关键词: 推荐系统, 时序效应, 奇异值分解, 项目流行度, 协同过滤

Abstract: Abstract: The traditional Singular Value Decomposition (SVD) recommender model based on sequential effect only considered scoring matrix and used complicated time function to fit item’s life cycle and user’s preferences, which led to many problems, such as difficult to explain model, inaccurate to capture user’s preferences and low prediction accuracy and so on. In view of the drawbacks, an improved sequential effect model which considered scoring matrix and item attributes and user rating labels comprehensively was proposed. The mole divided time axis into different periods at first and then achieved improvement of item bias through influence of period popularity which converted to [0,1] by sigmoid function. In addition, it also achieved improvement of user bias through nonlinear transformation of period deviation of user rating mean and overall rating mean. What’s more, this model achieved improvement of interaction through capturing users’ period preferences for items and items’ feedback rate from similar user groups. The experiment tested on Movielence 10M and 20M rating dataset shows that our improved model has yield excellent improvements in prediction of ratings with 2.5% RMSE increase on average.

Key words: recommender system, sequential effect, singular value decomposition(SVD), item popularity, collaborative filtering