A Privacy Anonymous Method Based On Constraint

In recent years, the protection of privacy in data publishing had become a hot researchtopic in the field of information security. Data anonymization, as an important implementationtechnology of privacy protection, had attracted more and more attention from scholars. Howto cut down the loss of information that conduced by data generalization had become thefocus of current concern.L-diversity model is an important privacy protection model. It requires that the sensitiveattributes in the equivalence class are diversify enough. This model not only could preventlinking attack, but also be able to prevent the homogeneous attack. In addition, it can preventbackground knowledge attack very well. Firstly, the article summarized the currentimplementation techniques of l-diversity model, and analyzed of its realization method. Theprocess of traditional implementation method is relatively simple, but its degree ofinformation loss is high. In consideration of this disadvantage, on the base of researching onclustering algorithm, the article brought up one kind of constraint-based hierarchicalclustering algorithm to implement l-diversity model. Due to the constraints of the l-diversitymodel, the data set must to meet the constraints of the model to operate clustering algorithm.So, this method changed the implementation of model into clustering problem withconstraints. When dividing the equivalence classes, we could not determine the number ofequivalence classes at one time. So, we first divided the equivalence class using hierarchicalclustering method, then, used the method of exchanging the items in the equivalence classesto adjust the center of each cluster class and divided of equivalence classes again to makeitems in the equivalence classes more similar. At last, we deal with quasi-identifier attributesusing anonymous method, in order to reduce the overall loss of information in the data sets.At the same time, the article gives the data generalizability strategy and the reasonablemethod to quantify the loss of information.At last, we carried out the experiment, then got the experiment results, and thencompared the result of other clustering algorithms experiment with results of this method. Thedata of the experiment showed that the algorithm that we presented could protect data privacyinformation not being disclosed well, and can also reduce the loss of information in data anonymous better, to keep the availability of data.