Analysis of Student Learning Depth Progression in Programming Courses Through Video Annotation

In this study, we propose quantifying students' learning depth in video programming courses by analyzing their video annotations using natural language processing techniques and Bloom's Digital Taxonomy (BDT). Utilizing a video-based Learning Management System that allows direct annotation of video frames, we collected 3,014 annotations from 42 students enrolled in a “Data Management” course. We optimized Japanese morphological analysis tools by integrating a custom IT terminology dictionary to enhance the recognition of domain-specific terms. A specialized corpus was constructed, and a domain-specific Word2vec model was trained to expand the BDT classification table through word embeddings. This expansion enabled the comprehensive classification of keywords extracted from student annotations into cognitive process categories. The results revealed a clear shift from lower-order to higher-order cognitive processes as the course progressed, indicating significant changes in students' learning depth influenced by instructional content and teaching strategies. The method effectively captured dynamic trajectories of cognitive development, and these findings provide educators with valuable insights to conduct nuanced assessments and enhance personalized learning experiences. This study provides a practical method for quantifying the depth of learning in online programming education, advancing the understanding of students' cognitive processes, and supporting the improvement of teaching practices through data-driven adjustments to instructional design.