collaborative filtering (CF) aims to predict users' ratings on items
according to historical user-item preference data. In many real-world
applications, preference data are usually sparse, which would make models
overfit and fail to give accurate predictions. Recently, several research