|
With the rapid development of computer networks, data mining and other informationtechnology, the importance of data sharing based on scientific research, commercialapplications and knowledge discovery gradually emerged. However, the shared data includesindividual sensitive information such as medical records, etc. Therefore, personal privacy is inthe risk of leakage, and how to better protect privacy will become the concern of the experts,scholars and information owners. As privacy preservation commonly used methods,k-anonymity and l-diversity are simple and practical, t-closeness model is more effective inprotecting privacy leakage, so it has been widely used and studied.This paper is based on the previous analysis and studies of t-closeness and enhancedk-anonymity, finding there are two shortcomings of t-closeness: it has no specific algorithmsand its semantic privacy can’t be defined as demand. The paper proposed (t,a)-closeness, anenhanced privacy protection method:(1) Considering a measure of semantic privacy a whosevalue is calculated according to the classification of sensitive degree, classification can becustomized, flexible and convenient;(2) Because it has no specific algorithms, t-closeness arevery open and wide research, the paper proposed two algorithms. Firstly, because top-downmethods may cause no distortion information, adapting top-down methods to find anonymoussolution to meet the (t,a)-closeness. Secondly, produce less loss of information in the clusterprocess, genetic algorithm in accordance with the principle of survival of the fittest andsurvival of the fittest, generation by generation evolution to produce increasingly betterapproximate solution, after coding, crossover, and mutation, to produce the best approximatesolution, which is in line with the process to produce (t,a)-closeness published table aftergeneralization and the suppression of the gradual, and so use genetic algorithm-basedclustering approach with less loss of information.Experiments show that the proposed (t,a)-closeness method is effective and feasible, to acertain extent, enhances the ability to resist similarity attack, but the quality of the data on thelack of protection efforts, there is still room for further strengthening.
|