As educational data mining (EDM) technology advances rapidly, educators can gain in-depth insights into and optimize the teaching process with advanced data analysis methods. This study combines Multidimensional Item Response Theory (MIRT) with educational data mining. It uses the MIRT package in R software to analyze college physics final exam data from three semesters, from the 2022—2023 spring semester to the 2023—2024 spring semester. The focus is on student academic performance in memory, conceptual understanding, and comprehensive application. By estimating students’ multidimensional abilities, it evaluates the teaching content and methods’ effectiveness. Results show a significant positive correlation between after class homework, classroom test scores and final exam performance across dimensions, particularly in comprehensive application ability, where homework design is key for improvement. Moreover, the study identifies underperforming classes in specific dimensions through analyzing anomalies in summative assessment data and suggests improvements. Based on cluster analysis, teaching classes are categorized into excellent, good, and needing-improvement groups, offering data-backed support for teaching improvement. Lastly, this study provides data-driven decision making support for teaching managers. It helps them identify teaching strengths and weaknesses, allocate resources better, and enhance teaching quality.

The rapid advancement of large language models (LLMs) has opened new avenues for innovation in physics teaching. This study focuses on a computational-thinking-driven methodology for creating digital and intelligent inquiry-learning scenarios with LLMs, enabling the rapid development of such scenarios for university-level smart physics courses. Using the projectile motion experiment as an example, we illustrate how AI-based inquiry learning can be employed to assess students' innovative capabilities. By mining the data generated during inquiry activities, we uncover the problem-solving strategies students adopt. Coupled with explainable AI, we can then derive scientifically grounded rubrics for evaluating their inquiry processes, thereby offering a quantifiable means of measuring innovation.

This study explores the impact of process evaluation on learning outcomes through classroom quizzes. At the student level, using a learning engagement questionnaire to measure the learning engagement of engineering students in college physics, it was found that process evaluation is an effective way not only to check classroom learning outcomes, but also to enhance student emotional engagement. At the teacher level, by analyzing college physics test data using the Rasch model, it was found that process evaluation can not only help students correct learning misconceptions in a timely manner, but also help test-making teachers optimize test question banks. In addition, course teachers can continuously construct and revise teaching content based on test feedback, which improves the quality of teaching.