Bayesian Poisson Factorization with SideInformation for User Interest Prediction inHierarchical Edge-Caching Systems

Edge-caching is an effective solution to cope withthe unprecedented data traffic growth by storing contents inthe vicinity of end-users. In this paper, we formulate a hier-archical caching policy where the end-users and cellular basestation (BS) are equipped with limited cache capacity with theobjective of minimizing the total data traffic load in the network.The caching policy is a nonlinear combinatorial programmingproblem and difficult to solve. To tackle the issue, we design aheuristic algorithm as an approximate solution which can besolved efficiently. Moreover, to proactively serve the users, itis of high importance to extract useful information from datarequests and predict user interest about contents. In practice,the data often containimplicit feedbackfrom users which isquite noisy and complicates the reliable prediction of userinterest. In this regard, we introduce a Bayesian Poisson matrixfactorization model which utilizes the available side informationabout contents to effectively filter out the noise in the data andprovide accurate prediction. Subsequently, we design an efficientMarkov chain Monte Carlo (MCMC) method to perform theposterior approximation. Finally, a real-world dataset is appliedto the proposed proactive caching-prediction scheme and ourresults show significant improvement over several commonly-used methods. For example, when the BS and the users havecaches with storage of25%and10%of the total contents sizerespectively, our approach yields around8%improvement withrespect to the state-of-the-art approach in terms of cachingperformance.