Improved Slope One Collaborative Filtering Predictor Using Fuzzy Clustering

Slope One predictor, an item-based collaborative filtering algorithm, is widely deployed in real-world recommender systems because of its conciseness, high-efficiency and reasonable accuracy. However, Slope One predictor still suffers two fundamental problems of collaborative filtering : sparsity and scalability, and its accuracy is not very competitive. In this paper, to alleviate the sparsity problem for Slope One predictor, and boost its scalability and accuracy, an improved algorithm is proposed. Through fuzzy clustering technique, the proposed algorithm captures the latent information of users thereby improves its accuracy, and the clustering mechanism makes it more scalable. Additionally, a high-accuracy filling algorithm is developed as preprocessing tool to tackle the sparsity problem. Finally empirical studies on MovieLens and Baidu dataset support our theory.