Deep Knowledge Tracking Integrating Programming Exercise Difficulty and Forgetting Factors

To address the limitations of existing deep knowledge tracing models that often consider the factors influencing learners’ knowledge state changes from a single perspective, lacking a comprehensive analysis of multiple dimensions such as problems, learners, and knowledge points, which fails to effectively reflect the complexity and diversity of the learning process, resulting in inadequate predictive performance and interpretability, a graph neural network knowledge tracing method that integrates programming problem difficulty and forgetting factors is proposed. First, an algorithm for analyzing problem difficulty based on GPT-3 is designed to enrich the feature information at the model’s input layer. Second, five factors affecting learners’ forgetting behavior are analyzed to fit the behavioral characteristics of learners. Finally, a knowledge structure graph is constructed using graph neural networks, enabling the model to consider the complex relationships between knowledge points while addressing problem difficulty and forgetting factors during the node updating process. Experimental results demonstrate that this method offers improved predictive performance and interpretability.