Correlation between information diffusion and opinion evolution on social media

Abstract

Information diffusion and opinion evolution are often treated as two independent processes. Opinion models assume the topic reaches each agent and agents initially have their own ideas. In fact, the processes of information diffusion and opinion evolution often intertwine with each other. Whether the influence between these two processes plays a role in the system state is unclear. In this paper, we collected more than one million real data from a well-known social platform, and analysed large-scale user diffusion behaviour and opinion formation. We found that user inter-event time follows a two-scaling power-law distribution with two different power exponents. Public opinion stabilizes quickly and evolves toward the direction of convergence, but the consensus state is prevented by a few opponents. We propose a three-state opinion model accompanied by information diffusion. Agents form and exchange their opinions during information diffusion. Conversely, agents' opinions also influence their diffusion actions. Simulations show that the model with a correlation of the two processes produces similar statistical characteristics as empirical results. A fast epidemic process drives individual opinions to converge more obviously. Unlike previous epidemic models, the number of infected agents does not always increase with the update rate, but has a peak with an intermediate value of the rate.