Human Behavior Dynamics in Online Social Media: A Time Sequential Perspective

We present an empirical analysis of human behavior dynamics in online social media from a time sequential perspective. By analyzing the users’ inter-event time distribution between two consecutive actions and fitting with accurate numerical method, we find that the users’ behavior pattern follows the heavy tailed or power-law distribution both in collective and individual scale, not the traditional Poisson processes hypothesis. With the estimating of the personal actions’ time sequential self-similarity, we find that the individual behavior can be predicted to some extent. We propose a more fine-grained method to eliminate the effect of individual non-regular behaviors on modeling in cascading non-homogeneous Poisson process model, and the simulated fitting results are more close to the actual human behaviors’ characters.