Abstract:

The collaborative filtering algorithm applied to the personalized recommendation system is a great success, which generate recommendations through the users rating data on the program and mutual cooperation between users or projects. However, the inaccurate similarity calculation among neighbors becomes the main obstacle to further improve the accuracy of the recommendation system. To improve the calculation accuracy of similarity among users, we propose an improved collaborative filtering algorithm by considering different characteristics, strengthening the mean effect and punishing the proportion of popular items, attempting to generate a more reasonable set of neighbor users. Finally we predict the scores according to the prediction equation  and ultimate recommendations are generated. Experiments on the MovieLens datasets show that the proposed algorithm can calculate the similarity among users more accurately, and the prediction accuracy is improved significantly.

Key words: collaborative filtering recommendation;recommendation accuracy;similarity;set of neighbor users