Research on personalized recommendation algorithm combined with time factor

The traditional collaborative filtering algorithm is widely used in the field of recommendation. This method recommends items to users based on what items other similar users have chosen, aiming to recommend the top N items that users most likely favorite. However, it often provides unsatisfactory results, because it ignores some contextual information such as social tagging and time changing. Tags are some keywords to annotate or categorize items. Incorporating contextual information in recommender systems can improve accuracy of recommendation and enhance the overall experience of users. For improving the current collaborative filtering algorithms, we propose an algorithm based on tensor factorization and combined with time change. This algorithm uses three typical entities, i.e., users, items and tags, to build recommendation model, it is presented as a three-dimensional tensor. The latent relation is performed by Higher Order Singular Value Decomposition (HOSVD) algorithm. Besides, the algorithm considers the change of user interests over time. We highlight the importance of users' recent data followed the Ebbinghaus memory curve, calculate data weight according its time. Then, the weight value is brought into the three-order tensor model. Finally, the algorithm makes the recommendation of items to users. Experimental results show that this approach has improvements compare to collaborative filtering algorithm and HOSVD algorithm in F1-Measure with two real world datasets.