Learning-Based Quality of Experience Prediction for Selecting Web of Things Services in Public Spaces

In the age of the Web of Things (WoT), an increasing number of WoT devices will be deployed over public spaces and provide various services to users. Therefore, discovering and selecting public WoT services by predicting the expected Quality of Experience (QoE) become critical to satisfying users. However, because of the uncertain and dynamic nature of public WoT environments, accurately predicting and continuously maintaining the QoE of the services is challenging. We investigated the limitations of the traditional model-based QoE prediction in WoT environments and the potential of the learning-based approaches to deal with the challenges. In this work, we propose a distributed algorithm powered by learning-based QoE prediction for selecting public WoT services. Service agents predict the long-term QoE for the corresponding service based on attention mechanism and multi-agent reinforcement learning. Service agents learn the influence of hard-to-observe influencing factors in the environment, such as physical obstacles and interference from other services.