Video Popularity Prediction Through Fusing Early Viewership with Video Content.

In this paper, an approach for video popularity prediction is proposed, by combining two different methods, video content analysis, and early viewership. Firstly, a multi-modal framework is used, which analyses video content features such as the number of presented people, poses, emotions of presenters and attendants, and an audio events detection algorithm to detect claps, pauses, and speech. Afterwards, all these features feed a linear regression model that predicts each video’s views. In parallel is utilized, a second method that uses interaction metrics such as early viewership to predict popularity. Viewership is crawled through an online tool developed based on Google Analytics. In order to go one step further, a fusion methodology of the views is proposed coming from the video content, with early viewership, while comparisons between models are investigated. The experimental results were based on real-life data, demonstrating that the proposed system is a promising tool for the prediction of the video popularity, having a Mean Absolute Error of less than 7.5.