Analysis And Modeling Of User Access Behavior Of Video-on-demand

With the development of computer technology and network bandwidth, video on demand service is developing rapidly, which has become the main share of network bandwidth, and human beings have entered the era of video on demand. The users as the service object of video on demand service that their access behavior to the system greatly affects the performance and presentation of video on demand system. How to effectively research and analyze the user’s access behavior in video on demand plays an important role in the design and maintenance of the video on demand system and the design of the user’s behavior generator.Based on the research of the user access behavior in the existing video on demand, this paper thoroughly analyzes and researches the user access behavior in video on demand based on the user access data of Jiangsu Telecom IPTV video-on-demand system.Firstly, the main features of the user’s access behavior in video on demand are analyzed, including the user access traffic, the user request arrival process and the length of user play. We found the user access patterns showing the periodicity of one day, and the number of user requests is related to service nodes and the number of online users; the user request arrival process follows Poisson distribution at a finer time granularity; the user behavior in video on demand is mainly watching behavior.Secondly, aiming at the current study of the number of requests mainly research on the distribution of video popularity from the angle of video, we put forward to study the distribution of the number of requests per user from the perspective of user object. This paper make statistical analysis on the video popularity and the number of requests per user in the system., and found the distribution of video popularity and the number of requests per user presents a heavy tailed phenomenon.Finally, aiming at current modeling on heavy tailed phenomenon in the number of user requests mainly user extended exponential model, this paper proposed use the drift power law model to model. Modeling and analysis of video popularity, found that the drift power law model can better describe the distribution of video popularity in the short-term and extended exponential distribution can better describe the distribution of video popularity in the long-term; the shape parameter of the extended exponential model depicts the random fluctuation of the video frequency; the scale parameter of the extended exponential model and video popularity mean showed a certain linear relation. Modeling and analysis of the number of requests per user, found that the drift power law model can better describe the distribution of the number of requests per user; the shape parameter of the drift power law model depicts the random fluctuation of the video frequency; the scale parameter of the drift power law model depicts the size of mean of number of requests per user.The main innovation of this paper is to study the distribution of the number of requests per user from the user’s point of view, and propose use drift power law model to model the distribution of the number of requests per user. Then by comparing and analyzing, the results show that the drift power law distribution can well describe the distribution of the number of requests per user.