Learning-Based QoE Prediction and Optimization for Video Streaming

It is always the top priority for network service providers to provide better Quality of Experience (QoE) for users around the world. Huge efforts have been devoted to Adaptive Bit Rate (ABR) streaming, but it is only a passive QoE amelioration method by adapting to bandwidth variations. In this paper, we design a Hierarchical Navigable Small World (HNSW) query algorithm and a classify method to help the content provider improve the users' QoE under the amount of users' playback logs with different network-context features. To make a better prediction for users' QoE and QoE metrics, we also designed a Long Short-term Memory (LSTM) prediction algorithm which can help us make a more optimal changeable features selection by an accuracy prediction capacity. To evaluate the performances, the extensive network streaming experiments are carried out over a controllable network test-bed and real Internet trace. The results demonstrate that the proposed learning-based QoE prediction and optimization video streaming algorithm is able to improve the average video bitrate up to 12%, decrease the startup delay up to 28%, reduce the stall ratio up to 29%, and increase the QoE up to 17% compared to the default scheme, which is the widely used scheme for the most video content providers.