User based Collaborative Filtering using fuzzy C-means

• We assess the applicability of Fuzzy Clustering to user-based Collaborative Filtering.
• Three clustering methods compared in terms of recommendation performance.
• For each of defuzzification and prediction methods we try two different approaches.
• COF defuzzified C-means with Pearson prediction yields best results.
• Greater cluster numbers don’t always achieve better results.

Today, users are surrounded by many items. Recommender Systems are used to help users find items of interest. Collaborative Filtering is one of the most successful techniques of Recommender Systems, which seeks to find users most similar to the active one in order to recommend items. In Collaborative Filtering, clustering techniques can be used for grouping the most similar users into some clusters. Fuzzy Clustering as one of the most frequently used clustering techniques, has not been used in user-based Collaborative Filtering yet. In this paper, a fuzzy C-means approach has been proposed for user-based Collaborative Filtering and its performance against different clustering approaches has been assessed. The MovieLens dataset is used to compare different clustering algorithms. They are evaluated in terms of recommendation accuracy, precision and recall. The empirical results indicate that a combination of Center of Gravity defuzzified Fuzzy Clustering and Pearson correlation coefficient can yield better recommendation results, compared to other techniques.