Deep-IRT with a Temporal Convolutional Network for Reflecting Students’ Long-Term History of Ability Data

This study proposes a new Deep-IRT with a temporal convolutional network for knowledge tracing. The proposed method stores a student’s latent multi-dimensional abilities at each time point and estimates the latent ability which comprehensively reflects the long-term history of ability data. To demonstrate the performance of the proposed method, we conducted experiments using benchmark datasets and simulation data. Results indicate that the proposed method improves the performance prediction accuracy of earlier Deep-IRT methods while maintaining high parameter interpretability. The proposed method exceeds the performance of earlier methods especially when the student’s ability fluctuates according to past abilities.