Data Analytics for Modeling User Behavior Within Massive Open Online Courses: A Comparison of Clustering Techniques

• In Event: 29.074 - Online Teaching and Learning SIG Paper Session 1

Abstract

Massive Online Open Courses (MOOCs) are a way for learners to engage with educational content (DeBoer, Ho, Stump, & Breslow, 2014). We examine the efficacy of three techniques in clustering users into groups based on participation level within a MOOC. Data from 1085 students were coded dichotomously (participated, or did not participate), and three clustering techniques were implemented; hierarchical, two-step, and latent class growth analysis (LCGA). Clusters from hierarchical method tended to overlap more than either two-step or LCGA clusters, and cluster agreement with other methods was poor. Overall, two-step and LCGA both identified patterns parsimonious with theory and practice. LCGA discovered a small, previously unidentified cluster that will be helpful moving forward with future research modeling user interactions.

Authors

• Megan Frankosky, North Carolina State University

• Jennifer London, North Carolina State University

• Isaac Benjamin Thompson, North Carolina State University

• Tara Behrend, The George Washington University

• Eric N. Wiebe, North Carolina State University