On the “familiar stranger” phenomenon in a large-scale VoD system

Video-on-Demand (VoD) services are having more and more impact on people's daily life. The huge number of view records generated during the services reveal lots of information about video popularity and user behavior. In this work, we seek to understand user behavior from a network approach. Users who overlapped view records are defined as “familiar strangers” of each other. With network science concepts, dynamic familiar stranger networks are established based on view records from one of China's largest VoD service providers. Analysis shows that users can be separated into light users and heavy users. Community detection results show that there is a threshold of familiarity that drastically deteriorates the community structure of the familiar stranger network. Further investigation on the evolution of the communities suggests that there exists considerate number of dynamic communities with low members turnover. Our findings about the interest-based familiar stranger network unveils interesting networked user behavior patterns and provides meaningful guidelines for Internet applications like online user communities and recommendation systems.