An activity window model for social interaction structure on Twitter

The development of online social environments has changed the manner of social interaction and communication, which are driven by individual human actions. Thus temporal variations in interaction networks are deeply impacted by the temporal dimension of human activity. In this paper, we address this issue through a detailed analysis on the retweets and comments of 550,000 Twitter users. We propose a temporal network model to represent the interaction network on Twitter, in which each node contains an activity window and the emergence of the edges between nodes are dependent on it. Specifically, the activity window is defined as the backtracking length from the message flow posted by the user's friend, which represents the user's social ability. It complies with a power-law distribution with an exponential cut-off. The interaction network is sparser and more clustered than the followee-follower network, in which the interaction stability and burstiness fluctuate with the activity window or with the degree to which the two users are involved in the communication. Finally, the effect of activity window on the aggregating degrees of the interaction network is examined.