Global dynamics of online group conversations

Abstract

Public online groups allow individuals to carry out conversations of common interests. Study of such group conversations provides a unique opportunity to study patterns of human conversations without violating individual privacy. The observational studies conducted in this paper are an attempt to identify the main correlates of continued growth of conversations, thereby clearing the path to developing predictive models user participation. We study temporal evolution of online group discussions. Surprisingly, we find that individual discussion groups display distinctively q-exponential shaped inter-message times to reply distributions, unlike the power law distributions seen in email conversations. We show, using simulations, that the heavy-tailed distribution of time to reply, which we also observe when all data is combined, originate from mixtures of q-exponentials. We also find that popular threads come to be so from the very beginning as opposed to evolving to be more popular as they grow. This raises new possibilities for developing generative models of thread growth. Copyright © 2012, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.