一种面向聚类的对数螺线数据扰动方法  

A Logarithmic Spiral Based Data Perturbation Method for Clustering

作　　者：黄茂峰[1] 倪巍伟[1] 王佳俊[1] 孙福林[1] 崇志宏[1] 

机构地区：[1]东南大学计算机科学与工程学院,南京211189

出　　处：《计算机学报》2012年第11期2275-2282,共8页Chinese Journal of Computers

基　　金：国家自然科学基金(61003057;60973023)资助~~

摘　　要：面向挖掘应用的隐私保护数据发布要求对数据集进行隐藏的同时维持数据的挖掘可用性,数据扰动是解决该问题的有效方法.现有的面向聚类的数据扰动方法难以兼顾原始数据个体隐私和维持数据聚类可用性,对此提出了一种基于对数螺线的隐私保护数据干扰方法.通过构建面向聚类的隐私保护数据扰动模型,利用对数螺线对原始数据进行扰动隐藏,维持原始数据的k邻域关系稳定,实现数据集聚类可用性的有效维护;进一步提出多重对数螺线扰动的策略,提高隐私保护强度.理论分析和实验结果表明:文中方法能够有效地避免数据隐私泄露,同时维持数据的聚类可用性.Privacy-Preserving Data Publication requires seeking tradeoffs among maintaining data utility and preserving privacy of the dataset, data perturbation is an effective method to meet this requirement. The existing data perturbation algorithms for clustering usually fall short in accommodating maintaining individual privacy and data availability simultaneously. In this paper, a novel logarithmic spiral based data obfuscation method is proposed. A clustering-oriented data per- turbation model is built to regulate the obfuscation process. The original dataset can be perturbed leveraging the logarithmic spiral, which can maintain k neighborhood relationship of the data set as well as its clustering utility effectively. Furthermore, for improving the strength of privacy protection, the paper proposes a multiple logarithmic spiral perturbation strategy. Theoretical analysis and experimental results demonstrate that this method can avoid leaking the data privacy meanwhile maintaining better clustering utility.

关 键 词：隐私保护 数据挖掘 聚类分析 对数螺线 数据干扰 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]