Expecting to be HIP: Hawkes Intensity Processes for Social Media Popularity

Explaining and predicting the popularity of online multimedia content is an important problem for the practice of information dissemination and consumption. Recent work advances our understanding of popularity, but one important gap remains: to precisely quantify the relationship between the popularity of an online item and the external promotions it receives. This work supplies the missing link between exogenous inputs from public social media platforms (Twitter) and endogenous responses within video content platforms (Youtube), by studying a 6-months feed of 1.06 billion tweets about 81.9 million YouTube videos. This is done via a novel mathematical model, the Hawkes intensity process, which is able to explain the complex popularity history of each video according to its content, network, and sensitivity to promotion. This model supplies a prototypical description of videos, called an endo-exo map, which allows us to explain popularity as the joint effect of two intrinsic measures and the amount of discussion from the outside world. This model also allows us to forecast the effects of future promotions more accurately than approaches based on popularity history alone, and to identify videos that have a high potential to become viral, or those for which promotions will have hardly any effect.