Understanding viewer engagement of video service in Wi-Fi network

详细信息    查看全文

• 作者：Yanjiao Chen ; ^a ; ^{chenyj.thu@gmail.com" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Qihong Chen^a ; ^{qchenag@ust.hk" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Fan Zhang^a ; ^{fzhangaf@ust.hk" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Qian Zhang^a ; ^{qianzh@ust.hk" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Kaishun Wu^b ; ^{wu@szu.edu.cn" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Ruochen Huang^c ; ^{ruochen90@qq.com" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Liang Zhou^c ; ^{13913936864@163.com" class="auth_mail" title="E-mail the corresponding author}Author Vitae
• 关键词：Viewer engagement ; Quality metrics ; Video content ; Viewer demography ; Personalized prediction model
• 刊名：Computer Networks
• 出版年：2015
• 出版时间：14 November 2015
• 年：2015
• 卷：91
• 期：Complete
• 页码：101-116
• 全文大小：1995 K

文摘

With the dramatic growth of online video services and video traffic, video service providers and network operators have keen interest in improving viewer engagement. Viewer engagement is mainly influenced by four aspects: service quality metrics (e.g., rebuffer time), network quality metrics (e.g., physical-layer data rate), video content (e.g., video length) and viewer demography. Previous works only partially consider some of these factors due to limitation of the dataset. In this paper, we develop an experimental platform with more than 50 self-deployed routers in our university campus, collecting information regarding all four aspects of engagement-related factors. Correlation and information gain analysis show that different viewer groups and video content types have different engagement patterns. Furthermore, we analyze each factor’s significance in determining viewer engagement. Finally, we propose to build personalized models to better predict viewer engagement, with bootstrapping customized models for new viewers.