Behavior Of Network Users Research Based On Time-Varying & Services

Today, with the advancement in information technology and the development of the conomic, internet has gotten a more and more important position in oue life.The rapid expansion of Internet, network traffic, network user and the number of host computers increases with the exponential growth. With the development of mobile devices, embedded systems and sensor networks, the new Internet element, Internet-scale will continue to grow in a long time. Internet applications have developed from simple traditional applications to real-time multimedia applications. Nowadays Internet applications have the new characteristics that sharing the resource and collaboration. However, the important reason for Internet reaching today's position is that the rapid development of Internet services, network application, diversified types of business and personal trends.However, the rapid booming Internet business also bring some troubles to telecommunication companies:if we do not know the mode of network service and its rules with time, we can not make better fees, targeted marketing strategies and monitoring the network; If we do not know the model of user online and offline behavior we can not get a reasonable server's load balancing, to give server the optimize performance.This paper studies the Internet user behavior. Using the actual backbone traffic data to analyze in order to grasping the prefer mode for the network user service, analyzing the rule of the model changing over time, and modeling the user online and offline behavior. In this case, the carriers can make the directional products marketing, according to the characteristics of customers; make the fees according to the user characteristics, providing the valuable reference to distinguish valuable clients and server load balancing, etc.1) This paper selects hierarchical clustering algorithms, which based upon the real provincial backbone network data and our purpose of analyzing the prefer mode for the network user service. And because of the defects of the hierarchical clustering algorithm, this paper introduces the improving the clustering algorithm to reduce the time complexity, the data results show that ,comparing our improved algorithm with the classical hierarchical clustering, our improved algorithm improve the efficiency of time has greatly increased by about 10 times,. Even comparing with the improved hierarchical clustering algorithm based on the minimum spanning tree, our algorithm is also faster than it about 3 times.2) According to the results of fast hierarchical clustering,this paper reveals the composition of the prefer mode for the network user service that based on the different time scales and the size distribution of every mode of network service.And this paper deeply analyzes the using frequency difference between the different modes of the network service, and therelationship between the modes of the network service and the users daily online duration, the daily flow of network users and the flow ratio of the up-flow and down-flow. This paper also analyzes and explains the forming reasons of the characteristics between distribution and relationship.3) This paper not only analyzes mode of the network service,using the improved hierarchical clustering algorithm, but also is the first paper that analyzes the time changing combining with the mode of the network service,studies the rules that the mode of the network service changes with time scales.By defining a series of indicators and processing the actual data, this paper reveals the relationship between the mode of the network service and the time scales, the critical point improvement,at the mean time, analyzes and explains the characteristics of the changing relationships . Then it summarizes and shows the changing sequence of the mode of the network service, which appears the most in a month's time series.4) This paper is the first to using the non-homogeneous Poisson process to model and analyze the network user online and offline behavior. This paper uses the method of hypothesis test to authenticate users online and offline meets non-homogeneous Poisson process, using the actual data. Then it uses the non-homogeneous Poisson process to model the user online and offline behavior, and we derived the probability formula of the user online and offline, based on the dependence assumptions. Finally, we make the theory verification and data validation whether the formula is right or wrong. The verify results confirm the conclusion reasonable. In addition, we also give the user online and offline log probability distribution picture based on the different mode of network service. It is a basis of the further research.