Distinguishing social ties in recommender systems by graph-based algorithms

Incorporating the social network information into recommender systems has been demonstrated as an effective approach in improving the recommendation performance. When predicting ratings for an active user, his/her taste is influenced by the ones of his/her friends. Intuitively, different friends have different influential power to the active user. Most existing social recommendation algorithms, however, fail to consider such differences, and unfairly treat them equally. The problem is that the friends with less influential power might mislead the rating predictions, and finally impair the recommendation performance. Some previous work has tried to differentiate the influential power by local similarity calculations, but it has not provided a systematic solution and it has ignored the propagation of the influence among the social network. To solve the above limitations, in this paper, we investigate the issue of distinguishing different users' influence power in recommendation systematically. We propose to employ three graph-based algorithms (including PageRank, HITS, and heat diffusion) to distinguish and propagate the influence among the friends of an active user, and then integrate them into the factorization-based social recommendation framework. Through experimental verification in the Epinions dataset, we demonstrate that the proposed approaches consistently outperform previous social recommendation algorithms significantly.