Research On Personalized K-anonymity Model

The development of Data mining technology not only provides a new technical support for machine learning and knowledge discovery, but also threatens people's privacy correspondingly. The researchers found that the privacy information will be leaked by linking various public data sources, although the data owners always take some measures to hide the identifiable information before they share or publish the source data. In order to avoid the link attack mentioned above, Sweeney proposed k-anonymity model, which has been experimentally proved that it is an effective solution to prevent the identity leakage in data publishing and sharing, but lack of appropriate protection mechanisms for sensitive property information. In order to enhance the effectiveness of privacy preserving, researchers put forward many improved k-anonymity models and methods. Although the models have been to some extent improved the privacy protection, they also have many deficiencies in protectinng individual sensitive privacy information. After analyzed and researched the current existing algorithms and models of k-anonymity, we find that all the methods were basically not taking the fact into account that the same sensitive attribute may be have different privacy degree for different individuals have different opinions and requirements. To solve the problem mentioned above we propose a new model of k-anonymity privacy preserving based on personal privacy decision degree. The main research content of this paper are as follows:Firstly, in the view of the different privacy-preserving decision degree based on personalized granularity of privacy preserving, we propose a new k-anonymity model of privacy preserving, which is also called ( g ,α)k- anonymity. We apply the new model on the dataset with one sensitive attribute. Experiments show that the model can effectively improve the precision of privacy protection, and avoid the leakage of the data with higher demand for privacy protection, and over protection of data with lower demand.Secondly, considering the dataset with multi-sensitive attributes, we combine the k-anonymity model of privacy preserving with granular theory. According to the different individual privacy-preserving decision degree, we cluster the whole dataset to different granular spaces. The corresponding definitions of the individual privacy-preserving decision degree and the algorithms of granularity anonymity are also given. Experimental results show that the algorithms can effectively reduce the information loss caused by data anonymous, and improve the privacy protection accuracy for the data with multi-sensitive attributes.Thirdly, based on the idea of cluster analysis, the generalization anonymous algorithm of the ( g ,α)k- anonymity is given in this paper. According to the principle that the distance between two tuples in the same granular space should be as small as possible, and in the different granular spaces should be as large as possible, a variety of distance formula, information loss formula and the generalization algorithm have been redefined and the new algorithm and its time complexity have been analyzed too. The correction of the experimental results verified that the ( g ,α)k- anonymity model and the algorithm must be an effective tool for individual privacy protection.