Information passing in online recommendation

Abstract

In this paper, we analyze the information passing in an online recommendation system. Our dataset consists of a "read" network between users and books, and a user-follower network. We first investigate in general, if one's recommendations have impacts on her followers' decisions. We then analyze the correlation between one's influence and her network centrality. Finally, we investigate how recommendation effectiveness changes with the recommendation number. This investigation is taken from both senders' and receivers' perspectives. Results show that a user does have influence over her followers' decisions. Such influence is not correlated with her centrality. The more a book is recommended, the more likely that one will accept it. However, there is a saturate point beyond which more recommendations will have no more impact. On the other hand, the more recommendations one makes, the more likely that her recommendations will be accepted. This trend has no saturate point. Copyright © 2013 ACM.