SOFA: Statistic Based Collaborative Filtering Algorithm

The classic user-based collaborative filtering algorithm has some shortcomes in its similarity calculation. In this paper, we propose a statistic based collaborative filtering algorithm (SOFA). The contributions are three-fold: 1) a threshold is used to filter those inaccurate similarities between users who have less intersection, 2) users´ statistics, such as mean, and variance, are used for similarity measurements, 3) two similarities are aggregated for more accurate prediction. The experiments are conducted on MovieLens data set, and the results show that the proposed method performs better than traditional ones in several popular metrics, i.e. MAE, Coverage, Precision, Recall, and F-measure etc.