Similarity of user ratings play a positive information filter in social networks

In the past decade, social recommending systems have attracted increasing attention from the physical, social and computer science communities. In this study, we use social networks to capture similarities of users' interest and, accordingly, recommending systems to explore latent similarities. We build similarity-ratings-prediction models for a dataset of books and reviewers from Douban.com. The similarities are measured by the Pearson and the Jaccard correlation coefficients. Using singular value decomposition, we evaluate the strengths and the weaknesses of the similarity measure, and discuss their effectiveness in recommending systems.