基于匿名化的隐私保护数据挖掘技术的研究

发布时间：2019-03-09 09:32

【摘要】：近年来,信息技术和数据科学发展迅猛,并逐渐被应用于各个行业。数据挖掘可以发现潜在的数据信息和数据间的微妙关系并将其用于决策制定,然而,敏感和隐私信息也同样会被挖掘,这将会给数据提供者造成安全风险。匿名化技术是一项通过将同一等价类中的用户准标志信息变的不可区分,从而保护用户身份或敏感信息不被识别的隐私保护技术。目前的相关算法多是针对关系型等结构型数据集而设计,无法直接用于事务数据集的匿名化。少数针对事务数据的匿名化算法面临着数据高维度,稀疏型数据敏感等问题,从而导致算法时间复杂度高,信息损失大。此外,现实数据中敏感信息往往因人而异,攻击者的背景知识也往往存在局限性。因此,本文的主要研究为:针对事务数据集的匿名化隐私保护问题,本文提出了一个K-匿名隐私保护数据挖掘算法PTA,针对性的解决了事务数据集匿名化时间复杂度高和信息损失大的问题。本文将最小化信息损失的问题转换为求解TSP问题中的最短环路问题,并通过一个类似Prim算法的思想求解最短环路,再通过设计的映射、投票和筛选操作对信息损失进行优化,从而实现了事务数据集的K-匿名隐私保护。此外,本文采用分而治之的思想,以用来降低算法的时间复杂度。实验表明,该算法在时间复杂度和信息损失量方面均优于现有算法。针对个性化的匿名化隐私保护问题,本文首次提出了个性化层次事务数据集的隐私保护问题。同时,考虑到传统的L-多样性隐私保护技术存在的缺陷,本文提出了一个(L,P)-多样性的隐私保护概念,并在此基础上提出了一个贪婪的Lnn-means算法。该算法首先通过层次泛化和矩阵化方法将原始数据转换为类似关系数据集的形式,并采用聚类技术将相似度较高的事务记录进行聚类。最后,通过优先对信息损失量小且满足(L,P)-多样性的事务记录进行等价类生成,从而实现了(L,P)-多样性隐私保护。事实上,Lnn-means算法不仅弥补了传统L-多样性隐私保护技术的缺陷,还在一定程度上避免了L-多样性隐私保护可能面临的语义攻击问题,隐私保护能力更强,安全性更高。总体来说,本文针对性的解决了事务数据集和个性化事务数据的匿名化隐私保护问题。大量的实验证明,本文所提出的框架与算法在实现匿名化需求上是可行且有效的。
[Abstract]:In recent years, the rapid development of information technology and data science, and gradually applied to various industries. Data mining can discover potential data information and the subtle relationship between data and use it in decision-making, however, sensitive and privacy information will also be mined, which will cause security risks to data providers. Anonymization is a privacy protection technology which can protect users' identity or sensitive information from identification by changing the user quasi-flag information in the same equivalent class into indistinguishable information. At present, most of the related algorithms are designed for relational and other structured datasets, and can not be directly used for anonymization of transaction datasets. A few anonymous algorithms for transaction data are faced with problems such as high dimension and sparse data sensitivity, which lead to high time complexity and large loss of information. In addition, sensitive information in real data often varies from person to person, and attacker's background knowledge often has limitations. Therefore, the main research of this paper is as follows: for the problem of anonymized privacy protection of transaction data sets, this paper proposes a K-anonymous privacy protection data mining algorithm, PTA,. It solves the problems of high time complexity and information loss in transaction data set concealment. In this paper, the problem of minimizing information loss is transformed into solving the shortest loop problem in the TSP problem, and the shortest loop problem is solved by an idea similar to the Prim algorithm, and then the information loss is optimized by mapping, voting and filtering operations designed. Thus, the K-anonymity privacy protection of transaction data set is realized. In addition, the idea of divide-and-conquer is used to reduce the time complexity of the algorithm. Experimental results show that the proposed algorithm is superior to the existing algorithms in terms of time complexity and information loss. In this paper, the privacy protection of personalized hierarchical transaction data set is proposed for the first time in order to solve the privacy protection problem of personalization anonymity. At the same time, considering the shortcomings of the traditional L-diversity privacy protection technology, this paper proposes a concept of (L, P)-diversity privacy protection, and then proposes a greedy Lnn-means algorithm based on the concept of (L, P)-diversity privacy protection. Firstly, the original data is transformed into similar relational data set by hierarchical generalization and matrix method, and the transaction records with high similarity are clustered by clustering technology. Finally, the (L, P)-privacy protection of (L, P)-diversity is realized by generating equivalent classes of transaction records satisfying (L, P)-diversity, which are small amount of information loss and satisfying (L, P)-diversity. In fact, the Lnn-means algorithm not only makes up for the shortcomings of the traditional L-diversity privacy protection technology, but also avoids the semantic attack problem that the L-diversity privacy protection may face to a certain extent, and the privacy protection ability is stronger and the security is higher. In general, this paper solves the privacy protection problem of transaction data set and personalized transaction data. A large number of experiments show that the proposed framework and algorithm are feasible and effective in realizing anonymization requirements.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309;TP311.13

【相似文献】