ICD: A new interpretable cognitive diagnosis model for intelligent tutor systems

Abstract

Numerous models have been proposed for cognitive diagnosis in intelligent tutoring systems. However, the existing models still have room for improvement: (1) they ignore the interaction among knowledge concepts and (2) they ignore the quantitative relation between exercises and concepts. Here, we propose a cognitive diagnostic model comprising three layers of novel neural networks called ICD to solve the above two problems. Specifically, the first layer fits the influence of exercises on concepts, the second layer fits the interaction between concepts, and the third layer fits the influence of concepts on exercises. The three layers allow ICD to effectively distinguish learners with different cognitive levels, that is, ICD has good interpretability. The experimental results show that both the performance and interpretability of ICD are better than those of the latest state-of-the-art CDMs such as RCD, NCDM, and CDGK, and classical CDMs such as DINA and MIRT.