Mining Social Influence in Microblogging via Tensor Factorization Approach

Microblogging has become an important social media for creating, sharing, or exchanging information and ideas. Social influence analysis in Microblogging is often exploited for different tasks such as information retrieval, recommendations, businesses intelligence. Most existing methods mostly rely on social links between users, failing to take advantage of characteristics of Microblogging. Furthermore, the size of Microblogging's user (i.e. Microblogger) is very large, which makes computing resource for social influence mining approach can't be satisfied by single computer. In this paper, a tensor factorization framework based on cloud computing platform is proposed for mining social influence in Microblogging. The framework has three components: a feature extraction component, a tensor factorization component and a user influence ranking component. In feature extraction component, features are extracted for capturing user social influence quantitatively through statistical analysis on the Microbloggers' relations. In tensor factorization component, tensor factorization based MapReduce model is presented to infer user's implicit user's relations. Finally, a user influence ranking function is constructed for computing user social influence in user influence ranking component. Experiments on Sina weibo dataset (Chinese Microblogging platform) show that our proposal significantly not only improves the prediction accuracy compared with two baseline methods, but also has competitive advantage for processing massive data from Microblogging.