Predictability of information spreading on online social networks

Abstract

With the rapid development of the mobile Internet, online social networks are playing an increasingly vital role in the dissemination of information. Accurately predicting the size of information cascades in advance has become a crucial issue, particularly in the realms of viral marketing, risk management and resource allocation. There are numerous studies that have tackled this prediction task, but the outcomes are unsatisfactory. In this paper, we explore the predictability of information cascade size through the lens of percolation theory. Our investigation reveals that the accuracy of cascade size prediction is notably diminished in the proximity of the threshold, evident in both artificial and empirical networks. Moreover, we observe a degradation and an user-level difference in prediction performance as social media platforms undergo evolution. Our findings underscore the necessity for additional factors to enhance prediction accuracy.