The title of this paper reminds me of many other papers. However, I soon realized that it is one of the most interesting pieces I’ve read in a long time--and I mean compared with any text, not just scientific literature. The paper is so relevant that, once the language becomes more widespread, it could easily appear in a popular newspaper.
Borghol et al. describe how to model the popularity of viral videos. They make four contributions to the field. First, they discuss impartial sampling techniques for gathering data. Collecting samples from a social networking site is often not a straightforward process because services such as YouTube often restrict how, including how many, videos can be retrieved.
Next, the authors investigate how popularity and media churn evolve for a concrete set of recently uploaded videos over a period of eight months. They note that the relative popularity of a video is often highly nonstationary. All of their observations are discussed in detail.
Their third contribution is what they call a “three-phase characterization of popularity evolution for our sample of recently uploaded videos.” The idea here is to explain the evolution of relative popularity and “the differences in how long it takes video popularity to peak.”
Finally, the authors actually present a model that aims at capturing “the popularity evolution of newly uploaded videos.” They describe their model:
Using only a small number of distributions based on the three-phase characterization, our model is able to generate synthetic datasets in which key characteristics and consequences of the video popularity dynamics match those observed in the empirical data.
Borghol et al. discuss both their theoretical and mathematical models in great detail, providing visual results along the way. The appendices present empirical results to verify the models.
This paper is a must-read for anybody interested in social media. Again, it’s one of the most interesting reads I’ve come across in a long time.