User Preference and Activity Aware Content Sharing in Wireless D2D Caching Networks

Abstract

Device-to-Device (D2D) content sharing has emerged as an important tool to alleviate the backhaul pressure. Most of prior works optimize D2D caching policies with known content popularity, which may not be the case in reality. In this paper, we investigate a D2D caching optimization problem with unknown content popularity in wireless D2D caching networks. To maximize the overall D2D caching hit rate, we propose a distributed caching policy by learning user preferences and user activity levels. For the first time, we exploit the sliding time window method to predict real-time user activity levels. And we employ a logistic regression model to describe the user preference. By predicting user activity levels and user preferences in real time, the proposed policy not only can significantly improve the overall D2D caching hit rate, but also reduce the traffic load of the base station compared to existing policies. Simulation results with MovieLens dataset further show that the overall D2D caching hit rate of our proposed policy is close to that of the optimal caching policy.