Machine Learning Based Popularity Regeneration in Caching-Enabled Wireless Networks

Abstract

Obtaining accurate content popularity in caching-enabled cellular networks can not only increase the caching profits in a large scale but also effectively improve quality of service (QoS). This paper investigates the content popularity based caching strategy optimization problem by maximizing the successful delivery probability under the premise of meeting the QoS. Based on the Dirichlet distribution, we developed a common interest model (CIM) by which the common interest properties of the mobile users (MUs) and the content popularity can be extracted from the content delivery history. In order to estimate the parameters of the CIM, a machine learning (ML) model is proposed by using the Gibbs sampling algorithm. Then, the content caching problem is transformed into a decision making problem which is solved by the branch and bound method. Numerical results demonstrate the effectiveness of the proposed scheme.