Retweet or not?

With Twitter being widely used around the world, users are facing enormous new tweets every day. Tweets are ranked in chronological order regardless of their potential interestedness. Users have to scan through pages of tweets to find useful information. Thus more personalized ranking scheme is needed to filter the overwhelmed information. Since retweet history reveals users' personal preference for tweets, we study how to learn a predictive model to rank the tweets according to their probability of being retweeted. In this way, users can find interesting tweets in a short time. To model the retweet behavior, we build a graph made up of three types of nodes: users, publishers and tweets. To incorporate all sources of information like users' profile, tweet quality, interaction history, etc, nodes and edges are represented by feature vectors. All these feature vectors are mapped to node weights and edge weights. Based on the graph, we propose a feature-aware factorization model to re-rank the tweets, which unifies the linear discriminative model and the low-rank factorization model seamlessly. Finally, we conducted extensive experiments on a real dataset crawled from Twitter. Experimental results show the effectiveness of our model.