Enhancing QoE for Multi-Device Video Delivery: A Novel Dataset and Model Perspective

Multi-device video streaming applications enable seamless playback across various devices, including large-screen TVs, tablets, and smartphones, revolutionizing digital content consumption and enhancing user experience. However, ensuring consistently high quality of experience (QoE) across these heterogeneous devices remains a substantial challenge due to intrinsic differences in screen sizes and viewing conditions. In this paper, we first build an open-source, multi-device, and time-continuous QoE dataset named
$MCQoE$
by conducting a large-scale subjective experiment to analyze QoE variations among different screen-size devices. Then, we thoroughly investigate the dataset and observe that video quality and rebuffering impact on TVs is more significant than on other devices, such as middle-size PC monitors and small-screen smartphones, emphasizing the importance of building specific QoE models for different devices. Furthermore, we propose a novel low-complexity but effective QoE model denoted as
$LiteDC$
, integrating a temporal dilated convolution network with a targeted pruning technique to align with the computational constraints of embedded platforms. Extensive results show that compared to a state-of-the-art baseline algorithm,
$LiteDC$
achieves a remarkable 20.9-fold improvement in execution speed while increasing prediction accuracy by 6.4%. The
$MCQoE$
dataset is available for download at https://github.com/yanghaocuc/mcqoe.