Addressing Label Leakage in Knowledge Tracing Models
This addresses a specific issue in educational AI for improving student performance prediction in intelligent tutoring systems, though it is incremental as it builds on existing KT models.
The paper tackles the label leakage problem in knowledge tracing models, where correlations between knowledge concepts (KCs) belonging to the same item can leak ground truth labels and degrade performance, especially in datasets with many KCs per item. The authors present methods to prevent this leakage, with their model variants consistently outperforming original counterparts and one variant surpassing all tested baselines on different benchmarks.
Knowledge Tracing (KT) is concerned with predicting students' future performance on learning items in intelligent tutoring systems. Learning items are tagged with skill labels called knowledge concepts (KCs). Many KT models expand the sequence of item-student interactions into KC-student interactions by replacing learning items with their constituting KCs. This approach addresses the issue of sparse item-student interactions and minimises the number of model parameters. However, we identified a label leakage problem with this approach. The model's ability to learn correlations between KCs belonging to the same item can result in the leakage of ground truth labels, which leads to decreased performance, particularly on datasets with a high number of KCs per item. In this paper, we present methods to prevent label leakage in knowledge tracing (KT) models. Our model variants that utilize these methods consistently outperform their original counterparts. This further underscores the impact of label leakage on model performance. Additionally, these methods enhance the overall performance of KT models, with one model variant surpassing all tested baselines on different benchmarks. Notably, our methods are versatile and can be applied to a wide range of KT models.