Title:Temporal scaling in information propagations

Abstract:For the study of information propagations, one fundamental problem is uncovering universal laws governing the process of information propagations. This problem, in microscopic perspective, is formulated as estimating the propagation probability that one piece of information is propagated by an individual to another. A propagation probability depends on three major factors: the intrinsic attractiveness of information, an aging factor characterizing the freshness of information, and the interactions between individuals. Despite that the temporal effects of the former two factors are widely studied, few studies consider the temporal laws underlying the interactions between individuals, causing inaccurate estimations of propagation probability and inaccurate prediction of information propagation. In this paper, we empirically study the effect of individuals' interactions on information propagations on the largest Chinese social media. We find a temporal scaling in information propagations: for two individuals, the propagation probability decays with the length of time interval since the latest interaction between them, following a power-law manner. Based on this temporal scaling, we design a temporal model to estimate propagation probabilities among individuals, significantly reducing the prediction error of propagations from 6.7% to 2.6% and improving viral marketing with 9.7% incremental customers.