Quality of Experience Inference and Prediction for Real-Time Communication

Quality of Experience refers to the degree of delight or annoyance perceived by the user of an application or service. The purpose of this thesis is to leverage QoS measurements collected in the network to model and predict the QoE in the context of real-time communication applications. This case study focuses on video QoE and uses Cisco WebEx as real-time multiparty conference system. Experiments are conducted in different scenarios, using both active and passive measurements, and the QoS information extracted from the traffic captures are fed to machine learning algorithms to infer the QoS/QoE relationship and make QoE predictions starting from QoS features. While the majority of previous work focuses on QoE extraction at the application and at the client (often mobile), leaving the underlying network with no/limited visibility about QoE behavior, this thesis considers multiple monitoring points. This is beneficial since an intermediate network node could then use the QoE information learned to optimize its behavior and drive traffic management decisions. The obtained results give a better understanding of the QoS/QoE relationship and show that the modeling conducted further from the client can lead to reduced yet interesting knowledge.