A new user similarity measurement based on a local item space in collaborative filtering recommendation

Collaborative Filtering (CF) is one of the most successful methods for automated product recommendation in recommender systems. The general process of CF is to compute similarities among users using users' ratings data on items and suggesting the items highly rated by a user to other similar users. The existing similarity computing algorithms assume that a pair of similar users have similar preferences on all items and take all items into account when compute similarities among users. However, the same two users have different hobbies for different types of goods. In other words, the similarity of the two users confined to a part of items. In this paper, we propose Local Item Space Distance (LISD), to improve the CF prediction accuracy. This new similarity measure distinguishes similar users in a local item cluster. That is, LISD finds different similar users for different item clusters. Experiments on two public datasets in recommend realm demonstrate the superiority of the new similarity measure in improving prediction accuracy for recommender systems. ©, 2015, Binary Information Press. All right reserved.