LSTM for Mobility Based Content Popularity Prediction in Wireless Caching Networks

Abstract

Caching has attracted a wide range of research interests due to its ability to reduce traffic load and latency. However, reasonable caching strategies are required to further improve caching efficiency and system performance. However, how to predict the content popularity evolution has become a major issue in the design of caching strategies. Moreover, user locations is a non-negligible factor since it is often coupled with content popularity in the practical scenarios, e.g., content popularity may vary along with user's location. Therefore, in this paper, a caching scheme is proposed based on a novel prediction model which jointly considers mobility and content popularity. In specific, Long Short-Term Memory (LSTM) method is utilized as a prediction tool due to its advantage of processing long sequences. Experimental results demonstrate the effectiveness of our proposed scheme with higher prediction accuracy and improved caching efficiency.