Graph Knowledge Tracking Interaction Model Combining Classification and Regression Tree

Knowledge tracking aims to assess learners' current level of knowledge based on their previous response performance. In recent years, graph neural networks have been successfully applied in the field of knowledge tracking. However, existing graph-based knowledge tracking interaction models typically combine a graph convolutional neural network with a Long Short-Term Memory network. Although there is an improvement in performance, the model only considers exercises, knowledge, and answers as inputs, ignoring the impact of rich learning behavioral features on the learner's knowledge state. In this work, we propose a Graph-based Interaction Model for Knowledge Tracing with Decision Tree (GIKT-DT) that fuses classification and regression trees. Especially, to effectively capture the effect of behavioral features on answer results, predicted responses are first obtained by pre-processing learners' behavioral features using classification and regression trees. And then the cross-sectional features of predicted responses and interaction sequences are calculated as inputs to the GIKT-DT to track learners' knowledge acquisition levels more accurately. Moreover, we validate the GIKT-DT model on three publicly available online education datasets. The experimental results show that GIKT-DT outperforms other baseline models and can better utilize the behavioral characteristics of learners to improve the accuracy and effectiveness of knowledge tracking with better prediction performance.