Research On Personalized Anonymity Method Based On The Semantic Of Sensitive Attribute Values

With the coming of information age, with the rapid development of Internet, data storage and data sharing technologies, a lot of information are collected, stored and released by all kinds of organization. Data mining and data publication are two important concerns of database applications, data mining can collect a lot of information and extract underlying valuable knowledge, while data publication is to illustrate the data to user directly, with the process of collection and release of these data, the problem of the loss of privacy become prominent increasingly. Therefore, how to protect individual privacy and sensitive information has become a research focus in academia.Since the concept of anonymity was put forward for the first time from 1998, the research on anonymous principle and technology has been made deeply. To meet the needs of individual privacy autonomy, personalized anonymous privacy become a hot issue. How to prevent the loss of privacy and reduces information loss in the process of anonymity, is the research focus of personalized anonymity protection, at the same time, k-anonymity and 1-diversity can not resist the similarity attack. Therefore, based on the semantic similarity of sensitive attribute values, how to make a personalized anonymity method which can meet the needs of personalized preservation and can resist the similarity attack with the less information loss, is the purpose of this paper.This paper proposes a personalized (α,l)-anonymity model based on sensitivity.According to different sensitive attribute values, this model provides different sensitivity, and defines the concept of equ-sensitivity group, implements the needs of personalized privacy anonymity cause setting the frequency constraints for different equ-sensitivity group in the equivalence class. The model can also withstand similarity attack by limiting the total frequency of sensitive attribute values whose sensitivity are same in any equivalence class and controlling the distribution of sensitive attribute values.This paper proposes a personalized (α,l) anonymity algorithm based on clustering, inspired by the principle of 1-diversity, first forming a new set of clustered by the clustering of the original data, and then deal with each clustered by local-recoding to achieve the purpose of anonymity, while the grouping process based on the personalized constraints.This anonymity algorithm can achieve the personalized (a,l)-anonymity model based on sensitivity with less information loss.Finally the paper make experiments to prove the effectiveness and rationality of the model and the algorithm. Experimental results show that the method and the algorithm provide better personalized privacy preservation which can defense similarity attack more effectively than the 1-clustering anonymity and general 1-diversity models with the similar information loss and time cost.