Going Multi-viral: Synthedemic Modelling of Internet-based Spreading Phenomena

Marily Nika, Thomas Wilding, Dieter Fiems, Koen De Turck, William J. Knottenbelt

Introduction

Standard epidemiological models, such as the Susceptible-Infected-Recovered (SIR) model, were originally designed to track single biological outbreaks. However, modern internet-based phenomena—ranging from viral videos to trending hashtags—rarely follow a single, smooth trajectory. Instead, they often exhibit "multimodality," characterized by multiple peaks of activity triggered by distinct real-world events like celebrity endorsements or media appearances.

Traditional "mono-epidemic" models struggle to capture these secondary pulses, resulting in poor fits and unreliable forecasts. To address this, researchers have introduced synthedemic modeling, a framework that treats complex, multi-peaked trends as a synthesis of multiple simple epidemic outbreaks.

The Synthedemic Framework

Inspired by Fourier analysis, which decomposes complex sound waves into simple sine waves, the synthedemic approach decomposes a complex spreading phenomenon into a sum of simple sub-epidemics.

The Synthedemic Modeling Framework architecture, illustrating the decomposition of a composed signal into multiple compartmental epidemiological models.

The framework utilizes two primary basis models:

  • SIR Models: Used to represent organic growth and spreading through communities.
  • IR (Infected-Recovered) Models: Used to represent sudden "spikes" of interest, such as mass-media events, followed by exponential decay.

The system employs an online fitting algorithm that processes data points as they arrive. It maintains a parsimonious model, adding new sub-epidemics only when the current model's fit (measured by $r^2$) drops below a target threshold. This is achieved through residual analysis, where sudden deviations from the current model trigger the instantiation of a new sub-epidemic.

Experimental Results

The framework was evaluated across three distinct datasets: synthetic data, the 2009 H1N1 outbreak in England, and BitTorrent download counts for artist Robin Thicke.

Biological and Technological Data

The model demonstrated remarkable versatility. In the case of the 2009 H1N1 (Swine Flu) outbreak, the synthedemic approach accurately captured the two-wave structure of the infection. As shown below, the model adapts in real-time as new data arrives, improving its fit as the second wave emerges.

Online synthedemic decomposition of the 2009 H1N1 outbreak in England. The left panel shows the initial fit to the first wave, while the right panel demonstrates the model's adaptation as the second wave emerges, reaching a final $r^2$ of 0.952.

For the Robin Thicke BitTorrent data, which is highly complex due to multiple promotional events, a standard mono-epidemic model achieved an $r^2$ of only 0.485. In contrast, the synthedemic model identified four distinct sub-epidemics—corresponding to events like a performance on Jimmy Kimmel Live and the 2013 MTV VMAs—achieving an $r^2$ of approximately 0.90.

Discussion and Future Directions

The "synthedemic" paradigm allows for high complexity while maintaining model simplicity. Its online nature enables real-time adaptability, making it a powerful tool for near-term forecasting of viral trends.

Future work will focus on increasing the rigor of predictive analysis and exploring "negative" terms to model events that suppress spreading, such as censorship or counter-campaigns. Additionally, the authors aim to account for interdependencies between waves, where one outbreak might exhaust the pool of susceptible individuals for subsequent ones. This method holds potential for broader applications in computer virus propagation and retail sales forecasting.

Summary updated

Abstract

Epidemics of a biological and technological nature pervade modern life. For centuries, scientific research focused on bio-logical epidemics, with simple compartmental epidemiologi-cal models emerging as the dominant explanatory paradigm. Yet there has been limited translation of this effort to ex-plain internet-based spreading phenomena. Indeed, single-epidemic models are inadequate to explain the multimodal nature of complex phenomena. In this paper we propose a novel paradigm for modelling internet-based spreading phe-nomena based on the composition of multiple compartmental epidemiological models. Our approach is inspired by Fourier analysis, but rather than trigonometric wave forms, our com-ponents are compartmental epidemiological models. We show results on simulated multiple epidemic data, swine flu data and BitTorrent downloads of a popular music artist. Our technique can characterise these multimodal data sets util-ising a parsimonous number of subepidemic models.
BibTeX
@article{nika2015going,
  author = {Marily Nika and Thomas Wilding and Dieter Fiems and Koen De Turck and William J. Knottenbelt},
  title = {{Going Multi-viral: Synthedemic Modelling of Internet-based Spreading Phenomena}},
  journal = {EAI Endorsed Trans. Ambient Syst.},
  year = {2015},
  doi = {10.4108/icst.valuetools.2014.258221},
}
Figure