Multi-criteria recommendation system based on deep matrix factorization and regression techniques

Recommender systems have emerged as effective solutions for managing information overload by providing personalized predictions. Nowadays a deep learning-based recommendation system has achieved remarkable outcomes, but the majority of these systems are single criteria recommenders that rely on a single rating. Multi-criteria recommender systems have recently become more popular as they provide better recommendations than single criteria. In this paper, our proposed work combines the power of deep matrix factorization and regression techniques for recommendation on multi-criteria ratings. First we have used deep matrix factorization to capture latent factors in user-item interactions on multi-criteria ratings. Secondly the recommendations generated by this model are aggregated using lasso and linear regression techniques to improve the overall recommendation accuracy. Experimental analysis is conducted on two datasets TripAdvisor and Yahoo!movie demonstrated that our proposed recommender system outperforms as compared to traditional recommendation models in terms of accuracy.