Inferring learners' affinities from course interaction data

Abstract

A data-driven model where individual learning behavior is a linear combination of certain stylized learning patterns scaled by learners' affinities is proposed. The absorption of stylized behavior through the affinities constitutes "building blocks" in the model. Non-negative matrix factorization is employed to extract common learning patterns and their affinities from data ensuring meaningful non-negativity of the result. The empirical learning patterns resulting from the actual course interaction data of 111 students are connected to a learning style system. Bootstrap-based inference allows to check the significance of the pattern coefficients. Dividing the learners in two groups "failed" and "passed" and considering their mean affinities leads to a bootstrap-based test on whether the course structure is well-balanced regarding the learning preferences.