Predicting who will retweet or not in microblogs network

Abstract

Retweeting often leads to fast and wide information propagation in microblogs. There has been research concerning predicting users’ retweeting behavior. However, it lacks of in-depth investigations on key features that contribute to the prediction accuracy. In this paper, we systematically examined four types of features: followee (the user who posted tweets) features, follower features, tweet features and interaction features in terms of their contribution to retweeting prediction accuracy. We collected Sina Weibo data and ranked the features extracted from original data on importance by using an information gain method. There were twelve features showing significant association with the retweeting predictive accuracy through lots of experiments. Experimental results showed that using these twelve features, we could train a novel predictive model and achieved the predict accuracy up to 98%.