Towards a unified framework for knowledge tracing with graph convolutional and neural architectures

This paper sets out to propose a unified theoretical framework for knowledge tracing (KT) that combines graph convolutional networks (GCNs) with neural sequence architectures in intelligent tutoring systems. While existing methods have achieved some success, they face limitations in modelling relational dependencies among concepts and the temporal progression of learner behaviour. Building on socio-constructivist views of knowledge as a network of relations and connectionist accounts of learning as adaptation over time, the framework integrates graph-based relational reasoning with sequence-based temporal modelling. The argument advanced here is that the integration offers interpretable representations of knowledge states while preserving predictive performance. The paper draws together recent developments in graph-enhanced KT and attention-based models and outlines design heuristics for scalable deployment. Key issues are identified, including computational cost, data sparsity, and explainability for classroom use. It is anticipated that the framework will inform the design of more systems and provide a tractable agenda for empirical validation across multiple domains and learner populations.