Clustering Algorithms And Its Application For Campus Network User Behavior Analysis

Campus network plays an important role in the field of Internet. Its users’ behavior have specific characteristics. Although clustering algorithm is widely applied in the analysis of Campus network users’ behavior, it still has some limits in analyzing the conversation behavior data of campus network user. This paper will choose an appropriate clustering algorithm which is suitable for analysis of campus network users,improve it and apply it to analysis.① The study of clustering algorithm includes classical clustering algorithm and existing improved one. The author will make study of them and also compare the performance of different typical clustering algorithms.② The author proposes the improved K-medoids algorithm which is based on initial cluster center which is chosen from the farthest distance and local search strategy.Considering the shortage of K-medoids algorithm, which reliance on initial cluster center and slow convergence speed, the author will improve it from those two aspects and analyze the time complexity of improved algorithm.③ The author puts forward the improved SOM algorithm.Firstly, adding the dimension reduction before the input layer. And then proposes a stop condition which bases on the change of weight. Considering the SOM algorithm cannot be applied to the analysis of campus network users, and the setting of its stop condition will affect the convergent speed, the author will improve it from two aspects: the input layer model and train stopping, and analyze the complexity of improved time algorithm.④ The author preprocesses the log record which is gotten from Dr. com server-side of campus network. The preprocessing operation includes data cleaning, data extraction,data conversion and data storage.⑤ Finishing the simulation experiment and performance analyzing of the two improved algorithms.Analyzing the behavior of campus network users based on the improved algorithms,and concluding the behavior rule of campus network users.The author tests the application performance through algorithm’s effectiveness and stability.Analysis the cluster result of different dimension of data set,and getting the rule which can reflect the behavior of campus network user.Summarizing characteristic of the two improved algorithms.The study shows that time complexity of improved K-medoids which mentioned in the paper is 1/k2 of traditional one. Its effectiveness and stability acquire obvious improvement, and the stability rate has been raised by 11 percent. At the same time, the improved algorithms have better interpretability on the analysis of behavior of campus network users. Although SOM algorithm is not well applied in the time complexity,compared to K-medoids algorithm, its effectiveness and stability have achieved similar improvement. So does its interpretability. After the study, both algorithm can be used in the analysis of campus network users’ behavior.