Research On Methods And Applications Of Negative Publication Of Data

With the rapid development of information technology, more and more people focus on data privacy protection issues in the process of publishing data. At present, most of data publication algorithms directly expose users’the original sensitive attribute value when protecting privacy. The attacker can obtain the users’sensitive attribute value from the published data by some attack ways. Negative representation of information, using the self-nonself theory of biological immune system for reference, has a good effect on privacy protection. Currently, researches, applying negative representation of information in publishing data, proposed algorithms of publishing data called negative publication of data (NPD). Aiming at the defects existing NPD algorithms, in this thesis, we put forward two methods of negative publication of data methods named SvdNPD and (k,p,l)-NPD, and study the relevant application.Works in this thesis include the following three aspects:(1) Based on users’sensitive attribute value distribution, we propose a negative publication of data method named SvdNPD. Existing methods of negative publication of data named (k,m)-anonNPD and (l, m)-divNPD will cost large memory when there’s huge users, because the two methods improve m to reduce reconstruction error. Taken sensitive attribute value distribution into consideration, which is calculated by bayes1 theorem. Based on l-diversity model, during the process of negative representation, we will amend the negative selection probability to the probability which is calculated according to the user’s sensitive attribute value distribution. Then we will obtain the final published form. SvdNPD directly get the aggregation query answer from the published data without reconstruction. SvdNPD has a good result of query answer compared with (l, 1)-divNPD.(2) Based on the negative representation of information we propose negative publication of data method called (k, p,l)-NPD. When implementing the ^-anonymous model and l-diversity model, both (k, m)-anonNPD and (l, m)-divNPD used abstract anonymous technology which need to establish abstract tree for each quasi-identifier attribute. To get the requirment of anonymity, the attribute value will be abstracted to a fuzzy value according to the abstract tree. Therefore, it is not convenient to mining the association rules from the final published table. This thesis puts forward a NPD method named the (k, p,l)-NPD. It can satisfy the requirment of anonymity without having to establish abstract tree to abstract attribute values. (k, p, l)-NPD selects appropriate quasi-identifier attribute to be the negative representation attribute. Then it is hard for the attacker to detmine whether some user is in published table. At the same time, grouping the sensitive attribute value and negatively representing the sensitive attribute value in each group can realize the l-diversity model. So (k, p,l)-NPD prevents users’ identity information from attackers, at the same time, prevents sensitive information leakage, it is convenient to mine rules from the published data produced by (k, p,l)-NPD.(3) Based on the (k, p,l)-NPD table, we propose a association rule mining method NPD-AR. This paper designs a association rule mining method for (k, p, l)-NPD. The experimental comparison between the association rules mining from the published table and the processed table shows that NPD-AR has good effect on mining rules. Besides, increasing support threshold and confidence threshold can reduce the ratio of false positives and the ratio of false negatives.This work not only is reference in research on methods of data publication on the basis of privacy protection, but also is reference in mining assosiation rule from published data which is on the basis of privacy protection.