Online Learning of Caching and Recommendation Policies in a Multi-BS Cellular Networks

Mobile edge computing is a key technology for the future wireless networks and hence, the efficiency of a cache-placement algorithm is important to seek the cache content which satisfies the maximum user demands. Since recommendations personalizes an individual's choices, it is responsible for a significant percentage of user requests, and hence recommendation can be utilized to maximize the overall cache hit rate. Hence, in this work, joint optimization of both recommendation and caching is proposed. The influence of recommendation on the popularity of a file is modelled using a conditional probability distribution. To this end, the concept of probability matrix is introduced and a Bayesian based model, specifically Dirichlet distribution is used to predict and estimate the content request probability and hence the average cache hit is derived. Joint recommendation and caching algorithm is presented to maximize the average cache hits. Subsequently, theoretical guarantees are provided on the performance of the algorithm. Also, a heterogeneous network consisting of M small base stations and one macro base station is also presented. Finally, simulation results confirm the efficiency of the proposed algorithms in terms of average cache hit rate, delay and throughput.