MVAC: A Multimetric Distance-Aware Variational Autoencoder Framework for Personalized Course Recommendation

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Purpose This research addresses limitations in existing course recommendation systems, including sparse interaction data, weak contextual understanding, and limited adaptability to diverse educational features. Design/methodology/approach This study proposes MVAC, a multimetric distance-aware hybrid Variational Autoencoder framework for personalized course recommendation in online learning environments. The proposed framework combines collaborative filtering with VAE-based latent representation learning and integrates cosine distance with Euclidean distance for content similarity estimation. The system processes course titles, organizations, ratings, and difficulty levels using TF-IDF feature extraction, embedding layers, convolutional blocks, and latent interaction modeling to generate top-K personalized recommendations. The framework captures latent user-course relationships while improving semantic similarity analysis between courses. An experimental evaluation was conducted using multiple Coursera educational datasets across varying training configurations. Findings The proposed MVAC framework outperformed MR-CNN, AdaBoost, collaborative filtering, and Bi-LSTM-based approaches. The model achieved an R 2 of 0.84, an NDCG@10 of 0.98, and an MAE of 1.24 on the Coursera Course Dataset. The integration of multimetric similarity learning and VAE-based collaborative filtering improved recommendation diversity, contextual relevance, and ranking quality under sparse interaction conditions. Research limitations The experimental evaluation was conducted using multiple Coursera educational datasets across varying training configurations. Practical implications The framework supports intelligent recommendation services for e-learning platforms by helping learners identify suitable courses based on their interests, course ratings, and difficulty levels while reducing information overload. Originality/value This study introduces a hybrid recommendation framework that improves the precision and ranking effectiveness of recommendations across heterogeneous educational datasets.