This study proposes a federated heterogeneous graph neural network framework for enhancing lecturer digital competence through privacy-preserving, cross-institutional collaboration. Traditional recommender systems frequently encounter challenges associated with data silos and privacy constraints, thereby limiting their capacity to deliver personalized professional development recommendations. The proposed framework addresses these challenges by modeling lecturer–institution–resource interactions as a heterogeneous graph, wherein nodes represent lecturers, institutions, courses, and resources, while edges capture their complex relational structures. A relation-aware graph attention network is employed to learn node embeddings locally, thereby enabling institutions to train models without sharing raw data. Furthermore, the framework integrates federated split learning with differential privacy, ensuring that intermediate outputs are perturbed with Gaussian noise prior to secure aggregation. The global model generates personalized recommendations by computing compatibility scores between lecturer and resource embeddings, subsequently ranking these to suggest relevant micro-courses or workshops. Experimental results demonstrate that the framework achieves 94.3% of centralized model performance while maintaining provable (1.0, 10⁻⁵)-differential privacy guarantees, significantly outperforming existing federated baselines in both recommendation accuracy and system efficiency. These findings contribute to the growing body of knowledge on data-driven human resource development and institutional data governance in higher education.
Copyrights © 2026