Abstract: With the rise of electronic reading in recent years,the use of collaborative filtering(CF) recommendation algorithm to recommend user personalized books has practical application value,and has become the important research content in the study of recommender systems.But many current e-reading systems for book recommendations lack of users’ rating data,which hinders the application of CF.To address this,we analyzed the massive users’ reading beha-vior,and proposed a reading time-frequency(T-F) model to profile the users’ interests to the book.Thus,the implicit ratings matrix can be derived from this model and then classical CF algorithm could be used in a natural way.The experimental results show that the user based CF with our proposed T-F rating model can improve the recommendation effectiveness,which is feasible for real scenarios.

Key words: Collaborative filtering,Recommendation system,User ratings matrix,User behavior,Time-frequency