The Anatomy of Social Media Popularity project studied popularity as a state of being liked or supported by many people. The approaches primarily stem from the computer science and mathematics disciplines. Examples include the development of the Hawkes Intensity Process (HIP) which factored three intuitions: magnitude of user influence, content quality, and decaying social memory to develop a predictive model. This model was then refined into Recurrent Neural Network models to forecast (verified with actual data) or predict (infer from past data only) popularity from heterogeneous streams to quantify average response to unit promotion and relative influence amongst users of varying fame levels. These models identified four sets of additional metrics to quantify the likelihood of an online event going viral, requiring the development of an algorithm to estimate user influence. The RNNMAS model outperformed YouTube popularity prediction system by 17 and also captured seasonal trends of unseen influence, though performance was sensitive to content type (e.g. super users in Gaming videos, cohorts of regular users in Activism videos).The experimental portion of this work also created five new video datasets. Code developed under this project is openly hosted on GitHub. The contributions of this work advances the science of understanding social influences in the cyber domain.
展开▼
