Research On Personalized Privacy Protection Method For Clustering Mining

With the rapid development of research on big data,data as a resource for the first time received a high degree of attention by each branch of social entities such as government,enterprises and academia.People have begun to benefit from the data which has an important application value.However,people may consciously or unconsciously reveal personal privacy information when they use a variety of digital services.Privacy security problem obviously stands out in nowadays society.In the process of fully exploring the great value of the data,how to protect personal privacy,especially how to avoid the privacy disclosure of data mining,is the key issue to be solved.Privacy protection problem in the process of data mining is gradually stepping into the sight of people.Researchers have proposed some beneficial results on privacy protection.However,these gained achievements failed to fully consider personalized demand of privacy protection in the process of data mining.Unlike general methods,personalized privacy protection could be more targeted.Hiding technology research aiming at meeting personalized demands,has become an urgent problem to be solve in the academic area.However,there are few researchers involved in the field of personal privacy protection technology for clustering mining.How to reduce the risk of privacy leak,caused by data mining from the personalized protection data,remains to be explored by academia.In order to solve the problem of personalized demands of privacy protection in the process of clustering,this thesis researches a kind of personal privacy protection algorithm for clustering mining.Firstly,in view of the personalization features of privacy,the concept of privacy degree was defined and its coding was given.Then a privacy data model,which is described jointly by the original data and privacy degree,was established based on the privacy degree.Secondly,for the sake of expressing the sensitivity differences of privacy data,a privacy relation was defined and a partial-order set was built accordingly.Afterwards,a topological classification algorithm oriented on privacy data was proposed to resolve the privacy partial-order relation.Thirdly,in view of data's multi-view features,a multi-view clustering method was established based on several views consist of original data,privacy degree,tuple sensitive degree and privacy partial-order set and so on.Fourthly,through the variable k-anonymity strategy,a multi-view privacy preserving algorithm that can meet personalized demands was proposed to apply different degree of individual protection on tuples with different sensitive degrees in different classers or within a same classer.Finally,the information loss and efficiency of the proposed algorithm was investigated on real data set.The proposed method has the following characteristics.On the one hand,this method can fully respect users' privacy protection aspiration,and can be able to reflect the differences of privacy cognition and privacy protection demand for different users.On the other hand,this method can provide different levals of protection for different sensitive data,which has the characteristics of great pertinence,lower information loss and high data availability.