Model-based non-Gaussian interest topic distribution for user retweeting in social networks

Retweeting behavior is critical to dissect information diffusion, innovation propagation and events bursting in networks. However, because of the various contents of tweets, recent work mainly focuses on the influential relationship while unable to derive different pathways of information diffusion. Therefore, our work tries to reveal the pattern by tracking retweeting behavior through user interest and categories of tweets. The key for modeling user interest is modeling topic distribution of tweets, which have non-Gaussian characteristics (e.g., power law distribution), thus we present the Latent Topics of user Interest(LTI) model which make full use of the non-Gaussian distribution of topics among tweets to uncover user interest and then predict users' possible actions. After dividing users into conceit users and altruism users by whether they have definite selection when retweeting, and categorizing tweets into repeated hot tweets and novel hot tweets by whether its topics always occur in the training set, we demonstrates a pattern - the conceit users promotes the diffusion of repeated hot tweets, whereas the altruism users expands the diffusion of novel hot tweets, and the pattern is evaluated by the correlation coefficient between types of users and tweets, which is greater than .61 for 10 and 100 million tweets of Weibo2 and Twitter with respect to 70 and 58 thousand users over a period of one month.