Content Request Prediction with Temporal Trend for Proactive Caching

In this paper, we aim to improve the performance of proactive caching policies by presenting an accurate content request prediction algorithm. We develop a Bayesian dynamical model through which a latent temporal trend structure in the content request can be accurately tracked and predicted. The dynamical model also leverages tensor train decomposition to capture content-location interactions to further enhance the accuracy of predictions. We derive an approximation of the posterior distribution based on variational Bayes (VB) and Kalman smoother algorithms to infer the model’s parameters. Moreover, using a real-world dataset, we examine the impact of prediction accuracy of our proposed scheme on a designed cooperative caching policy. The numerical results show that our algorithm substantially outperforms reference methods which ignore the temporal trends and content-location interactions.