Mechanistic modelling of viral spreading on empirical social network and popularity prediction

Online social networks are becoming major platforms for people to exchange opinions and information. While spreading models have been used to study the dynamics of spreading on social networks, the actual spreading mechanism on social networks may be different from these previous models due to users’ limited attention and heterogeneous interests. The tractability of the spreading process in social networks allows us to develop a detailed and realistic model accounting for these factors. In addition, the empirical social networks have higher order correlations among node degrees, especially for directed networks like Twitter, that could affect the dynamics of spreading. Based on the analysis of the retweet process in the empirical Twitter network, we find both non-trivial correlations in network structures and non-standard spreading mechanisms for viral tweets. In particular, there is a strong evidence of information overload for retweeting behaviors that is not in line with the standard spreading model like the SIR (Susceptible, Infectious and Recovered) model, and can be described by a sublinear function. From these empirical findings, we introduce an intrinsic variable “attractiveness” to the message, describing the overall propensity for any node to retweet the message, and present the analytical equations to solve such an empirical process, validated through numerical simulations. The results from our model is consistent with findings from the empirical Twitter data. Our analysis also indicates a close relationship between the spreading sub-network structure and the final popularity of the information, leading to a method to predict the popularity of tweets more accurately than existing models.