Research On Privacy Protection Based On K-anonymity

Database Technology is one of the most important research fields in information technology over the the past decade. Data stream and uncertain data have been the hot area of research for several years. Data mining makes all sorts of data which concerns individuals privacy leak at any time, data security has become an important research subject, and how to protect the privacy of enterprises and individuals was focused by researchers during the data publishing and using. K-anonymity is one of the most important technology of privacy protection, and can prevent privacy information leak efficiently. Much attention has been focused on k-anonymity.Aiming at the goal of improving the utility of anonymous data, and the anonymous data was of more practical value, a series of research work has been done for the k-anonymity, and the main contributions of this paper may include:In order to prevent the various inference attack and pull down the chances of success of inference attack, and to avoid privacy disclosed, we propose improved strategies of preventing privacy inference attacks based on value range, value relation, anonymity rules, anonymity relation etc. on k-anonymity data set:the diversity of sensitive attribute, less than predefine threshold; combining various forms of anonymity rules, anonymity relation; pulling down inference probability mean and so on.Analyse sensitive attribute deeply, summarize its characteristics, propose the concept of sensitive degree, and build sensitive degree weighted matrix, relation sensitive degree matrix, uncertain relation matrix based on background knowledge, then, we use them in k-anonymity model. We propose CBK(L,K)-anonymity model and algorithm, multidimensional CBK(L,K)-anonymity algorithm for statical data set based on defense strategies of privacy inference attacks mentioned above, and achieve annoymization by sensitive degree weighted matrix and clustering. Experimental results show that CBK(L,K)-anonymity algorithm is effective and efficient, and proves CBK(L,K)-anonymity algorithm can make anonymous data effectively resist background knowledge attack and sample attack, and can solve diversity of sensitive attribute. Especially the practicality of anonymous data is improved.We propose an improved RSLK-anonymity model and algorithm for dynamic stream data set, in order to solve the problem of privacy information leakage in streaming data publishing, the main idea is anonymizing the streaming data set based on specialization tree and relation and sensitive matrix of background knowledge. Experimental results show that RSLK-anonymity algorithm is utility, effective and efficient. It can make anonymous streaming data effectively resist background knowledge attack and homogeneity attack, and can solve diversity of sensitive attribute of dynamic stream data.We propose UDKattr(tuple)-anonymity model and algorithm, UDAK-anonymity model and algorithm for uncertain data, in order to solve the problem of privacy information leakage in relational uncertain data publishing, UDKattr (tuple)-anonymity was achieved by specialization tree and uncertain relation matrix of background knowledge, UDAK-anonymity was achieved by clustering and anatomy. Experimental results show that UDKattr (tuple)-anonymity and UDAK-anonymity algorithms are utility, effective and efficient. They can make anonymous uncertainty data effectively resist background knowledge attack and homogeneity attack, can solve diversity of sensitive attribute, at the same time, they can keep the original uncertainty of the uncertain data.