Online Learning Models for Content Popularity Prediction in Wireless Edge Caching

Abstract

In the geographical edge caching, where base stations (BSs) and users are distributed as Poisson point process (PPP) and the caching performance is measured using average success probability (ASP), we consider the content popularity (CP) prediction problem to maximize the ASP. Two online learning (OL) models are proposed based on weighted-follow-the-leader (FTL) and weighted-follow-the-regularized-leader (FoReL). Regret analysis concludes that OL methods results in sub-linear MSE regret and linear ASP regret. With MovieLens dataset, simulations verify that the FTL yields better MSE regret while FoReL has lower ASP regret. © 2019 IEEE.