Capacity Tracing-Enhanced Course Recommendation in MOOCs

Abstract

Massive open online courses (MOOCs) have been an important learning tool in education. In order to reduce the high dropout rate and improve learners' satisfactions, it is urgent for MOOCs platform to provide course recommendation and tutoring service. To achieve it, it is necessary to determine and trace learners' learning state. Cognitive diagnosis in psychometric is a good way to quantify learners' capacities, but it demands explicit learner feedback, which does not always exist in MOOCs platform, such a typical weak-interaction scenario. Therefore, in this article, multidimensional item response theory (MIRT) is exploratively integrated into recommendation models in MOOCs by introducing a time-effectiveness hypothesis to obtain the implicit response on a followed course. To dynamically update learners' capacities by considering real-time and capacity multidimensionality, MIRT is extended to a capacity tracing model. The estimation for learner capacity is treated as attributes and integrated into collaborative filtering framework in course recommendation. To the best of our knowledge, this is the first work to integrate capacity tracing into course recommendation in MOOCs. Extensive experiments are conducted on a real-world dataset, demonstrating that the capacity tracing-enhanced course recommendation has improved effectiveness and explainability in MOOCs.