Scalable Detection of Viral Memes from Diffusion Patterns

Abstract

Social media and social networking platforms have flourished with the rapid development of mobile technology and the ubiquitous use of the Internet. As a result, memes, or pieces of information spreading from person to person, can be reshared among users quickly and gain huge popularity. As viral memes have tremendous social and economic impact, detecting these viral memes at their early stages of spread is a worthy, yet challenging problem. Here we review the literature on predicting viral memes, and present empirical results from Twitter and Tumblr datasets. We demonstrate how diffusion patterns of memes, in the context of network communities, play an important role in predicting virality. We show that it is feasible to obtain predictive features based on community structure even at the massive scales that common social media services need to process. Our results may not only enable practitioners to make predictions about meme diffusion, but also help researchers understand how and why different factors, in particular diffusion patterns in communities, affect online virality.