Research On Privacy Perserving Data Publication With Multiple-dimentional Relevance Of Sensitive Attribute

In the open Internet Era, the data/micro data related to personal information increases exponentially on the Internet. The data sharing and publishing can be used for Massive data analysis, with the rapid development and wide application of data mining technology, data publication has great utilization in scientific research, social investigation and public opinion monitoring and so on. However, at the same time micro data publication also brought the privacy disclosure problem, in today’s information Era, the concept of privacy has changed a lot, and society and the public pay more and more attention to the need of privacy protection, when we keep the information availability, how to protect the privacy of user is the main problem in the data publication research. In the current research of privacy preserving data publishing,most of research mainly consider the condition of single sensitive attribute dataset, and the reseach of multiple sensitive attributes is usually based on the simple extension of existing single sensitive data publication method, which results in large amount of information loss of the published data table in practical application, and the utility value of data table dclines. Expecially the exist research about multiple sensitive attribute have never considered multidimensional association description between different sensitive attributes,and adversary can use this to attack publishing table. To solve the problem that the relevance of multiple sensitive attributes may generate privacy disclosure, this paper reseach multiple sensitive attributes data publication from the theory of multidimensional sensitive association privacy discosure、privacy sensitive data release model for Multidimensional Association and privacy protection method., we not only try to reduce information loss effectively, but also consider the situation that adversary use the correlation between different attributes to attack and provide the corresponding protection. The main work of this paper will be introduced as followed:①Make a in-depth research and analysis to two existing classic data publication k-anonymous and1-diversity. and we also point out the shortcomings and deficiencies when single sensitive attribtue data publication method are directly used in multiple sensitive attribtues will bring. For example, if we use the method used in1-diversity to dispose multiple sensitive attributes directly, equivalence group will become larger and larger with the growing of number of sensitive attributes, which results in more and more information loss. So the basic idea of this paper’s pirvacy protection is provided. ②Make a analysis to the privacy relevance of multiple sensitive attribute data, give the relational description of multiple sensitive attributes in data, provide a privacy attack mode which use association rules, and a new privacy protecion model with multidimensional privacy relevance is provided accordingly.③According to the privacy discosure mode, this paper provides two new attack models which increase the attacker’s background knowledge, in order to takes association between different sensitive attributes into account adequately the two models introduced the association rules into multiple sensitive attributes and they can avoid that adversary use association rule to attack user’s privacy, furthermore they can reduce information loss effectively with generalizing the sensitive attributes. The two models is different in strength and emphasis of protection, which provides more choice for Data publishing of different situation.④Provide the two models’ own algorithm, respectively. Test and verify the model with real data set Adult. Through the experimental results, the model provided by this paper not only can solve the problem that adversary use association rule to attack privacy of user, but can keep the quality of published data table high..