Research On Privacy Preserving Methods For Multiple Sensitive Attributes Based On Attributes Classification

Massive data of practical application was produced in each organization with therapid development of information technology, especially the Internet and databasetechnology. The collection and sharing of these data will help improve the quality ofservice, to promote scientific research; however, use of these data containing personalinformation, such as data mining, will increase the risk of leakage of personal privacyinformation. Therefore, how to publish data about individuals without revealingsensitive information has become a widespread problem.Privacy protection is to ensure that adversaries can’t infer sensitive informationwith high probability. In the process of data publishing, in order to prevent disclosingindividuals’ sensitive information, published data is usually anonymous data aftertreatment. On the other hand, the ultimate aim of data publishing is to perform dataanalyses and researches, so, the publisher should guarantee the availability ofanonymous data. So the focus is to balance the conflict between the privacy protectionand utility of anonymized data in privacy-preserving data publishing.In practical applications, the data to be released often contain multiple sensitiveattributes, but the existing privacy principles are not suitable for multi-sensitive attributedata publishing. The privacy protection method towards multiple sensitive attributesprevents adversaries from inferring individuals’ sensitive information accurately bychanging the correlation among sensitive attributes. Currently, most privacy protectionmethods towards multiple sensitive attributes extend k-anonymity or l-diversity model,but there are still some defects, such as privacy information protection inhigh-dimensional data and high information hiding rate, these problems will be moreprominent especially when the diversity of each sensitive attribute values are quitedifferent.In view of these problems and the practical application, a sensitive attributesclassification based privacy preserving model was proposed. The core idea is to classifythe sensitive attributes according to the diversity and importance of each sensitiveattribute, and then set different value of l for them, and group according to somestrategies so as to satisfy l-diversity. Meanwhile, two algorithms were proposed toimplement the model, they use different strategies to achieve data sets grouped and meetprivacy requirements of the model. Experimental results show that this approach can protect privacy of data and reduce the hide ratio and enforce the usability of data.