Comparison of various metrics used in collaborative filtering for recommendation system

Collaborative filtering technique for generating recommendation uses user's preferences to find other users most similar to the active user and recommends new items to the user. The task of calculating the similarity is the heart of collaborative filtering approach. In this paper, we have compared various similarity metrics which are used in collaborative filtering approach for recommendation system. We have studied these metrics for both user-based approach, which determines relationships among users of similar taste and item-based approach, which aims to determine the relationships indirectly, by considering the relationship among different items. For each of the two approaches, we have compared similarity and distance metrics like Euclidean distance, Tanimoto coefficient, Pearson correlation etc. To evaluate these metrics for both user-based and item-based approach of collaborative filtering, we conducted a simple data mining experiment on MovieLens dataset for building a movie recommendation system. Finally for performance evaluation we compared our result against performance measures like accuracy, sensitivity, Mathew's coefficient etc.