Learning at the Edge: Smart Content Delivery in Real World Mobile Social Networks

The explosive growth of network traffic in recent years is largely attributed to the rising demand of content delivery in mobile networks. MSN has become the top mobile application, ranked by the number of daily active users and the daily time spent by users. This introduces new challenges for content delivery due to highly dynamic distribution of mobile users and their requests for content. It is important for content providers to understand how edge users' mobility and their consumed social content can affect the network traffic in the underlying content distribution network. To study this problem, we present a machine learning based framework for predicting the dynamics of content demands of MSN users at the network edge. It employs a smart content delivery scheme to dynamically deliver the content to users according to their future demands as predicted. We compare four regression models in predicting the number of requests to certain content. Then we evaluate the performance of the smart content delivery scheme over a realworld commercial content distribution network for WeChat.