Mining Temporally-Interesting Learning Behavior Patterns
John S. Kinnebrew, Daniel Mack, Gautam Biswas · Educational Data Mining · 2013
Identifying sequential patterns in learning activity data can be useful for discovering, understanding, and ultimately scaffolding student learning behaviors in computer-based learning environments. Algorithms for mining sequential patterns generally associate some measure of pattern frequency in the data with the relative importance or ranking of the pattern. However, another important aspect of these patterns is the evolution of their usage over the course of a student’s learning or problem-solving activities. In order to identify and analyze learning behavior patterns of more interest in terms of both their overall frequency and their evolution over time, we present a data mining technique that combines sequence mining with a novel information-theoretic, temporal-interestingness measure and a corresponding heat map visualization. We demonstrate the utility of this technique through application to student activity data from a recent experiment with the Betty’s Brain learning environment and a comparison of our algorithm’s pattern rankings with those of an expert. The results support the effectiveness of our approach and suggest further refinements for identification of important behavior patterns in sequential learning activity data.