Using Topic Modeling to Research Student Diversity in Higher Education

Abstract

Abstract: Groups in computer-supported collaborative learning (CSCL) are characterized by diversity regarding several sociodemographic and task-related features (e.g., students differ with respect to their first language and their prior experience with online learning simultaneously). Understanding the effect of this multiattributional diversity on CSCL groups is necessary to inform the design of interventions aimed at managing CSCL group diversity. We used topic modeling to investigate how multiattributional diversity affects discussion topics and subsequent group performance. Twenty topics were derived by Latent Dirichlet Allocation and assigned to 17,720 forum posts by 342 groups of psychology freshmen. Multiattributional sociodemographic diversity was negatively related to the number of posts in two topics concerned with time management and feedback mechanisms in groups where multiattributional task-related diversity was also high. These topics were positively related to subsequent group performance ratings. Findings suggest that this combination of diversity presents a risk for group communication, potentially negatively impacting performance.