Research On Privacy Protection Methods For Resisting Similarity Attacks With Sensitive Attributes

With the rapid development of information age,data mining technology is widely used in human life.Data mining can be explained as to obtain the implicit,unknown,and potentially useful data information from micro data at first,then publish and share the data information which provide great convenience for scientific research among various organizations.At the same time,as the people's attention to personal privacy increases,resource sharing and data distribution should not only guarantee the availability and real-time performance of data,but also should prevent individuals from being exposed to the threat of privacy leak.Through the research and analysis of many existing privacy protection anonymous models,it is found that most of the models ignore the semantic similarity between sensitive attribute values that can be susceptible to similarity attacks leading to privacy leakage and safety threats.Therefore,this article do some research to solve the problem of existing models.The main work is as follows:(1)We propose a (p,k,d)-anonymity model that resists similarity attacks for sensitive attributes.The current Sensitivep-k-anonymous model has no regard for the semantic similarity of sensitive attribute and can be susceptible to similarity attacks.In order to solve this problem,a (p,k,d)-anonymity model that can resist sensitive attribute similarity attacks is proposed.The model requires that each equivalence class in the data anonymous table satisfies k-anonymity,constrains the semantic dissimilarity of sensitive values by d-different,and usespto control the number of sensitive values that satisfy d-different in each equivalence class.Finally,each equivalence class has several sensitive values with a large difference in semantics,which reduces the risk of privacy leakage.Considering the availability of data,the model divides the equivalence class by means of the distance-based measurement methods to reduce the loss of information.(2)A (l,m,?)- anonymity model for multiple sensitive attributes similarity attacks is proposed,where m is the dimension of sensitive attributes.Most of the existing anonymous models are oriented to single sensitive attribute and cannot be directly applied to multiple sensitive attributes.However,the published data often contains multiple sensitive attributes,and multiple sensitive attribute similarity attack problemsalso exist in large numbers.Therefore,on the basis of the previous work,an (l,m,?)- anonymity model is proposed to solve this problem.This model can flexibly set the d-dissimilarity of each dimension-sensitive attribute,and then calculate comprehensive dissimilarity e for the sensitive attribute of m-dimension.So that each equivalence class in the publication table has at least l sensitive attribute values that satisfy comprehensive e-different in sensitive attributes to resist similarity attacks.In order to realize this algorithm,the KACA clustering method is used to generate the equivalence classes and the data availability is improved.We study the privacy protection anonymous technology and do some improvements for the existing anonymous models.Experiments demonstrate that models which we proposed can effectively resist similarity attacks,and has a great improvement in privacy protection and data availability.