Knowledge ontology enhanced model for explainable knowledge tracing

Abstract

Knowledge Tracing (KT) aims to predict learners’ future learning outcomes based on their past learning interactions. Deep Knowledge Tracing (DKT) is a technology developed in recent years that employs deep learning techniques to dynamically track students’ learning progress and offer personalized learning support. Yet, existing research often miss the complex interplay between student characteristics and knowledge components, limiting interpretability and adaptability in assessments. Addressing this gap, we introduce the Ontology Perceptible Knowledge Tracing (OPKT) model, a novel approach that reconceptualizes student-knowledge interaction by integrating both student and knowledge ontologies directly into the tracing process. This integration allows for a more nuanced representation of learning interactions, significantly enhancing the predictive precision and interpretability of the model. The OPKT model extracts and utilizes ontology features—deriving student attributes from behavioral data and aligning them with knowledge aspects through sophisticated relationship mapping. This dual-layer integration captures the evolution of knowledge states more accurately, facilitating targeted educational interventions. The results of prediction on two publicly available datasets demonstrate that our proposed OPKT model achieves at least a 2% improvement in AUC, accurately assessing students’ learning outcomes and knowledge proficiency. Additionally, we conducted dimensionality reduction experiments and ablation studies to validate the interpretability of the model.