Approaches For Privacy Preserving Based On Individual Correlation

With the development of various social networking and personalized recommendation services, personal information is often collected, managed and used by service providers, which causes the risk of personal information leakage. The existing privacy preserving methods face more severe challenges in front of attackers with different background knowledge. Therefore, the study and improvement of the state-of-the-art privacy protection methods is of great significance to adapt to the new attack scenarios.The present privacy preserving methods are not competent to handle known individual correlation attack scenarios. In this paper, starting with a commonly known individual correlation attack scenario, we design a privacy protection model that can resist this attack, namely r-anti-known-tuple-relation-attack privacy protection model. To get this model, we firstly not only extract the essence of the attack and abstract it into known tuple relation attack model but also model and bound the scope of the relationships among the individuals in the attack model. Then we explain what conditions should the groups satisfy to resist the attack model called r-anti-known-tuple-relation aggressivity, this property constrain that the intersection size between the candidate sets of sensitive attributes about the related tuples after anonymization should be at least threshold r. Finally we define the anti-known-tuple-relation-attack privacy protection model by exert different privacy constraints to different groups of anonymous table according to whether including associated tuples and prove the security of this model theoretically.Based on r-anti-known-tuple-relation-attack privacy protection model, we design the algorithm to publish anonymous datasets, which includes the ExtractRelation algorithm to extract the background knowledge and AnatomyAntiRelation algorithm to generate safe anonymous publishing table. The AnatomyAntiRelation algorithm includes group creation, group complementary and group partition. We also give the theoretical analysis of the algorithm's correctness, security, availability and cost. Experimental results show that the anonymous data generated by the algorithm satisfying r-anti-known-tuple-relation-attack privacy protection model proposed in this paper not only has similar usability compared with Anatomy algorithm satisfying l-diversity, but also has better security.