Research On Network User Classification Based On Multi-Dimensional Characteristics

With the development of network technology, network application and network scope, the Internet has already become an essential part of people's daily life in almost every aspect. The Internet is more an extension of real world than a tool of acquiring and exchanging information. Network users can not only read news, search information, listen to music, watch videos and download files by using the applications, but also can buy things of all kinds, publish opinions, play online games or even act as different roles in virtual online communities. Network users interact with the network with network applications, and at the same time show different behaviors in interaction with the information and other users. The study of the network user behavior is very important for network planning and tuning, network service design and improvement.Network measurement evolved as a set of methods and technology with development of the Internet. Network measurement generally consists of network traffic measurement, network performance measurement and network topology measurement. Network traffic measurement, also called network traffic monitoring, keeps playing a very important role in network planning and improvement as basic way to evaluate network status. With the development of the broadband network, the network operators intend to provide better and diversified network services to a larger population of users instead of only providing basic network access and transport service. In such an environment, network operators are keen to learn the principles of user behaviors in network interaction. Thus, the study of network user behavior calls more and more attention.In this paper, we applied the network traffic monitoring method to user behavior data collecting, and analyzed the behavior data with data analysis and data mining method to classify network users according to their behavior. Traditional traffic monitoring method is faced some challenges for the purpose of user behavior data collecting, such as dynamic user identification, network application identification, massive data collection, high link speed and traffic volume and huge user number. We evaluated the different data collecting methods and considered that a dedicated user behavior data collecting system with DPI (Deep Packet Inspection) function is necessary. During the research, all the data were collected from a typical metropolitan network managed by one of largest broadband network operators in China, with our hardware based DPI data collecting system. We collected and analyzed most of the broadband dial-up users'behavior data and the result reflected the current situation of broadband users'behavior.1) Principles of network traffic monitoring and network user behavior analysis. In this paper, we summarized the network traffic monitoring and network behavior analysis method and the key technology, such as DPI based network application identification. Since the broadband dial-up user has already been the main part of broadband users, we focused on the analysis of broadband dial-up user behavior. We raised a multi-dimensional behavior model based on user online activity, application behavior and Web behavior. With the improvement of behavior data collecting technology, we can collect and analyze the behavior in several dimensions. We in this paper focused on user behavior identification and classification based on multi-dimensional characteristics, which is a key subject in user network behavior analysis.2) Broadband dial-up user behavior analysis. In this section, we first analyzed the general principle of user activity data, including online times, duration, traffic sent and received, and time of the day. Then we presented a general approach for identification and profiling user online activity behavior model by using K-means clustering method based on the data above. We identified several main behavior groups and further summarized behavior profile of each user group as main user behavior models. The results showed that our approach could properly identify behavior models and classify users by their behaviors.3) Broadband user application behavior analysis. The development of network applications accelerated the interaction between users and networks as well as among users each other. Application usage behavior is an important aspect of user behavior. In this section, an algorithm combining entropy concept and clustering method is proposed to analyze current broadband users'usage modes of network services. By analyzing the broadband user application usage data with the entropy based method, we indentified the main application usage patterns. What is more, we analyzed the user group distribution, probability and transition probability of the application usage pattern. At last, we discussed the application usage model of single users.4) Broadband user web behavior analysis. Web has been the killing application of Internet for a long time. Currently, the Web sites are so many and diversified that they provide us the basic and effective way to acquire and exchange information. Web application is the most popular and widely used service in the Internet, so the web visiting behavior is also a key aspect of network user behavior. In this part, we compared the different web behavior data collecting method, analyzed the data collecting with traffic monitoring method. Then based on the data come from the metropolitan area network broadband users, we summarized the user interest data from web sites classification, and indentified the user web interest patterns.5) At last section, we discussed the relationship of three behavior dimensions as user online activity, application behavior and Web behavior, and put forward some future research points.In short, we focused on analysis of user behavior data in different dimensions, and classification of network users by identification of their behavior patterns in multi-dimensions, which solved the practical problem of network user behavior analysis. The results verified the validity of the method and gave us a deeper understanding of broadband user behaviors.