Modeling Temporal Dynamics of User Preferences in Movie Recommendation

Users in movie recommender systems are likely to change their preferences over time. Modelling the temporal dynamics of user preferences is essential for improving the recommendation accuracy. In this paper, we propose an approach to model temporal dynamics of user preferences in movie recommendation systems based on a coupled tensor factorization framework. We weigh the past user preferences and decrease their importance gradually by introducing an individual time decay factor for each user according to the rate of his preference dynamics. We exploit users' demographics as well as the extracted similarities among users over time, aiming to enhance the prior knowledge about user preference dynamics, in addition to the past weighted user preferences to generate movie recommendations. Our experiments on the public benchmark dataset, MovieLens show that our model outperforms other competitive methods and is more capable of alleviating the problems of cold-start and data sparsity.