Early prediction of the future popularity of uploaded videos

Abstract Predicting the popularity of videos on video sharing sites is important for the formulation of online advertising strategies and commercial marketing. The predicted popularity value can help the system decide which videos to recommend to users and determine the recommended order of related videos. However, the weakness of most previous methods has been that the prediction of video popularity is based on past usage data after upload. Early predictions cannot be made because they are dependent on past access data. To solve this problem, we propose a new method for the early prediction of video popularity based on the data available at the initial time of video upload. This study first uses a supervised learning approach to develop a video popularity prediction model. Then, to further improve the overall accuracy of the prediction, an ensemble model is developed for integration of these classification results to get the most accurate predictions. Empirical evaluation shows that these models can effectively predict the future popularity of videos at the time of uploading.