Personalized Knowledge Graph Embedding for Deep Adaptive Learning in Higher Education

Adaptive learning in higher education requires accurate modeling of both curriculum structures and student-specific learning behaviors. This study proposes a Personalized Knowledge Graph Embedding (PKGE) framework that integrates course concepts, prerequisite relations, assessments, and skill dependencies with multimodal student interaction signals. Personalization experiments show substantial learning benefits: students receiving PKGE-based recommendations achieved an average mastery gain of +20.4 points compared to +9.2 in the non-adaptive control group. Recommendation relevance, validated through instructor scoring, resulted in a mean NDCG@10 of 0.72, indicating that high-priority concepts were ranked effectively. Error pattern analysis further revealed a reduction in repeated conceptual errors from an average of 4.3 to 2.2 per cycle, demonstrating the model’s ability to remediate misconceptions efficiently. Longitudinal performance tracking across four learning cycles shows quiz accuracy improvements from 62% to 82% in the adaptive group, compared to a smaller increase from 61% to 70% in the non-adaptive group. Visualization of the embedding space using UMAP confirms clear semantic clustering among concepts and distinct personalization trajectories for students. Case-study analysis highlights the model’s ability to generate coherent learning paths that balance prerequisite coverage, difficulty progression, and cognitive pacing. The findings collectively demonstrate that PKGE provides a scalable, interpretable, and highly effective foundation for deep adaptive learning in higher education, surpassing traditional sequence-based and matrix-based personalization methods. This research establishes PKGE as a robust approach for future intelligent learning systems that require both semantic alignment and behavior-driven adaptation.