本研究基于生成式人工智能技术，构建了一套大学物理实验课程答疑系统，并对试点班级学生的对话数据进行了统计分析．此系统结合了DeepSeek-V3大语言模型与动态课程知识库，优化了通用模型在教学场景下的适配性；通过双重角色设计和作答溯源机制，增强了学生的兴趣并提高了作答规范性；后台数据库完整记录下学生对话内容，可以非介入性观察的方式了解学情．该系统可助力个性化答疑和分层次教学，为破解面上实验教学的规模化与个性化之间的矛盾提供了一种解决方案，长期收集的对话语料也将为“智数驱动”教学评价和教学改革提供有力支持．

This study introduces a tutoring system for college physics experiment courses powered by generative artificial intelligence technology, with statistical analysis on the dialog collected from pilot classes. The system integrates the DeepSeek-V3 large language model with a dynamic knowledge base, enhancing adaptability of the general model in educational settings. Through dual-role interaction design and answer tracing mechanisms, it inspires students’ interest while improving the standardization of response content. The backend database archives users’ dialogues, enabling unobtrusive research of learning process of students. This platform facilitates personalized and differentiated instruction, offering a viable solution to reconcile the tension between sizable student population and personalization in general physics laboratory courses. The dialogue corpus collected through prolonged observation further provides data support for teaching evaluation and reform.