M-3R: A Memory Based Approach for Streaming QoE Prediction under 3R settings

The rapid growth of video traffic and demand for high quality videos has increased in the recent years. The service providers emerge out to be successful only when they provide satisfactory end user viewing experience. Due to the channel throughput fluctuations, the end users in Hypertext Transfer Protocol (HTTP) streaming endure video quality variations with time (due to bitrate adaptations) and rebuffering events (once the received video data in the buffer is played out). Hence, it is necessary to evaluate the Quality of Experience (QoE) in video streaming scenario in a continuous time (per frame) manner in order to regulate the quality deteriorations. In this paper, we have proposed M-3R predictor- A network of Long-Short Term Memory (LSTM) that evaluate the time varying streaming QoE resulting from the effects of rate adaptation and rebuffering under the 3R settings. The "3R" settings is a combined effect of several Full Reference (FR), Reduced Reference (RR), and No Reference (NR) Video Quality Assessment (VQA) metrics. Our proposed M3R predictor was found to be effective in modelling the temporal dynamics of the streaming video QoE, that got reflected in the performance evaluation measures.