆内容、查询结果及结果融合。实验表明,在Agent框架中引入文本分类模型可提升用户交互效率,降低大模型的调用开销;强化学习机制虽增加了系统复杂度,但用户满意率会随着用户交互不断提升。
The integrated retrieval enhanced generation technology for large language models can effectively solve the problems of knowledge updating and"illusion"in large language models,and is suitable for the field of academic inquiry.Considering the general and specific issues in the field of academic inquiry,the system is continuously optimized based on the different needs of the big language model and user feedback.Firstly,use LLM based agents to combine the text classification model with RAG to form a processing flow;Secondly,based on the classification results of the problem,integrate the memory content of the agent;Again,call the big language model to generate system responses;Finally,a reinforcement learning mechanism based on user feedback is introduced into the framework to optimize the memory content,query results,and result fusion of the agent.Experiments have shown that introducing text classification models into the Agent framework can improve user interaction efficiency and reduce the overhead of calling large models;Although reinforcement learning mechanisms increase system complexity,user satisfaction rates will continue to improve with user interaction.
作者
刘志忠
LIU Zhizhong(School of Communication Technology,Communication University of Nanjing,Nanjing 210007,China)
出处
《软件导刊》
2025年第10期149-154,共6页
Software Guide
关键词
AGENT
文本分类
强化学习
大语言模型
检索增强生成
教务查询系统
Agent
text classification
reinforcement learning
large language model
retrieval augment generation
academic affairs inquiry system
