Adapting Bayesian Knowledge Tracing to a Massive Open Online Course in edX

Abstract

Massive Open Online Courses (MOOCs) are an increasingly pervasive newcomer to the virtual landscape of higher-education, delivering a wide variety of topics in science, engineering, and the humanities. However, while technological innovation is enabling unprecedented open access to high quality educational material, these systems generally inherit similar homework, exams, and instructional resources to that of their classroom counterparts and currently lack an underlying model with which to talk about learning. In this paper we will show how existing learner modeling techniques based on Bayesian Knowledge Tracing can be adapted to the inaugural course, 6.002x: circuit design, on the edX MOOC platform. We identify three distinct challenges to modeling MOOC data and provide predictive evaluations of the respective modeling approach to each challenge. The challenges identified are; lack of an explicit knowledge component model, allowance for unpenalized multiple problem attempts, and multiple pathways through the system that allow for learning influences outside of the current assessment.