ESTABLISHED AT THE BEGINNING: VOD USERS' LOYALTY, LIFESPAN AND VISITING PATTERN

Watching 'on-demand' video on the Internet is popular. Providers are paying more attention on mining user behaviour data to improve service quality and attract more loyal users. It is desired to characterize evolution process of user engagement. While there are models characterizing engagement of traditional customers and web users, there still lacks an appropriate metric of user's loyalty for VoD systems. In this paper, we prove that users can be categorized into three groups based on the following three weekly visiting patterns: routine, random and rare visiting. We introduce a model from Web Data Analysis theory to present loyalty of a VoD user. It integrates the following four metrics for a user’s engagement: 1) recency, i.e. number of weeks the user does not visit the system since his last visit; 2) frequency, i.e. number of days the user visits the system per week; 3) number of movies that the user watches per week and 4) mean finish rate of the user’s all watching sessions per week. We find that finish rate is the most important feature to characterize user loyalty; and users of routine pattern visit more frequently and watch more videos with higher finish rate every week than others. Moreover, we find that a user’s visiting pattern, lifespan and engagement can be established from his initial watching behaviour of the first week after he joined the system. This suggests that a VoD system paying more efforts to provide newcomers who are in their first week arriving the system with better watching experience may foster more loyal users.