从“负能”到“赋能”:基于LLMs的思维链提示设计与教研AI智能体构建——以课堂教学智能分析为例  

From“Negative Energy”to“Empowerment”:Designing Chain of Thought Prompts Based on LLMs and Building AI Agents for Teaching and Research——A Case Study in Intelligent Classroom Teaching Analysis

作　　者：王冬青[1] 陈自力 邵文豪 张粤芳 李赞坚[2] 任光杰[1] Wang Dongqing;Chen Zili;Shao Wenhao;Zhang Yuefang;Li Zanjian;Ren Guangjie(School of Information Technology in Education,South China Normal University,Guangzhou 510631,Guangdong;Guangzhou Educational Technology Center,Guangzhou 510091,Guangdong)

机构地区：[1]华南师范大学教育信息技术学院,广东广州510631 [2]广州市电化教育馆,广东广州510631

出　　处：《中国电化教育》2025年第3期111-117,125,共8页China Educational Technology

基　　金：广东省哲学社会科学规划项目“教育数字化转型下促进知识建构的人机智能协同方法与适应性干预研究”(项目编号:GD24XJY35);广东省哲学社会科学创新工程特别委托项目“广东省高校人工智能学科发展现状与对策研究”(项目编号:GD24WTCXGC10)研究成果。

摘　　要：课堂教学智能分析是人工智能技术赋能循证教研的新趋势,通常以报告的形式呈现给一线教师,但其往往包含巨大认知负荷且数据呈现复杂,使得一线教师难以把握问题关键点并用于教学改进,以数据“赋能”为出发点,却反而给教师带去了数据“负能”。该文基于思维链提示逻辑,提出了教研AI智能体赋能课堂教学分析报告解读的构建框架,实现从数据解析到反馈生成的循环,并以此为导向模块化构建了基于开源大语言模型(LLMs)的智能体框架,个性化开发教研AI智能体。通过63份真实报告数据,验证了“基于思维链提示的回复”相较于“基于LLM的普通回复”的有效性,结果表明前者在多项评价维度上均表现出更高的评分,尤其是在准确性、逻辑性和专业性方面具有显著提升。该文通过聚焦智能体在教研中的垂直应用,探索从数据负能到赋能转变的新路径。Intelligent analysis of classroom teaching represents a new trend in leveraging artificial intelligence(AI)to empower evidencebased educational research.Typically presented to frontline teachers in the form of reports,these analyses often impose a high cognitive load and involve complex data presentations,making it challenging for teachers to identify key issues and apply findings to improve teaching practices.What begins as an effort to“empower”teachers with data often results in a“data burden.”This study introduces a framework for enhancing the interpretation of classroom teaching analysis reports using Chain of Thought Prompting(CoT)logic,facilitated by AI agents.The proposed framework modularly constructs AI agents based on open-source large language models(LLMs),enabling the development of AI agents for Teaching and Resarch.Using 63 real-world reports,the study evaluates the effectiveness of“CoT-based responses”compared to“standard LLM-based responses”.Results indicate that CoT-based responses achieve higher scores across multiple evaluation dimensions,particularly in accuracy,logical coherence,and professionalism.By focusing on the vertical application of AI agents in educational research,this study explores a novel pathway to transition from data burden to data empowerment.

关 键 词：大语言模型 智能体 思维链提示 课堂教学智能分析 循证教研 

分 类 号：G434[文化科学—教育学]