On Celebrity, Epidemiology and the Internet

The proliferation of the internet has created new opportunities to study the mechanisms behind the emergence and dynamic behaviour of online popularity and celebrity. In this paper we examine how common epidemic models, specifically SIR and SEIR models, can be applied to model the evolution of outbreaks of celebrity interest on the internet. A major challenge when using such models is to parameterise them to fit data as an outbreak unfolds over time, without knowing the initial number of susceptibles in the target population. We present a methodology capable of fitting the model's parameters from a single trace, while the outbreak unfolds, and of forecasting the epidemic's progression in the coming days. We present results on three kinds of data: simulated epidemic data, data from a real Influenza virus outbreak and data from music artists BitTorrent download and YouTube video views activity.