人工智能赋能高职适应性学习的实证分析——基于两轮教学医学生心理课程图谱学习数据

Empirical Analysis of Artificial Intelligence Empowering Adaptive Learning in Higher Vocational Education—Based on Learning Data from Two Rounds of Psychological Curriculum Knowledge-Graph-Based Teaching

摘要: 目的：线上教学平台普遍面临建成后长期按初始设计运行、真实学习数据未高效使用的问题。本研究尝试用平台自身产生的教学数据驱动平台迭代，并检验迭代前后的教学效果差异。方法：以《大学生心理健康教育》课程为载体，对2024级223名学生和2025级183名学生先后开展两轮教学：第一轮使用平台初版，运行结束后依据数据诊断出知识点颗粒度失衡、题库测评失真、路径缺乏分层三类问题，针对性完成三项迭代后开展第二轮教学，比较两轮学习行为与期末成绩。结果：两轮学生的掌握度(96.68%与94.27%)和进度(98.96%与98.70%)均处于高位且基本持平，第二轮期末成绩却由87.44分升至94.96分；行为指标与成绩的相关系数趋近于零(p > 0.05)。结论：成绩提升与行为投入脱钩，指向平台迭代而非学习时长，为“数据诊断–靶向改进–跨轮验证”的闭环模式提供了实证支持；完成型行为指标在平台成熟期已经失效，平台建设应及时转向过程性指标。

Abstract: Objective: Online teaching platforms commonly suffer from being operated long-term under their initial design after construction, with the authentic learning data they generate left underutilized. This study attempted to use teaching data generated by the platform itself to drive platform iteration, and to examine the differences in teaching effectiveness before and after the iteration. Methods: Taking the course “Mental Health Education for College Students” as the carrier, two rounds of teaching were conducted successively among 223 students of the 2024 cohort and 183 students of the 2025 cohort. The first round was run on the initial version of the platform. After the round, data-based diagnosis identified three categories of problems—unbalanced granularity of knowledge points, distorted assessment by the item bank, and lack of tiering in learning paths—based on which three targeted iterations were completed before the second round of teaching. Learning behavior and final examination scores were compared between the two rounds. Results: Mastery rates (96.68% and 94.27%) and learning progress (98.96% and 98.70%) remained at high levels and were largely comparable across the two rounds, whereas the mean final examination score in the second round rose from 87.44 to 94.96. The correlation coefficients between behavioral indicators and examination scores were close to zero (p > 0.05). Conclusion: The improvement in scores was decoupled from behavioral engagement, pointing to platform iteration rather than increased learning time as the underlying driver, which provides empirical support for the closed-loop model of “data-based diagnosis → targeted improvement → cross-round verification”. Completion-type behavioral indicators have become ineffective in the mature stage of platform operation, and platform construction should shift toward process-oriented indicators in a timely manner.