Collaborative Filtering using Fuzzy Rank-based Similarity Measures

Collaborative filtering has been a successful technique used by many commercial recommender systems. One of the critical factors affecting the performance of these systems is the similarity measure. Various methods have been developed for similarity computation, but there are still much to be improved. This study proposes a new similarity measure for user-based collaborative filtering systems which reflects the rating behavior of users on each item onto similarity. It computes fuzzy ranks of user ratings on an item and combines them with a traditional similarity measure. Performance of the proposed measure is investigated through extensive experiments to find that it outperforms state-of-the-art weight-based similarity measures under various data environment, especially when ratings data are sparse.