Authors: Edmund Evangelista, Syed M. Salman Bukhari Source: Computers and Education: AI, Vol 11 — Open Access (CC BY 4.0)
Key Findings
Hybrid HKG-GRU framework for educational recommendations combining heterogeneous graph embeddings with sequential modeling (152 students, 59 resources, ~150K interactions). Multi-objective training with GroupDRO for fairness, MMR reranking for diversity, and built-in explainability through path-based and counterfactual analyses. HR@10=0.68, MRR=0.41. Addresses popularity bias and cold-start fairness in educational recommenders.