autoTimeSVD++: A Temporal Hybrid Recommender System Based on Contractive Autoencoder and Matrix Factorization

Abstract

Matrix factorization is one of the successful approaches used largely in Recommender systems to provide recommendations to users based on their historical preferences. In recent years, many approaches based on deep learning, such as autoencoders, were used alone or combined with other methods to extract non-linear relationships between items. But most of these models are static and do not capture dynamic changes regarding the rating process which is dynamic and may change over time. In this paper, we propose a new hybrid model autoTimeSVD++ which combines timeSVD++ and autoencoder to extract item side information including time effect. The experimental results show that the proposed model achieves competitive results compared to many baselines models.