The Approximate Association Based On The Network Data Stream For The User Identity

With the rapid development of the network, the network data is increased by theexponential order; it contains wealth information of user, and how to extract the valuableinformation from massive data becomes big data research hotspot, which causes widespreadconcern in various industries. The paper chooses massive network data streams as the researchobject, through processing and analysis, gets the stable network behavior modes of the users’,makes the approximate association with their identities and group division. Firstly, the paperaims at the DDOS attacks on the application layer, proposes secondary cleaning datapreprocessing algorithm based on the principle/auxiliary mass function of the D-S evidencetheory; then in the process of data processing, proposes the user network fingerprint concept anddesigns a pyramid-shaped network fingerprint update framework, clustering the user networkfingerprint unscheduled, to form the stable network fingerprint of the user over time evolution, toachieve user’s associated identity and marked its uniqueness; due to the heat value of the useraccess network is an important part of the user network fingerprint,the paper adopts two types ofwebpage classification technology to take apart in the calculation of heat value for the useraccess WebPages, the database matching classification and the webpage classificationtechnology which is based on the Na ve Bayes algorithm,when the match is unsuccessful in thedatabase then adopts the second webpage classification algorithm, reduce the time complexity ofthe system; For network users’ group partition problem,the paper proposes the user networkfingerprint clustering algorithm based on hybrid distance, through clustering the networkfingerprint of all users on the system to obtain a plurality of user groups and the probability ofindividual user belonging to each user group, to achieve the approximation of the identity of theuser group division. Based on the above algorithm, we achieved an analog system, theperformance and accuracy of the algorithm has been verified through experiments.