Collaborative filtering recommendation based on fuzzy clustering of user preferences

In recent years, extensive researches have been conducted to develop approaches to answer two major challenges for collaborative filtering problems, namely sparsity and scalability. In this paper, we propose a novel collaborative filtering recommendation approach to alleviate these challenges. Our approach firstly converts the user-item ratings matrix to user-class matrix, and hence increases greatly the density of the data in the resulted matrix. Next, we fuzzily partition users into different groups by using Fuzzy C-Means (FCM) algorithm. We believe this is a more reasonable and natural way of partition by preferences. Finally, we propose a novel CF top-N recommendation algorithm to generate the recommendation list directly. We provide results and evaluations of computational experiments to demonstrate that our approach does provide better computational accuracy and efficiency, and does outperform other CF approaches with respect to the metrics of precision, recall and F1.