差分隐私合成数据研究

发布时间：2018-03-19 02:27

本文选题：隐私保护　切入点：数据发布　出处：《安徽理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：过去的几十年里,互联网的发展彻底改变了我们的生活方式。网络也逐渐成为人们生活的中心,网上购物、查资料、看新闻逐渐成为人们日常生活离不开的一部分。这种日常生活的网络化塑造了一个网络时代和一大批与我们息息相关的互联网公司。这些互联网公司往往提供优质而免费的服务,并且拥有巨量的用户和资源。互联网技术的飞速发展,网络为各种数据的发布、采集、存储和分析带来了一定的便捷。比如医疗单位长时间建立的患者诊断数据集记录,电子商务网站收集的用户个人基本信息等。对这些数据进行大量挖掘分析能够得到大量的关于现实生活的许多有用信息。但这些数据中通常会包含很多个人隐私信息,直接发布有利于数据挖掘分析的实现,但将给个人隐私信息带来一定的威胁,因此,为了更好的解决好这个矛盾,数据发布中的隐私保护技术问题引起了学术界广泛的研究和探讨。近年来,我们已经目睹了多起用户隐私泄漏事件,多家公司深陷其中;而这些事件全都是由于数据拥有者分享数据不当引起的。因此,如何保护数据发布中的隐私问题也成为一项引人注目的话题。本文主要研究了如何在基于差分隐私保护条件下有效发布合成数据集技术,文中对隐私保护、差分隐私保护领域已有的研究成果进行了总结概述,对隐私保护领域内的基本原理和特征进行了阐述,通过研究现今几种差分隐私数据集发布技术,对比总结,最终分析和提出了一种基于差分隐私的分类数据匿名化合成发布方案,使得在基于差分隐私保护条件下分类数据匿名化发布技术既能够保护用户的隐私信息,又不影响数据的可用性,达到了研究的目的。在数据发布方面,分析介绍了交互式和非交互式两种差分隐私保护数据发布方法,并着重从精确度和样本复杂度的角度对这些方法进行分析和对比,介绍了各种非交互式方式下数据发布技术。本文以理论分析为基础,综合实验平台,对提出的一种分类数据合成匿名化发布技术进行实验设计、实验数据论证,其实验结果表明:当发布数据集满足一定条件时,通过添加更少的噪音,可以达到降低源数据的失真程度和提高数据可用性的目的,使得数据的可用性和隐私性相结合达到更好的平衡,从而实现了差分隐私保护条件下数据发布技术。
[Abstract]:In the past few decades, the development of the Internet has completely changed our way of life. The Internet has gradually become the center of people's lives. Watching news is becoming a part of people's daily life. The network of daily life has shaped a network era and a large number of internet companies which are closely related to us. These internet companies often provide the best. Quality and free service, With the rapid development of Internet technology, the network has brought a certain convenience for the release, collection, storage and analysis of various data, such as the patient diagnosis data set records established by medical units for a long time. A lot of useful information about real life can be obtained by mining and analyzing these data. But these data usually contain a lot of personal privacy information. Direct release is beneficial to the realization of data mining and analysis, but it will bring a certain threat to personal privacy information. Therefore, in order to better solve this contradiction, Privacy protection technology in data release has caused extensive research and discussion in academic circles. In recent years, we have witnessed a number of user privacy leaks, many companies are trapped in it; These events are all caused by improper sharing of data by data owners. How to protect the privacy in data release has also become a noticeable topic. This paper mainly studies how to effectively publish composite data sets under the condition of differential privacy protection, and how to protect privacy in this paper. This paper summarizes the existing research achievements in the field of differential privacy protection, expounds the basic principles and characteristics in the field of privacy protection, and compares and summarizes several kinds of differential privacy data set publishing techniques. Finally, this paper analyzes and proposes a scheme of classified data anonymous composition and release based on differential privacy, which can protect the privacy information of users under the condition of classified data anonymous publishing under the condition of differential privacy protection. At the same time, it does not affect the availability of data, and achieves the purpose of research. In the aspect of data release, two kinds of interactive and non-interactive differential privacy protection data publishing methods are analyzed and introduced. This paper analyzes and compares these methods from the angle of accuracy and sample complexity, and introduces the techniques of data release in various non-interactive ways. This paper is based on theoretical analysis and synthesizes the experimental platform. An experimental design of a classified data synthesis and anonymous publishing technique is carried out. The experimental data demonstrate that: when the published data set meets certain conditions, less noise is added to the data set. It can reduce the distortion of source data and improve the availability of data, and make the combination of data availability and privacy achieve a better balance, thus realizing the technology of data release under the condition of differential privacy protection.
【学位授予单位】：安徽理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP309

【相似文献】