The Research And Analysis Of User Group Behaviors In Campus Network

Because of different interests and habits, users of the Internet differ from one another in terms of information focus, time consumed and services selected. Thus different user groups will inevitably take on behavioral characteristics of their own. Knowing about the so-called group behaviors could gain, in a macro sense, an informed knowledge of the state of network use. On the one hand, this can provide some guidance for Internet users, enabling them to make better use of network resources; on the other hand, this can also serve as theoretical basis for network administrator to carry out monitoring and similar works. Campus network users, considered as one part of the Internet users, are supposed to be more professional and with strong purpose. Thus they may bear more obvious characteristics. This thesis focuses on this problem to find out the group behaviors to supply important theoretic foundation for network management, such as anomaly detection. As a result, the analysis of group behaviors on campus network is even more important.This thesis brings forward a method of analyzing user group behaviors from the viewpoint of campus network users. As the using of Internet is affected by a good many factors which changes sharply, it's difficult to utilize an effective mechanism to summarize it. So this thesis does some research on the analytic method of group behaviors. First, the flow data of campus border net are collected and the data source are got by flow aggregation; then by pretreatment to data and data selection, the attribute sets which are of the highest correlation and can describe the user group behaviors are acquired. Afterwards, the thesis puts the groups into different categories and marks them, and then establishes behavioral characteristics of user groups from both horizontal and vertical perspectives according to time consumed, service selected, and visit current capacity. Finally the cluster analysis of network traffic flows using a new clustering algorithm is present. Then the relatively comprehensive behavioral characteristics of user groups are obtained.The thesis conducts an in-depth research and discussion into to the application of the cluster analysis to the address attributes. The popular clustering algorithms, such as k-means and DBSCAN, not taking into account the feature attributes of IP addresses, make the results incomplete and can't achieve maximize difference between clusters. A new algorithm, which can effectively improve IP addresses clustering, is proposed in respond to those limitations. The advantages are as follows: Firstly, the initial clusters are got by the longest prefix algorithm and adapted version of the nearest neighbor clustering algorithm. Then the thought of stepwise-optimal hierarchical clustering is applied to the mergence of the nearest groups of initial clusters. The similarity between initial clusters is determined by the longest prefix of IP addresses contained in these clusters. Finally, the algorithm automatically and meaningfully yields clusters that are in accordance with the characteristics of IP addresses on traffic flows. The researcher could obtain the user behavioral trend and custom to the website.This research and analysis on the behavioral characteristics of the user groups of campus network can provide considerable theoretical basis for the recognition of abnormal behaviors, the monitoring and anomaly detection as well as better design of campus network.