A Collaborative Filtering Algorithm of Selecting Neighbors for Each Target Item

Abstract

Traditional User-based collaborative filtering recommendation algorithm in the calculation of similarity between users only considers the users' score to the item, but not takes the difference of rated items into account. Aiming at the shortcomings of the traditional method, with the practical application of recommendation system, a new collaborative filtering recommendation algorithm is proposed which selects neighbors for each target item. Ratings based on item type determine preliminary neighbors from the users, for each target item computing neighbors of the target user, and in the case of not rating the target item, the expanded neighbors are considered, finally predicting and recommending target items. The experimental results show that the algorithm improves the accuracy of similarity calculation and the error performance when comparing with other classic algorithms, and effectively alleviates the user rating data sparsity problem, while improving the accuracy of the forecast.