Modeling the time to share fake and real news in online social networks

In this paper, we address the problem of predicting the time to share (real or fake) news items on online social media. Specifically, given the scenario where a user u is influenced on some given (real or fake) news item n by at least one of the people they follow v, we predict when the user u will re-share the news item n among their followers. We model the problem as a survival analysis task, which is a statistical analysis method aimed at predicting the time to event (the re-sharing event in our case). Survival analysis differs from other methods such as regression in that it also considers the data where the event (sharing) never occurs (censored data) in the considered time window. We considered Twitter data containing information on real and fake news shares to test our proposed survival analysis approach and modeled different aspects of the problem including user, news, and network characteristics. We show the superiority of survival analysis as compared to regression to model this problem in both the cases of real and fake news sharing.