Convolutional Recurrent Neural Networks with Attention Mechanism for Streaming QoE Prediction

Abstract

Cloud performing arts businesses has been accelerated by the advent of the 5G era and the COVID-19 pandemic, so there is a growing demand for a quality of experience (QoE) predictive model. However, QoE is a time series factor with nonlinear relationship influence, including subjective and objective factors named Quality of Service(QoS), which leads to a high complex prediction. To solve this problem, existing studies have utilized Long Short-term Memory Networks (LSTM) and Convolutional Neural Networks (CNN) to effectively capture this kind of complex dependency, respectively, to obtain excellent QoE prediction accuracy. However, they can not take into account the accuracy and computational efficiency at the same time. So we proposes CGRU-QoE, that is, using CNN to extract global information, using the variant of LSTM--Gate Recurrent Unit (GRU) to extract context information, and then following the Attention Mechanism. In addition, we introduced a new input factor representing bitrate. The proposed method is mainly validated in the LFOVIA database and is superior to the baseline method in terms of prediction accuracy and computational complexity.