Monitoring web QoE based on analysis of client-side measures and user behavior
Nawres Abdelwahed,Asma Ben Letaifa,Sadok El Asmi
DOI: https://doi.org/10.1007/s11042-022-13427-5
IF: 2.577
2022-08-06
Multimedia Tools and Applications
Abstract:An increasing number of web applications in all fields for different types of use incites developers and researchers to enhance the web applications as well as the network conditions to satisfy the final user. That's why the estimation of the Quality of Experience for web applications (web QoE) remains necessary. The web QoE gives service and content providers an idea about the perceived quality by users as it helps them to determine issues for improvement. Waiting time influences user satisfaction. For this reason, we found many studies in web QoE interested in the first part of the browsing operation, such as the waiting time until the first page is completely loaded. However, these measures do not include the interactions of the user with the web application lately. We measured the interactions of the user with the web application via the user engagement metrics. This new field of research has attracted many researchers these last years due to its efficacity in determining the satisfaction level of the user. The contribution of our study is the research and use of user engagement metrics for QoE prediction. We have noticed that user engagement metrics (generally used to evaluate a user's commitment in social media) are more precise in expressing the QoE of the user. For this reason, we used user engagement metrics to predict user web QoE that is the novelty of our work. In this study, we elaborated our dataset, which contains the three types of measurements; Quality of Service (QoS), QoE, and user engagement metrics. The obtained dataset reflects the user experience from several perspectives (the network quality, the loading process, and the interaction of the user with the web application). This reflection makes our dataset an exhaustive one. After collecting the data, we visualized our different metrics. Besides, we tried to predict the Mean Opinion Score (MOS) with Machine Learning (ML) algorithms, but we obtained low accuracy due to the small number of lines in our dataset. Finally, we tried to profile the users using the K-means clustering algorithm. In this clustering, we recuperated three user information metrics (age, gender, and study level).
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering