Unraveling the Dynamics of Stable and Curious Audiences in Web Systems

Rodrigo Alves, Antoine Ledent, Renato M. Assunção, Pedro Vaz-de-Melo, Marius Kloft

Keywords: Temporal Dynamics in Web Systems, Time series, Point process, EM- ALgorithm, Gibbs Sampler

TL;DR: We developed a model that is able to disentangle the slowly-varying regular activity of the stable from the curious audience activity occurring in bursts such as viral threads of web systems

Abstract: In this paper, we propose Burst-induced Poisson Process (BPoP), a parsimonious model to analyze time series, such as Twitter feeds or Youtube search queries. BPoP is able to disentangle the slowly-varying regular activity of the stable audience from the curious audience activity occurring in bursts such as viral threads. Our model is a mixture two hidden and interacting processes. The first component is a self-feeding-process (SFP), and the second is a stochastically driven Poisson process with a random step function as intensity function, whose transitions are caused by the bursty behavior of the first component. The SFP generates bursty behavior corresponding to viral threads caused by sudden external events, whereas the non-homogeneous Poisson process models normal background behavior that is influenced only by the overall popularity of the topic (the stable audience). We performed extensive empirical work that shows that our model fits and characterizes a large number of real datasets with better results than state-of-art models. More important, we show that BPoP is able to quantify the stable audience of media channels over time, which may serve as a good indicator for their popularity.

Track: Social Networks, Social Media, and Society

Submission Guidelines Scope: Yes

Submission Guidelines Blind: Yes

Submission Guidelines Format: Yes

Submission Guidelines Limit: Yes

Submission Guidelines Authorship: Yes

Student Author: No

Submission Number: 986