AFGAKT: Forgetting Law Guided Knowledge Tracking Model by Adversarial Training
Haonan Li, Linlin Zhao, Zhenguo Zhang · 2024
With the development of smart education, a large amount of student interaction data is generated, and we can trace students' learning status from this data. It predicts students' next answers based on their previous responses, serving as a foundation for teachers' decision-making. The current knowledge tracking models lack intuitive modeling of forgetting laws, and the relationships between exercise questions have not been fully utilized. Additionally, there has been relatively little research on the model robustness. In this paper, we propose the Forgetting law guided knowledge tracking model by adversarial training (AFGAKT), which consists of an embedding module fused with forgetting laws, an feature extraction module, a pre- and post-relationship extraction module, and a prediction module. To enhance the model's robustness, we introduce adversarial perturbations into the embedding layer of the model. Finally, we propose a scoring method to provide teaching feedback. Experimental results on four educational datasets demonstrate the effectiveness of the proposed model for the knowledge tracking task.