Content Popularity Prediction via Deep Learning in Cache-Enabled Fog Radio Access Networks

Abstract

In this paper, the content popularity prediction problem in cache-enabled fog radio access networks (F-RANs) is investigated. In order to make the popularity prediction accurate and adaptive, we propose to learn the corresponding popularity prediction model for every content class from the preprocessed popularity series by training a simplified bidirectional long short-term memory (Bi-LSTM) network, and further use it to help build a content classifier in terms of content popularity trend in the training phase. Then, content popularity can be predicted by the right prediction model with respect to the corresponding content class in the predicting phase. Considering that it is hard to collect enough data about numerous content features through F-APs, we propose to only use the number of requests. Our proposed content popularity prediction policy offers a high prediction accuracy with low computational complexity by transferring the high complexity tasks from the predicting phase to the training phase. Simulation results show that the cache hit rate of our proposed policy approaches the optimal performance.