Research On Enhanced P-Sensitive K-Anonymity Techniques In Data Publishing

In recent years,the rapid development of information technology and network accelerates the growth of data and makes it show an explosive trend.Due to the huge value contained in data,the demand for data sharing,information disclosure and data mining from all social circles is more and more urgent,and the data publishing process for these purposes is often accompanied by privacy leakage.Therefore,how to effectively protect privacy and minimize the damage to data utility is the focus of this paper.To our best knowledge,p-sensitive k-anonymity model and its existing extended models can effectively resist linking attacks and homogeneous attacks.However,they are not capable of resisting skew attacks and sensitive attacks,and even face the risk of semantic similarity attacks.In addition,the privacy requirements of different sensitive values are not always identical.If the "one size fits all" unified privacy protection level is used in the anonymous process,it will result in over protection of some information and insufficient protection of other information.To address these problems,this paper divides sensitive groups according to the sensitivity of sensitive values and measures the privacy requirements of sensitive values and sensitive groups quantitatively.Two enhanced p-sensitive k-anonymity models,namely(p,aisg)-sensitive k-anonymity model and(pi,aisg)-sensitive k-anonymity model,and the global search clustering algorithm,called CGS,are then proposed.Among them,the former model restricts the total frequency of sensitive groups differentially and realizes personalized protection for sensitive groups.On this basis,the latter model sets different diversity constraints according to the sensitivity of sensitive values to achieve the effect of personalized protection from the perspective of sensitive values and reduce unnecessary information loss caused by over protection of low sensitive information.However,the two enhanced models still have the threat of semantic similarity attacks.Based on these,this paper proposes another two enhanced p-sensitive k-anonymity models by semantic categories partition,namely(psc,aisg)-sensitive k-anonymity model and(pisc.aisg)-sensitive k-anonymity model,and a suitable local search clustering algorithm,called CLS.Similarly,the former model has the characteristic of personalized protection for sensitive groups,while the latter model adds personalized protection for semantic categories.Furthermore,the experimental results show that the proposed enhanced models and algorithms have outstanding advantages in better privacy at the expense of less data utility.