Large-scale interpretability and accuracy of RegKT

Determine whether the RegKT model can maintain its interpretability while achieving greater predictive accuracy when applied to large-scale educational datasets.

Background

The paper presents RegKT as a hybrid of Deep Knowledge Tracing and Item Response Theory, using an IRT-based regularization term to improve interpretability and mitigate overfitting. The reported experiments use deliberately small datasets, while real-world educational systems may involve millions of students and diverse learning tasks. The authors therefore identify the unresolved question of whether RegKT can preserve its interpretability under large-scale conditions while also improving predictive accuracy.