WebQMon.ai: Gateway-Based Web QoE Assessment Using Lightweight Neural Networks

Abstract

AbstractUsers’ perception of their experience accessing web pages greatly affects users’ willingness to continue browsing the website. However, it is difficult to assess user perception through a generic Quality of Experience (QoE) model. Web content consists of a large variety of static as well as dynamic objects, with some of them coming from the remote sites. This makes QoE assessment a challenge for the traditional methods. To build a generic QoE model, we introduce WebQMon.ai, a lightweight Web QoE assessment architecture using machine learning methods without setting any specific formula or threshold. WebQMon.ai can evaluate web-browsing QoE using mostly network-layer data with only one piece of application-layer information, the referer in the HTTP header, which is used to aggregate the packets associated with the same web page. The distribution of the arriving packets requested by the web page is used to construct WebQMon.ai. WebQMon.ai requires little storage space (80KB~6MB). More importantly it can be deployed directly at edge routers/gateways, due to the weak dependence on the application-layer payload. We further improved our algorithm by ensemble learning combining multiple orthogonal features, to generate a stronger classifier. We evaluated WebQMon.ai on three popular websites. It shows that the QoE assessment results for 4,800 unknown samples can be obtained within just 0.07 s and reach an average accuracy of 97%.KeywordsWeb-browsing QoENeural networks