A novel cognitive diagnosis model with multiple attributes and multidimensional discrimination

In intelligent tutoring systems, cognitive diagnosis represents a crucial task. The advancement of machine learning, particularly deep neural networks, has significantly propelled the field of computer-assisted education. Although neural network cognitive diagnosis models (NNCDMs) capture non-linear interactions between students and exercises more effectively, they may neglect essential information, such as the relations between knowledge points in exercises and students' latent abilities, with skill level characterization often oversimplified. Furthermore, although fully connected layers in NNCDMs capture nonlinear relationships more effectively, they may also result in overfitting on educational response data sets. To address these challenges, this paper introduces a neural network cognitive diagnosis model(MACD) that incorporates multiple attributes and multidimensional discrimination. The model integrates students' explicit knowledge proficiency and latent potential abilities, investigates the intrinsic interrelationships between knowledge points via a dedicated relation matrix, and refines the discriminative power of each exercise for individual knowledge points through multidimensional discrimination to improve the fine-grained representation of student cognitive abilities. The experimental results on three public educational data sets (ASSIST0910, ASSIST17, FrcSub) with core evaluation metrics (accuracy (ACC), root mean square error (RMSE), area under the curve (AUC)) indicate that the proposed model demonstrates superior performance in predicting student outcomes, achieving an ACC of 0.7570 and an AUC of 0.8024 on the ASSIST0910 data set and a minimum RMSE of 0.3453 on the FrcSub data set–outperforming all traditional statistical and state-of-the-art neural CDM baselines on all metrics.

Actions (login required)

View Item