【作者】 程超；

【导师】 李海波；

【作者基本信息】 华中科技大学 ， 计算机软件与理论， 2015， 硕士

【摘要】 随着信息技术的发展与普及,大量个人信息被发布以用于数据挖掘,这些信息在为各行业提供知识及商业价值的同时,也给个人隐私信息的安全造成了威胁。因此,研究新的、实用的隐私保护技术具有重要的意义。对于现有的面向分类挖掘的隐私保护技术存在安全性不足的问题,和现有的满足独立L-多样性原则的隐私保护技术不适用于分类挖掘应用的问题,首先,给出了面向分类挖掘的独立L-多样性隐私保护算法,算法分为三个操作步骤:初始化操作是为不同类型非隐私属性生成语义树,并对相应语义树进行初始化以提高算法效率,划分操作是基于Top-Down Specialization算法中自顶向下的泛化思想的两层划分,在划分操作中的划分节点选取操作,使得划分后的匿名数据可以获得较大的信息增益并降低泛化带来的信息损失,使匿名数据可用于分类挖掘,同时,在第一层划分后为分组内每个元组添加L-1个互不相同的噪音隐私属性值,扰动操作使匿名后的数据满足独立L-多样性原则;其次,对给出的算法在安全性、可用性与代价上的分析,理论上证明了经算法匿名后的数据满足分类挖掘的要求,且满足独立L-多样性原则;最后,给出了算法的特性实验、分类挖掘相关算法与现有工作的对比实验,实验结果表明,匿名后的数据可用于分类挖掘,同时有较好的分类效果。更多还原

【Abstract】 With the development and popularization of information technology, a lot of personal information is published to be used for data mining, however, it will threat the privacy of individuals’ with providing people a powerful knowledge and profit. Therefore, the study of new and practical privacy protection technology which aims at the balance between privacy and utility of data is meaningful.For problems that some privacy protection technologys for classification exist the problem of inadequate security, and the existing privacy protection technologys which meet the principle of Independent L-Diversity do not apply to classification, a Privacy-Preserving method with Independent L-Diversity for Classification is proposed. The method has three options: the initialized option is done to improve the efficiency of the algorithm, the partition option based on the method of Top-Down Specialization has two layer division,through the option of Cut Node choosing, we will obtain more information gain and reduce the information loss caused by the generalization, by the way, after the first layer division, each tuple in divided groups will be added noisy values of SA, and the distorb option is done to make the anonymous data meet the principle of independence L-diversity.The anonymous data is proved that it can be used in classify mining applications, and satisfies the principle of Independent L-Diversity by the analysis of algorithm. Finally, results of characteristics of the experiment and related algorithms comparative experiments, show that the algorithm can be used in classification, and it can preserve good utilities in classification with high security.更多还原