Video on demand (VoD) service in IPTV is a bandwidth-hungry application. It has been argued that the distribution of popularity of videos can be well measured using a Zipf-like distribution in which top 10% of the videos account for nearly 90% of requests. In this article, we propose a neural network based method to predict the popularity of videos in an IPTV system. The popularity prediction can be used by service providers for video placement in content delivery systems or hierarchical servers and hence it can lead to bandwidth save. Simulation-based performance evaluation of our proposed method confirms a significant accuracy in the prediction of the popularity of the videos.
