Context-Aware Prediction of QoS and QoE Properties for Web Services

Web Services are commonly used for integrating applications between partners over the Internet. Since services with the same functionality are advertised with different Quality of Service (QoS) levels and are assessed with different Quality of Experience (QoE), choosing the right service may be quite challenging. It is essential for a user to predict QoS and QoE values as accurately as possible in order to find a suitable service. Usually collaborative filtering is applied using similar users and services for predictive purposes. We hypothesize a correlation between context data and QoS and QoE dimensions which can be additionally incorporated to improve predictive accuracy and scalability. In this paper we present the two algorithms PredReg and PredNet in order to predict QoS and QoE values for Web Services. The PredReg algorithm is based on multiple linear regression. The PredNet algorithm uses additionally a neural network for prediction. Both algorithms include context data of users and services generating personalized predictions for the requesting user. In addition, PredNet is able to process categorical variables so that user profiles can also be considered for predictions. We evaluated PredReg and PredNet and compared them with the state-of-the-art approach WSRec [1] which is a memory-based collaborative filtering approach. Our experiments demonstrated that PredReg and PredNet provide a higher predictive accuracy and a significantly improved scalability. Therefore, we recommend the application of PredReg and PredNet for future personalized predictions.