Collaborative filtering via temporal euclidean embedding

Recommender systems are considered as a promising approach to solve the problem of information overload. In collaborative filtering recommender systems, one of the most accurate and scalable algorithms is matrix factorization. As an alternative to this popular latent factor model, Euclidean embedding model presents the relationship between users and items intuitively, and generates recommendations fast. In this paper, a temporal Euclidean embedding (TEE) model is proposed by incorporating temporal factors of rating behavior. Through experiments on Netflix and Movielens data sets, we show the improvement of prediction accuracy, while keeping the efficiency of recommendation generation.