QoE Estimation Method with Time-series Features Extracted from Packet Flows for Video Streaming

Recently, video streaming services have generated a considerable amount of network traffic. The quality of experience (QoE) plays a crucial role in evaluating the condition of video streaming services; it depends on metrics such as encoding bitrate, resolution, and stalling. To improve the users’ experience while video streaming, understanding the users’ QoE in real time and maintaining a healthy network is essential. Herein, we proposed a method for estimating video metrics based on the features of packet flows. We also proposed a method for estimating the users’ QoE as a mean opinion score (MOS) using the estimated video metrics as inputs. To evaluate the effectiveness of the proposed methods, we constructed an experimental environment that can extract the video metrics and features of the packet flows during video streaming by HTTP Live Streaming and MPEG-DASH. We generated an extensive dataset of hundreds of hours of video streaming with various network quality. Through effectiveness evaluation using these datasets, we established that the proposed method could estimate video metrics more accurately than the conventional method. Moreover, we observed that the coefficient of determination of the proposed MOS estimation method is >0.89.