Comparative Action Sequence Analysis with Hidden Markov Models and Sequence Mining

Computer-based learning environments produce a wealth of data on student learning interactions. This paper presents an exploratory data mining methodology for assessing and comparing students’ learning behaviors from these interaction traces. In the first phase of this methodology, hidden Markov models (HMMs) are generated to model learning behaviors of the student groups being compared (e.g., high versus low performers). In this paper, we supplement the HMM technique with a novel combination of sequence mining techniques to identify differentially frequent patterns (between the student groups) in a finergrained analysis. We demonstrate the complete methodology through the analysis of learning trace data from a recent middle school classroom study with the Betty’s Brain learning environment. The results illustrate the effectiveness of this exploratory methodology and suggest further refinements of the HMM generation and sequence mining techniques.