中国电化教育

针对数学课堂中学生提问“有数量,缺质量”的现实困境,该研究以Know-Want-Learn(KWL)元认知策略为理论内核,构建了深度融合教育理论的数学问题评价智能体。该智能体设计了“动态记忆库—任务规划—行动反馈—教育大模型”的协同架构,将KWL元认知策略的认知框架转化为可计算、可执行的评价逻辑,并依托基于扎根理论与德尔菲法研制的“三维度及十要点”量规与检索增强生成(RAG)技术实现自动化诊断与反馈。在初中数学教学中的准实验结果表明,该智能体能显著提升学生的问题提出能力与问题提出态度。过程性日志分析进一步揭示,智能体通过支持“提问—反馈—优化”的迭代循环,促进了学生元认知策略的内化与思维深化。同时,智能体承担了规模化认知诊断职能,减轻了教师负担,推动了其角色向“引导者”转型,为人机协同支持课堂个性化评价提供了理论印证与实践范例。

Addressing the practical dilemma of students' problem-posing in mathematics classrooms characterized by “quantity over quality,” this study takes the Know-Want-Learn(KWL) metacognitive strategy as its theoretical core and constructs a mathematical problem-posing evaluation agent that deeply integrates educational theory. The agent adopts a collaborative architecture of “knowledge memory base-task planning-action feedback-educational foundation model,” transforming the cognitive framework of the KWL metacognitive strategy into computable and executable evaluation logic. It leverages a “three-dimension and ten-aspect” rubric developed through grounded theory and the Delphi method, along with Retrieval-Augmented Generation(RAG) technology, to achieve automated diagnosis and feedback. Results from a quasi-experiment in middle school mathematics teaching show that the agent significantly enhances students' problem-posing ability and attitudes toward problem-posing. Process log analysis further reveals that the agent supports an iterative cycle of “ posing-feedback-refinement,” facilitating the internalization of metacognitive strategies and the deepening of students' thinking. Meanwhile, the agent undertakes large-scale cognitive diagnostic functions, reduces teachers' workload, and promotes their role transition toward “facilitators,” providing theoretical validation and a practical paradigm for human-AI collaborative personalized assessment in classrooms.