Improving Knowledge Tracing via Considering Conceptual Structure and Individual Differences
Aihua Mao, Jiaming Chen, Yong‐Jin Liu · 2024
The Knowledge Tracing (KT) task aims to track changes in students’ knowledge state based on their historical answer sequences and predict student’s future performance. Existing KT models have two limitations. One is that these models do not fully utilize the associative structure between knowledge concepts. Knowledge concepts are not isolated but are interrelated and have an important influence on each other, making their correlations a critical piece of information. The other limitation is that many KT models do not pay sufficient attention to individual differences among students in the learning process. Different students have different learning abilities, whose impact on students’ knowledge state should not be overlooked. In this paper, we propose a Concept Structure Key-Value Memory Network knowledge tracing model (CSKVMN) that considers the associative structure of knowledge concepts and students’ learning abilities. Experimental results on two public datasets show that CSKVMN can better track students’ knowledge state and outperforms existing models in terms of Area Under the Curve.