Modeling Information Diffusion via Reputation Estimation

We tackle the problem of predicting information diffusion in social networks. In this problem, we are given social data and would like to infer the diffusion process in the near future. Although this problem has been extensively studied, the challenge of how to effectively combine user activities, network structures and diffused information in social data remains largely open. In addition, no prior work judged the effect of user reputation on the diffusion process. Availability of such reputation score is really important for a user to decide whether he might share information. In this paper, we first devise a novel method for estimating user reputation. Our approach integrates network structure with user features, link features and the content of items shared by the users, then measures the strength of each of these factors. Based on this estimation approach, we develop a model predicting the tendency of a new information item as well as the number of participants of this diffusion process. We conduct several experiments on a snapshot of Twitter which show that our proposed model outperforms other baselines.