Research Of Network Service-based Classification Technology And Platform Building

With the development and evolution of the network technology, the scale of the network and the number of users increased rapidly. The development direction of future network based on contents and differentiation customized service strategy requires classification analysis of user behavior. On one hand, we need to analyze the users’network behavior to study structure of future network. On the other hand, we need to partition the users based on their network behavior, so we can provide personalized service.This thesis designs a user classification platform based on their network behavior. It has the ability of high-speed data acquisition, mass data storage and distributed processing. This paper designs a high-speed data capture program, which can continuously capture high-speed data packets on network core router.In order to carry out adaptive clustering analysis of business behavior, this paper proposes a new cluster validity index. Compared with traditional clustering validity index, new index has two major improvements in the following aspects:The new index does not depend on fuzzy membership matrix. The new index analyzes geometric structure of class by mining neighboring objects, rather than analyzes fuzzy membership. It makes the new index apple not only to fuzzy clustering algorithm, but also hard clustering algorithm. And it expands the scope of application of the clustering index.The new index estimates a data object is at the edge of the class or not by analyzing its neighbors. According to the fuzzy degree of the edge, the new index evaluates how a class separated from other classes. Unlike traditional clustering index, the separation measure of new index independent of the distance between class centers. And this feature makes the new index has a better performance than the conventional index when the structure of the class is in various shapes.In order to verify the performance of the new index, simulation experiments have been executed on both artificial data sets and real-world data sets. Its accuracy, robustness has been proved by experimental results.