Research On User Behavior Modeling And Management In Intelligent Access Network

With the steady growth of the scale of Chinese netizens,their demands for network services are increasing,and the data of access network user behavior(ANUB)has also become massive and complicated.In order to extract useful knowledge from the massive data of ANUB with insufficient prior knowledge and noise,so as to analyze the massive ANUB information intelligently and elaborately,it is necessary to select the data mining technology that fits the application scenario and make a series of improvements.Clustering algorithms based on partition,such as K-means algorithm,can quickly process massive datasets with high dimension,so it is widely applied to the analysis of regional users' online behavior data with large quantities and rich attributes.However,this algorithm needs to preset the number of clusters,and it is not suitable for datasets with noise.We can find clusters of different sizes and shapes in density-based clustering algorithms,such as DBSCAN,OPTICS and SNN,and they are all robust to noise.However,most of these algorithms have the problem of selecting prior parameters.Density peak(DP)algorithm can detect any cluster without specifying the number of clusters.However,DP cannot directly identify all clusters in the dataset with clusters of greatly varying densities,especially for ANUB data where the user scale of different types varies greatly.It is worth mentioning that the density ratio estimation(DRE)method was not developed until 2016.DRE proposes a rescaling method called Rescale,and it has driven DBSCAN,OPTICS and SNN,well-practiced in finding clusters with different densities under certain conditions.However,these improved algorithms are still affected by prior parameters,so they cannot be applied in the analysis of ANUB.In order to analyze and model for ANUB,so as to provide the basis of traffic control for the network traffic control layer,by introducing DRE method,this thesis proposed a density ratio peak(DRP)algorithm in the research of ANUB.In order to make a quick analysis of the massive regional online behavior datasets,so as to develop a deep understanding of the network performance,and then continuously adjust the network structure and bandwidth,this thesis proposed an improved DRP-means algorithm in the study of regional online behavior.This algorithm integrates the DRP algorithm with low noise sensitivity and the K-means algorithm with fast computing big data.Based on DRPmeans algorithm,the Apriori association algorithm is used to mine the correlation and difference of regional online behaviors,so as to provide a basis for improving the quality of regional networks.Compared with the performance of traditional clustering algorithms,the DRP algorithm proposed in this thesis is more applicable to the analysis of users' online behavior.The DRP-means algorithm proposed in this thesis to preprocess data,can not only obtain the optimal partitions,but also improve the efficiency of the clustering process.It compensates for the defect of the unstable clustering results based on the inevitably artificial presupposition of cluster numbers in K-means algorithm.The DRPmeans algorithm can depict such massive high-dimensional datasets as regional online behaviors more accurately and meticulously.