Extracting Access Patterns with Hierarchical Latent Tree Analysis: An Empirical Study on an Undergraduate Programming Course

Abstract

Students commonly access learning resources with online systems nowadays. The access logs of those learning resources potentially provide rich information for course instructors to understand students. However, when there are many learning resources, the insights on students could be obscured by the large amount of unprocessed log data in its raw form. In this paper, we propose to extract the access patterns using a recently proposed method for topic modeling. The method, named hierarchical latent tree analysis, can capture co-occurrences of access to learning resources and group related learning resources together. We further show a way to consider the access periods during a course from the log data to reflect different behaviors of the students. We empirically test the proposed method using the log data captured in an undergraduate programming course with 63 students. We evaluate the performance of the proposed method based on the task for predicting students’ scores. The regression model built on the features extracted by the proposed method shows significantly better performance over three baseline models. We further demonstrate meaningful insights on the student behaviors can be obtained from the extracted access patterns.