ResearchPod Summary
Traditional Knowledge Tracing (KT) methods are primarily designed for single-domain learning, often failing to account for the complexities of real-world scenarios where students engage with multiple subjects simultaneously. This paper addresses the challenge of accurately tracking student knowledge states in multi-domain environments, specifically focusing on how cognitive load and knowledge transfer impact learning outcomes.
The authors propose a novel method called LT-MKT (Learning-based cognitive Load and knowledge Transfer for Multi-domain Knowledge Tracing). The framework consists of three main components:
Extensive experiments on four real-world datasets (including proprietary data from iFLYTEK) demonstrate that LT-MKT consistently achieves state-of-the-art performance compared to 11 baseline models. The ablation studies confirm that both the cognitive load module and the knowledge transfer mechanism are essential for accurate performance prediction. Furthermore, the model shows superior generalization in cold-start scenarios, where it successfully infers knowledge states for concepts not seen during training by leveraging the hierarchical graph structure.
As online learning platforms increasingly offer multi-disciplinary curricula, standard KT models that treat domains as isolated silos become less effective. This work provides a scalable, LLM-assisted approach to bridge these domains, offering a more nuanced understanding of how students manage cognitive effort and transfer knowledge across subjects. This is a significant step toward more adaptive and personalized intelligent tutoring systems.
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.