DescriptionOnline streaming is one of the most popular services available over the Internet. Today video is increasingly consumed over wireless networks with their far higher packet loss rate compared with wired networks. It generates video impairments that degrade user experience.
We studied video impairments caused by packet losses with a realistic setting. We found that viewers prefer high-resolution videos with some impairments to smooth low-resolution videos. It disagrees with the HTTP adaptive streaming protocol, which sacrifices resolution for smoothness. Additionally, viewers ignore some short impairments and feel that block-artifacts impairment occurring after a freeze is acceptable. We are the first to reveal that impairment occurrence order and inter-impairment interval length influence user experience differently. These findings show the feasibility of improving user experience by reordering and balancing impairment occurrences.
In addition, we conducted experiments on observing users streaming video viewing behaviors under packet loss wireless networks. We have observed nine types of behaviors including system level behaviors and video player level behaviors. We noticed that low video quality and perceivable video impairments are key factors to motivate users take actions and users usually choose behaviors they believe can improve viewing experience. Also, the sequence of behaviors user has taken follows some particular orders. According to these observations, we propose a novel user video watching behavior prediction model that achieves 94.7% accuracy. Moreover, we created cognitive models that explain how video quality variations and human memories play roles in users’ behavior decision making. Findings show the feasibility of improving user experience by reordering and balancing impairment occurrences.
Our study results and the user behavior prediction model provide promising ideas and tools for enhancing user experience under crowded networks where video impairments are unavoidable and optimizing network resource and user management via human engineering.