The Research About The Preprocessing And The Clustering Methods Of Behavior Information On Campus Network Users

With the rapid development and popularity of Internet,the web applications have come to human life deeply.As a crucial member of Internet,the campus network plays an important role in our college education.However,we can clearly see the fact that on the one hand,the quantity of campus network users is increasingly growing;on the other hand,some problems on web applications and internet regulations still exist.Based on the fact that the crowd of campus network users will bring about amounts of data on network activities,we could analysis log file of user behavior on campus network by using the method of data mining,which is well applicable to college education.This article dedicates to dig the network behavior data in log file to obtain network behavior rules to support administrators to set out a series of effective strategies to guide users conducting network activities reasonably.Currently,the analysis method of user behavior information on campus network is usually inclined to the algorithm design of clustering algorithms,instead of less consideration on connection to attributes of user behavior on campus network.And the clustering algorithms are mostly limited to application and improvement in the traditional algorithms during the analysis of user behavior on campus network.This article proposes a preprocessing method effectively bonding data features of network users,and on this basis,this article introduces the method of subspace clustering upon the graph theory,combining with the PSO clustering algorithms upon the linear inertia weight,which applied completely to the preprocessing data,to conclude valid network behavior pattern on campus network users.The specific research contents follow:The analysis method of user behavior information on campus network includes the preprocessing and clustering process in statistical and clustering method.And it gives an explanation of the common clustering algorithm and function.According to data features of user behavior information on campus network,the new preprocessing method which combines with current preprocessing method is advised and applied to access the user behavior data on campus network.As a result,it can be concluded that the fusion analysis for different attributes is achieved,replacing the current sole attribute analysis method.Account for the data features of user behavior information on campus network,the subspace clustering method upon graph theory is introduced during the clustering process,contributing to the solution to mainly object focusing on vertex partition based on undirected weighted graph rather the common clustering method.The characteristic matrix extracted from distance matrix after preprocessing original research data could accomplish dimension reduction of high dimensional data on campus network and achieve the behavior information analysis on campus network users by introducing the clustering algorithm upon the linear inertia weight.The innovations of this article are included:It proposes a new preprocessing and normalized weighting method of user behavior information on network campus,incompatible with the analysis of different attributes data.In the analysis of user behavior information,the proposing clustering algorithm on the connection between the subspace conception and the PSO clustering algorithm makes the research data effectively dimensional reduction,to some extent,It avoids the problem that the utilization of clustering algorithm as a rule may result in the local optimal solution,contributing to the clustering process of the study.The results show that the preprocessing method mentioned above could accomplish the combination of different attributes among the research data and the fusion analysis of objects in campus network.Furthermore,this article introduces a new method,the clustering method based on subspace PSO,Compared with the available method about behavior information of campus network users,the PSO method greatly lower the time complexity of algorithm,avoid the problem of sensibility on center point selection and enhance the ability of dealing with high-dimensional data.It achieves a more truly clustering result and the result’s stability,which proposes a new way researching the behavior information of campus network users.