关键词: 协同过滤, TF-IDF, 指数衰减函数, 时间窗口, 差异性

Abstract:

Aiming at the inaccuracy of single rating similarity calculation in traditional collaborative filtering algorithm, a collaborative filtering recommendation algorithm based on user interest and the ratings difference is proposed. Firstly, the TF-IDF is integrated into the tag weight calculation. An exponential decay function and a time window are used to capture the change of user interest. Secondly, according to the rating matrix, ratings difference similarity algorithm is defined by considering the difference of ratings, evaluation criteria, user and item influence. Finally, user interest similarity and ratings difference similarity are weighted and fused to obtain the nearest neighbors more accurately, which predicts item rating and makes recommendations. Experiments on the Movielens dataset show that the proposed algorithm can effectively improve the recommendation accuracy.

Key words: collaborative filtering, TF-IDF, exponential decay function, time window, difference