Efficient content delivery through popularity forecasting on social media

Ubiquitous social networks have in recent years become significant for sharing of content generated in online video platforms. Our work investigates how the predictability of video sharing is associated with the underlying social network of the initial sharer of the video and the context of the media platform it was uploaded. In particular we combine user-centric data from Twitter with video-centric data from YouTube to give insights that neither dataset (social network and media service dataset) individually gives. We propose a simple model to predict future popularity of a video resource with a small and easily extracted feature set, based on the notion of influence score of a user and its fluctuation through time, as well as the distance of content interests among users of both datasets. We further demonstrate how the incorporation of our prediction model into a mechanism for content delivery results in considerable improvement of the user experience.