Prediction of Content Success and Cloud-Resource Management in Internet-of-Media-Things Environments

Abstract

In Internet-of-Media-Things (IoMT) environments, users can access and view high-quality Over-the-Top (OTT) media services anytime and anywhere. As the number of OTT platform users has increased, the original content offered by such OTT platforms has become very popular, further increasing the number of users. Therefore, effective resource-management technology is an essential aspect for reducing service-operation costs by minimizing unused resources while securing the resources necessary to provide media services in a timely manner when the user's resource-demand rates change rapidly. However, previous studies have investigated efficient cloud-resource allocation without considering the number of users after the release of popular content. This paper proposes a technology for predicting and allocating cloud resources in the form of a Long-Short-Term-Memory (LSTM)-based reinforcement-learning method that provides information for OTT service providers about whether users are willing to watch popular content using the Korean Bidirectional Encoder Representation from Transformer (KoBERT). Results of simulating the proposed technology verified that efficient resource allocation can be achieved by maintaining service quality while reducing cloud-resource waste depending on whether content popularity is disclosed.