用户响应式场景下大模型驱动的AI问答研究:以医疗分诊为例  

Research on AI Question Answering Driven by Large Language Models in User Responsive Scenarios:A Case Study of Medical Triage

作　　者：张君冬 刘江峰 王震宇 黄奇[1] 刘艳华 李娜 Zhang Jundong;Liu Jiangfeng;Wang Zhenyu;Huang Qi;Liu Yanhua;Li Na(School of Information Management,Nanjing University,Jiangsu Nanjing 210023;School of Health Economics and Management,Nanjing University of Chinese Medicine,Jiangsu Nanjing 210023)

机构地区：[1]南京大学信息管理学院,江苏南京210023 [2]南京中医药大学卫生经济管理学院,江苏南京210023

出　　处：《情报理论与实践》2025年第2期188-197,共10页Information Studies:Theory & Application

基　　金：江苏省研究生科研与实践创新计划项目“图模驱动的在线医疗健康智慧问答服务研究”(项目编号:KYCX24_0107);江苏高校哲学社会科学研究重大项目“中医古籍文献预训练模型构建及其应用研究”(项目编号:2023SJZD084)的成果。

摘　　要：[目的/意义]当前大语言模型驱动的AI问答仍以用户主动式提问场景为主,面对复杂或模糊场景时,模型难以有效引导对话。[方法/过程]文章提出了用户响应式场景下大语言模型驱动的AI问答框架,包含继续预训练、有监督微调、人类反馈强化学习、模型评估4个步骤,最后以医疗分诊为例进行实证研究,验证其可行性。[结果/结论]构建的模型具备主动引导信息收集、个性化互动、用户体验优化三大优势;创新了大语言模型的场景化应用的形式,可为后续AI问答服务的发展提供借鉴。[Purpose/significance]The current AI question answering driven by large language models still focuses on user initiated questioning scenarios,and when faced with complex or ambiguous scenes,the model is difficult to effectively guide the conversation.[Method/process]This article proposes an AI question answering framework driven by large language models in user responsive scenarios,which includes four steps:continuing pre-training,supervised fine-tuning,human feedback reinforcement learning,and model evaluation.Finally,an empirical study is conducted using medical triage as an example to verify its feasibility.[Result/conclusion]The model constructed in this article has three major advantages:actively guiding information collection,personalized interaction,and enhancing user experience.This study innovates the scenario based application of large language models,which can provide reference for the development of AI question answering services in the future.

关 键 词：用户响应式场景 模型主动性 AI问答 医疗分诊 大语言模型 

分 类 号：G63[文化科学—教育学]