Discerning Influence Patterns With Poisson Factorization In Microblogging Environments

Social influence analysis in microblogging services has attracted much attention in recent years.However,most previous studies were focused on measuring user's influence.Little effort has been made to discern and quantify how a user is influenced.Specifically,the fact that user i retweets a tweet from author j could either because i is influenced by j,or simply because he is "influenced"by the content.In order to mine such influence patterns,we propose a novel Bayesian poisson factorization model,dubbed Influence Poisson Factorization(IPF).IPF jointly factorizes the retweet data and tweet content to quantify latent topical factors of user preference,author influence and content influence simultaneously.It generates every retweet record according to the sum of two causing terms:one representing author influence,and the other one derived from content influence.If the author is influential on the involved topics,the former term would dominate,while the latter term will be in charge if we observe that many similar tweets are retweeted from non-influential authors.We develop efficient variational inference algorithms for IPF.We demonstrate the efficacy of IPF on two public microblogging datasets collected from Twitter and Sina Weibo respectively.We also explore applying IPF on retweet prediction and tweet recommendation.The results show that,when integrated with state-of-art features,IPF can boost their performance.