Anomalous structure and dynamics in news diffusion among heterogeneous individuals.

Abstract

Previous research has suggested that well-connected nodes in a network (commonly referred to as hubs) are better at spreading information than those with fewer connections (ordinary users). Here we investigate the roles of nodes with different numbers of connections by studying how people share news online. Quantitative analysis shows that users without many connections can sometimes spread news more effectively than well-connected users when the diffusion pattern has dendrite-like paths that reach far into the network, leading to a non-Gaussian distance distribution. When the hubs dominate, however, the distribution is Gaussian. Enhanced interactions among ordinary users are the key to the emergence of non-Gaussian characteristics. Finally, we introduce a message-passing model that reproduces the observed diffusion features. This model shows that patterns dominated by either hubs or ordinary users can be clearly demarcated by measuring the average number of forwards.