Characterizing viral videos: Methodology and applications

• A framework for modeling on-line video views.
• A definition of “viral videos” based upon relative function performance and fitted function parameters.
• Analysis of video views with respect to video categories and time scale granularities.
• A means of analyzing on-line video channels.
• Applications to marketing decision support systems.

It is estimated that the online video advertising market will be worth $8 billion by 2016, up from $4 billion by 2013. Consequently, there is a need for metrics to help analyze and understand this rapidly growing market. This paper helps fill this need by describing a framework for modeling online video behavior, which is based on growth curve modeling. Different categories of video behavior are defined, including “delayed viral” behavior and “initial viral” behavior. The framework described in this paper can be used to analyze online video behavior, categorize videos based upon growth patterns, and predict future views. In associated empirical work, video views are analyzed for four different datasets. The first is an empirical video set, compiled from media lists of viral videos. The second is a “population” sample of videos, selected randomly from YouTube. The third is a dataset measured at different levels of time scale granularity. The forth is a set of videos released for specific YouTube channels. Managerial uses for the framework are described and specific scenarios are given for both content design and revenue prediction for online advertising.