Research And Impletation Of Network User Association Analysis Method Based On Traffic Flow

Analyzing user information and summarize user behavior from large-scale network traffic,which has attracted a lot of research and attention of industry is now a key technology in the Internet era.How to find potential relationships between users and convert it into useful knowledge of network security construction and Internet development is a pressing problem at this stage.With the in-depth research on the structure and communication mode of network protocol and data mining algorithm,this paper analyze which characteristics of users in each protocol can be extracted and use these feature data identify the network users from the complex flow,then puts forward a analysis model of the user association based on attribute similarity.Finally,a user analysis system based on net traffic is designed and implemented.The main research contents and achievements of this paper including the following aspects:(1)Aiming at the needs of structure for variety of mainstream network protocals and user association analysis,an analysis model designed in this paper achieves data mining and feature extraction of network traffic.Based on the structure of various mainstream protocols,the model clarifies the corresponding analytical method and automate the extraction of key fields.(2)In this paper,we extract the user account in traffic,and extract the corresponding account attribute characteristics for judging whether there is correlation between users.Besides,we propose four conversational features to enhance the association effect.The N-Gram,Jaro distance,and other different methods are proposed,which can measure feature similarity between different users.(3)Using SVM machine learning theory and extracted user data features,this paper proposed method using attribute similarity to judge user association.This paper derive features by calculating the similarity of different user attributes,which improved by feature segmentation and One-Hot coding.Based on this model,we propose the IP-SimRank algorithm to analyze the similarity between users,and then improve the effect of the association model.Besides,this paper designed a pruning strategy to optimize the computing performance of the model.(4)The user analysis system based on network traffic is designed and implemented and the components of each module of the system are designed in detail and coding implementation of the whole system is compeleted.(5)Finally,Building the system testing environment and test the performance of each model and the function of the user associassion analysis system.The system module includes the following parts:formulates a traffic analysis and mining module parsing strategy;implements the specific process of user identification;works out related models of SVM classification and association analysis;designes the table structure of the Mysql database to complete the data and results of storage.The experimental results shows that the system can solve the problem of low user associate accuracy rate and inability to associate different users.