An effect-size-based temporal interestingness metric for sequential pattern mining

Sequential pattern mining is a useful technique for understanding learning behavior. However, it can be challenging to select the most “interesting” patterns discovered through sequence mining. The work presented in this paper proposes an effect-size-based (ESB) method to help researchers identify temporally interesting sequential patterns. ESB is extended from the Temporal Interestingness of Patterns in Sequences (TIPS) technique [4] and distinguishes itself by 1) considering a different association direction between the sequential pattern usage and time, 2) providing a more interpretable ranking metric, and 3) providing a different ranking order for temporally interesting sequential patterns. Both ESB and TIPS are applied to interaction log data to demonstrate their differences in selecting sequential patterns.