Collaborative Filtering Based On Error Feedback Algorithm

With the rapid development of Mobile Internet technology, social networks and e-commerce, the rapid growth of the information seriously influences the choice and access of interested information. Personalized recommendation system could help users find the items they are interested in, at the same time make most Long Tail Items be able to accurately show in front of the users who are interested in it, realize win-win situation for users and items’ producer. As the most widely used personalized recommendation algorithm, Collaborative filtering take advantage of the collective wisdom thought, it mines similar interests from the behavior information of the user community and uses in personalized recommendation, obtains good results.In practice, though, collaborative filtering recommendation system still faces the problem of cold start and sparse data. On the one hand, when a user or a project is joining the system for the first time, since there is no corresponding information so it can’t give a recommendation; On the other hand, the rapid growth of the number of the users and items makes the score matrix more sparse, the traditional similarity measurement becomes inaccurate, leading poor performance of the forecast. In order to make the measurement more reasonable and solve the problem of sparse data, researchers had improved the similar measurement from different sides, such as using users’ information and the score of time to improve the traditional similarity measurement.By studying the uniformity between user based collaborative filtering and k-Nearest Neighbor algorithm, this paper introduced the idea of the error feedback based collaborative filtering that adjust the similarity according the neighbor’s prediction performance: If the error of neighbor’s prediction is bigger, we should reduce the similarity, on the contrary, increase the similarity. On the basis of this idea, this paper improves three kinds of traditional methods, including two kinds of traditional collaborative filtering and Slope One algorithm. Finally, this paper apply the idea of the error feedback based collaborative filtering in an algorithm which has concern the time and common score factor.Through the experimental results and analysis, when measuring the similarity between users or items, the error feedback based collaborative filtering proposed in this paper can continuously adjust the similarity according to the neighbor’s recommended performance, make similar measurement more accuracy. At the same time, the idea of error feedback based collaborative filtering has a good fusion performance.