增强型身份保持的隐私保护方法研究

发布时间：2018-06-12 02:22

本文选题：数据发布 + 隐私保护　；参考：《广西师范大学》2017年硕士论文

【摘要】：随着信息时代的到来,越来越多的数据被人们利用起来。如网络购物产生的数据,可以用于推荐系统;患者的医疗数据,可以用于对医疗并发症和疾病之间关联性的研究等。但是这些数据通常包含个人的隐私信息,如果直接发布的话,将会导致个体的隐私泄露。同时,简单的删除这些数据中个体的身份信息并不能达到保护个人隐私的目的,攻击者可以通过数据间的关联性来推断或分析出来个体的敏感信息。因此,数据发布中的隐私保护问题越来越受研究者重视。关于数据发布的隐私保护方法,最早的保护方法为k-匿名模型,后面在它的基础上提出了一些扩展模型。这些保护方法虽然能保护用户的隐私,但是它们往往会将个体标识属性删除掉,忽略了个体间的关联性。童云海等人提出了身份保持的隐私保护方法,但是他所提出的匿名模型没有充分考虑单一个体对应多条记录间的联系,仍然造成隐私泄露。因此,本文进一步研究身份保持的隐私保护方法,主要研究工作如下:首先,分析现有的身份保持的隐私保护模型中存在的问题,虽然身份保持的(k,l)-匿名和身份保持的(α,β)-匿名,能够较好地保护个体的记录链接攻击,但是对敏感属性值的约束没有考虑一个个体对应多个记录之间的关系,仍然存在隐私泄露。其次,针对身份保持的隐私保护模型中存在的信息泄露的问题,本文引入了推理空间和推理集的概念,提出了增强型的身份保持l-多样性和增强型的身份保持(α,β)-匿名模型,并将检测QI分组是否满足增强型的身份保持l-多样性模型和增强型的身份保持(α,β)-匿名模型分别转变为最小碰集和敏感属性值出现最大概率问题。针对准标识符属性中的有序属性和无序属性,给出了更为合理的概化信息损失度量方法。进而从信息损失的角度,给出个体间的距离、个体到类的距离以及类间的距离。然后,提出一种通用的基于聚类的身份保持匿名框架,对于不同的匿名模型,在判断一个等价组是否满足给定的匿名约束时会有所不同。首先对原始数据的身份属性进行重编码;然后判断存在数据是否满足给定的隐私要求,若满足,则利用聚类的方法生成满足匿名要求的等价类;否则,对剩余数据进行处理。并对算法进行复杂性分析。最后,从原有身份保持匿名模型的脆弱性,同时从信息损失和运行时间两方面对增强型身份保持和原有身份保持方法进行比较,实验结果表明,本文的增强型身份保持的隐私保护方法提供更强的隐私保护,并且在信息损失和运行时间方面非常接近原有的身份保持匿名方法。
[Abstract]:With the arrival of the information age, more and more data are used by people. For example, the data generated by online shopping can be used for recommendation system, and patients' medical data can be used to study the relationship between medical complications and diseases, etc. But these data usually contain personal privacy information, which, if published directly, can lead to personal privacy disclosure. At the same time, simply deleting the identity information of individuals in these data can not achieve the purpose of protecting personal privacy. Attackers can infer or analyze the sensitive information of individuals through the correlation of data. Therefore, more and more researchers pay attention to privacy protection in data release. Regarding the privacy protection method of data publishing, the earliest protection method is k- anonymous model, and then some extended models are proposed based on it. Although these methods can protect the privacy of users, they often delete the identity attributes of individuals and ignore the relationships between individuals. Tong Yunhai and others have proposed a privacy protection method for identity preservation, but his anonymous model does not fully consider the relationship between individual and multiple records, which still results in privacy disclosure. Therefore, this paper further studies the privacy protection methods of identity preservation. The main research work is as follows: firstly, the problems existing in the existing privacy protection models of identity preservation are analyzed, although identity preserving anonymity and identity preserving (伪, 尾 -anonymity) are analyzed. It can better protect the individual's record link attack, but the constraint on the sensitive attribute value does not take into account the relationship between one individual and multiple records, and there is still privacy disclosure. Secondly, aiming at the problem of information leakage in the privacy protection model of identity preservation, this paper introduces the concepts of reasoning space and inference set, and proposes an enhanced identity preserving (伪, 尾 -anonymity) model with l-diversity and enhanced identity preservation (伪, 尾 -anonymity). The problem that the QI group satisfies the enhanced identity preserving ldiversity model and the enhanced identity preserving model (伪, 尾 -anonymity model) is transformed into the minimum collision set and the maximum probability of sensitive attribute values, respectively. For the ordered and unordered attributes in the quasi-identifier attribute, a more reasonable method to measure the loss of information is given. From the angle of information loss, the distance between individuals, from individuals to classes and between classes is given. Then, a generic cluster-based identity preserving anonymity framework is proposed. For different anonymous models, there are differences in determining whether an equivalent group satisfies a given anonymous constraint or not. Firstly, the identity attributes of the original data are re-coded; then, the existence data is judged to satisfy the given privacy requirements. If the data is satisfied, the equivalent class satisfying the anonymity requirement is generated by clustering method. Otherwise, the remaining data is processed. The complexity of the algorithm is analyzed. Finally, the vulnerability of the original identity maintenance anonymous model and the comparison between the enhanced identity retention and the original identity maintenance methods are compared from the aspects of information loss and running time. The experimental results show that, The enhanced identity preserving privacy method in this paper provides stronger privacy protection and is very close to the original identity anonymous method in terms of information loss and running time.
【学位授予单位】：广西师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309

【相似文献】