QCCDM: A Q-Augmented Causal Cognitive Diagnosis Model for Student Learning

Shuo Liu,Hong Qian,Mingjia Li,Aimin Zhou

DOI: https://doi.org/10.3233/faia230434

2023-01-01

Abstract:Cognitive diagnosis is vital for intelligent education to determine students’ knowledge mastery levels from their response logs. The Q-matrix, representing the relationships between exercises and knowledge attributes, improves the interpretability of cognitive diagnosis models. However, completing the Q-matrix poses an expensive and challenging task due to the fine-grained division of knowledge attributes. Moreover, a manually sparse Q-matrix can also compromise the accuracy and interpretability of deducing students’ mastery levels, especially for infrequently observed or unseen knowledge attributes. To address this issue, this paper proposes a Q-augmented Causal Cognitive Diagnosis Model (QCCDM) for student learning. Specifically, QCCDM incorporates the structure causal model (SCM) to capture the causality between students’ mastery levels on different attributes, which enables to infer their proficiency on rarely observed knowledge attributes with better accuracy and interpretability. Notably, with SCM, one can guide students on how to realize their self-improvement through intervention. Furthermore, we propose to augment the Q-matrix in QCCDM, which uses the manual Q-matrix as a prior to deduce the relationships between exercises and explicit as well as latent knowledge attributes, resulting in a complete and comprehensive assessment of students’ abilities. We assess the efficacy of Q-augmentation across the widely-used Q-based cognitive diagnosis models and conduct the ablation study. The extensive experimental results on real-world datasets show that QCCDM outperforms the compared methods in terms of both accuracy and interpretability.