Analysis Of Network Users' Behavior Based On Clustering

With the computer network spreading and developing, the number of network user is increasing, and the users' behavior is more and more complex.One hand,the method is needed to quickly detect and effectively control the destroy,the cheat and the incursion; the other hand, to well serve for the users, it is needed to grasp the mode of users'network behavior. Therefore, the studying for network users'behavior is more and more important. In this paper, some researches as follows for this are completed:The detection for users' abnormal behavior is achieved. First, users' behavior is needed to be modeled. Different from the existing research, in this paper, it is seperately modeled by source IP statistical attributes and dest IP statistical attributes. Then, K-Medoids in Data Mining is chosen because of its fast speed. Some work is done for improvement basing on the cluster numbers. For successfully picking out the abnormal behaviors from the large number of the network behaviors, a method for standard is designed. In a word,the method in this paper is faster and more effective than the existing research.The analysing method of network users' business havior is achieved. First, users'business behavior is modeled by counting the percent of all kinds of business flow. Then, For mastering the business using characteristic, the gradation clustering is chosen. For improving the classical gradation clustering method, the knowledge about entropy is used, and multiple similar single point clusters are one-time merged. The experiment shows that the improved method in this paper is greatly faster than the existing researches.Last, a system frame for network uses' behavior analysing is designed. There are some modules such as network data collecting,behavior extracting,behavior analysing,data storing,management and control platform.