Real-Time Knowledge Tracing in Online Learning Environments Using Interpretable Transformer-Bayesian Models

Abstract

Knowledge tracing in online learning environments requires sophisticated systems capable of continuously monitoring student knowledge states while providing real-time feedback and personalized learning recommendations. Traditional knowledge tracing approaches struggle to balance computational efficiency necessary for real-time operation with the interpretability requirements essential for educational applications. The challenge lies in developing models that can process continuous streams of student interactions, maintain accurate knowledge state estimates, and provide transparent explanations of learning progress that support both automated adaptation and human educational decision-making.

This study proposes an Interpretable Transformer-Bayesian (ITB) framework that integrates transformer architectures with Bayesian inference mechanisms to enable real-time knowledge tracing while maintaining interpretability essential for educational applications. The framework employs transformer attention mechanisms to capture temporal learning patterns while utilizing Bayesian networks to model probabilistic knowledge states and provide uncertainty quantification. The integrated approach enables continuous knowledge state updating through streaming data processing while generating interpretable explanations of learning progress and knowledge mastery patterns.

Experimental evaluation using large-scale online learning platform datasets demonstrates that the proposed framework achieves 41% improvement in knowledge tracing accuracy compared to traditional real-time methods. The ITB approach results in 35% better prediction of future student performance while maintaining average processing latency under 150 milliseconds for real-time requirements. The framework successfully combines high predictive accuracy with interpretable knowledge state representations, achieving 39% improvement in explanation quality ratings from educational practitioners while supporting real-time adaptive learning applications.