WebQMon.ai: Threshold-Oblivious On-Line Web QoE Assessment Using LSTM Neural Networks
Enge Song, Tian Gong Pan, Chenhao Jia, Tao Huang, Yunjie Liu · 2019
The user perception of visiting a web page greatly affects the user willingness to continue surfing the website. However, it is difficult to construct a general Quality of Experience (QoE) model via traditional methods (e.g. deducing a formula or setting thresholds) because the contents of different websites are in endless varieties. To handle such an issue, we introduce WebQMon, a web QoE evaluation method using machine learning without deducing any formula or threshold, which is easy to deploy for different websites. WebQMon can evaluate the user experience through little application-layer data and most network-layer data, which makes it possible to deploy WebQMon on the gateway. Specifically, at the application layer, we use the referer field in the header to aggregate the packets generated by visiting a website, and in the network layer, we use the size and arrival time of packets to construct the WebQMon. In addition, we improved WebQMon by reversing the input data and evaluated two methods on three popular websites. The QoE prediction accuracy of the improved method can reach 99% for 16,000 samples.