Learning Popularity for Proactive Caching in Cellular Networks

Abstract

Video data have been showed to dominate a significant portion of mobile data traffic and have a strong influence on a backhaul congestion issue in cellular networks. To tackle the problem, proactive caching is considered as a prominent candidate in terms of cost efficiency. In this chapter, we study a novel popularity-predicting-based caching procedure that takes raw video data as input to determine an optimal cache placement policy, which deals with both published and unpublished videos. For dealing with unpublished videos whose statistical information is unknown, features from the video content are extracted and condensed into a high-dimensional vector. This type of vector is then mapped to a lower-dimensional space. This process not only alleviates the computational burden but also creates a new vector that is more meaningful and comprehensive. At this stage, different types of prediction models can be trained to anticipate the popularity, for which information from published videos is used as training data.