Memory Attentive Cognitive Diagnosis for Student Performance Prediction

Abstract

As the core of intelligent education, cognitive diagnosis aims to capture the proficiency of students on specific knowledge concepts. The Neural Cognitive Diagnosis (Neural CD) exploits an elegant method to simulate the interactions of student exercising process with deep neural networks. However, Neural CD still treats the student factor as static, which doesn’t change after the exercising process, against the common sense that a student will gain better proficiency after practices. Furthermore, the Neural CD focuses on the current exercise despite leveraging multi fully connected layers to model the complex interactions, ignoring the relationship between exercises. In this paper, we propose the Memory Attentive Cognitive Diagnosis (MACD) for student performance prediction. Specifically, MACD introduces memory-augmented neural networks to express the student factor, which can be updated with the process of solving exercises. Moreover, MACD replaces the multi fully connected layers with multi-heads attention layers to consider the relationship between the current exercise and the past exercises. We conduct experiments on several real-world datasets and the experimental results show that our model outperforms the state-of-the-art approaches.

Supported by the National Natural Science Foundation of China (U1811261).