Abstract: As the accuracy and real-time requirements of users continue to increase , it is a very valuable research direction to mine the accurate information that users need from massive historical data of users .The collaborative filtering algorithm based on fuzzy clustering must first solve the problem of data sparsity. Firstly, the original user rating data is preprocessed, and the data is filled by the SMOTE method to effectively solve the data sparsity problem. Then, the classification data is classified using fuzzy clustering. By combining Ebbinghaus's forgetting curve, the timestamp of user evaluation is used as a factor to score and predict the clustered data, in order to improve the impact of user preferences over time on the recommendation effect and solve real-time problems. Through experiments on MovieLens-100k dataset, the results show that fuzzy collaborative filtering recommendation with time  can significantly improve the recommendation accuracy.

Key words: synthetic minority oversampling technique, fuzzy clustering, collaborative filtering, scor- ing matrix, time factor