【作者】 王超；

【导师】 杨静；

【作者基本信息】 哈尔滨工程大学 ， 计算机应用技术， 2015， 博士

【摘要】 近年来,随着信息技术、移动通信技术、位置感知技术和互联网技术的迅速发展,各领域产生了海量的用户数据,尤其是个人位置和轨迹数据,对其进行数据分析、挖掘,能获得数据潜在的规则及商业价值。然而数据是把双刃剑,只要有数据,就必然存在安全与隐私的问题。为了对数据进行分析,常常需要对存储在服务器上的数据进行发布,这无疑会给心怀恶意的用户(称为攻击者)带来可趁之机,威胁用户的隐私。针对用户信息的推理性攻击可能导致个人的兴趣爱好、行为模式、社会习惯等隐私信息暴露,严重威胁到用户的个人生命、财产安全。因此,为了保护发布数据中个人的隐私信息,需研究数据发布的技术和方法,使得发布的数据既能保护其隐私性又能保持较高的数据效用。本文以隐私保护领域的两种典型数据类型——关系数据和轨迹数据为研究背景,对隐私匿名方法展开研究,在保证数据安全性的前提下,通过相应的匿名技术,以适当的信息损失为代价,换取匿名数据更高的可用性,最终达到平衡数据可用性和隐私保护度的目的。本文的研究内容主要包括以下四个部分:首先,现有的许多基于限制发布的技术在实现k匿名算法时,采用基于分治策略的泛化技术,减少了潜在的匿名组数量;虽然基于取整划分函数的划分策略,避免了“可能减少潜在匿名组数量”这一情况的发生,但是没有考虑临时匿名组中相邻数据点的距离,在划分过程中极易产生许多不必要的信息损失,从而影响发布的匿名数据集的可用性。同时,在p-敏感k-匿名模型中,存在由于敏感属性值在聚类结果中分布不均匀,而遭到背景知识攻击,泄露敏感信息的情况。针对上述问题,本文提出基于投影区域密度划分的k匿名算法和基于敏感属性熵的微聚集算法,分别从数据可用性和隐私保护度两方面解决k匿名模型存在的问题。其次,在轨迹数据中,最大的隐私威胁就是“敏感位置泄露”,如果攻击者能够了解某人在哪些时间访问了哪些位置,那么攻击者就能够确定此人在发布数据库中的真实记录,并且能够了解此人的其他轨迹信息,进而推理得到此人的行为模式、生活习惯等敏感信息,造成个人隐私信息的泄露。对此,本文提出了一种新的轨迹相似性度量模型,和基于轨迹位置形状相似性的隐私保护算法,在轨迹聚类过程中使用贪心的思想,最大限度地提高了聚类内部轨迹之间的相似性;并且在位置聚类过程中,使用真实的原始位置信息形成数据“面罩”,满足了轨迹k-匿名,在保护轨迹数据的同时,有效地提高了轨迹数据的可用性。再次,在现实应用中,不同的移动用户对个人隐私需求的定位是不同的,例如:有的用户将自己的住址看作是个人隐私,而有的人则不这么认为。简单地将所有移动用户的隐私保护级别视为等同是不合理的。满足移动对象个性化的隐私需求,不仅会提高移动对象的隐私保护程度,还能有效地降低匿名过程中不必要的信息损失。同时,现有的轨迹匿名算法在计算轨迹相似性时还存在没有充分考虑轨迹内外在特征信息的情况。针对上述问题,本文提出了个性化轨迹k-匿名的概念,和轨迹结构相似性度量模型,并提出了基于稀疏化最小生成树聚类的个性化隐私保护算法,通过贪婪的策略生成一个近似最优的轨迹k-匿名集合,显著地提高了轨迹数据的可用性。最后,现有的轨迹匿名算法在度量相似性时,仅考虑轨迹在采样点位置的邻近性,属于移动对象的静态邻近性研究。在形成移动对象的轨迹k-匿名集合时,往往会遇到这样的情况:开始时刻相邻的移动对象可能会随着移动而逐渐远离,而开始时刻相距较远的移动对象可能会随着移动而逐渐靠近。为此,本文提出了邻域扭曲密度和邻域相似性的概念,并分别提出了基于邻域相似性和邻域扭曲密度的轨迹匿名算法,在聚类过程中充分考虑轨迹位置的动态邻近性,来解决上述问题。更多还原

【Abstract】 In recent years,as the rapid development of information technology,mobile communication technology,location-aware technology and Internet technology,all the field have produced a vast user data,especially the personal position and trajectory data.Analyzing and mining data can help us obtain potential rules and commercial value.However,the data is double-edged sword,as long as there is data,there must be security and privacy issues.In order to analyze the data,it is often necessary to release data stored on the server,and it is no doubt that the malicious user(attacker)will take advantage of this opportunity to threaten users’ privacy.Speculative attack against the user information may lead to the leak of personal interests,behavior patterns,social habits and other private information,and threaten individual’s life and property safety of users,seriously.Therefore,in order to protect the personal privacy of released data,we need to research the techniques and methods of data releasing,to protect the privacy of released data,as well as maintain high data utility.In this research,we use two typical data type of the privacy protection field---relational data and trajectory data as the research background,and study privacy anonymous methods.In the premise of ensuring the data security,through corresponding anonymous technology,our research losses appropriate information to exchange the higher availability of anonymous data.Eventually,it achieves the purpose of balancing the data availability and privacy protection degree.Specifially,the main contents of the paper can be divided into four parts:Firstly,while achieving k-anonymity,lots of existing algorithms based on limited release use generalization technology based on divide and conquer strategy,and reduce potential number of anonymous groups;Although the partition strategy based on rounded partition function avoid the situation that it is likely to reduce the number of potential anonymous groups,it does not consider the distance of adjacent data point in the temporary anonymous group,and can easily produce much unnecessary information loss during the division process,and affect the availability of released anonymous data sets.Meanwhile,in the p-sensitive kanonymity model,the uneven distribution of the sensitive attribute values in the clustering results may cause sensitive information disclosure.Therefore,we propose the algorithm for k-anonymity based on projection area density partition and a micro-aggregation algorithm based on sensitive attribute entropy,to solve these problems of k-anonymity,from the two aspects of data availability and privacy,respectively.Secondly,in the trajectory data,the biggest privacy threat is the "sensitive position leakage",If the attacker is able to understand the position and time that someone visits at,the attacker will be able to determine the person’s true record in the publication database,and understand the person’s other trajectory information,then get the person’s interests,hobbies,behavior patterns,social customs and other private information,and result in the leakage of the person’s privacy information.Therefore,we present a trajectory similarity measure model and a privacy preserving algorithm based on trajectory location and shape similarity,which maximizes the trajectory similarity in the clusters,and forms data "mask" which is formed by fully accurate true original locations information to meet the trajectory k-anonymity.While protecting trajectory data,the algorithm effectively improves the availability of trajectory data.Thirdly,in real applications,different mobile users’ privacy need are different,for example: some users will see his address as privacy,and some one will not think so.Simply seeing the level of privacy protection for all mobile users as equivalent is unreasonable.Meeting moving objects’ personalized privacy needs will not only improve the privacy protection level of moving objects,but also effectively reduce unnecessary information loss,during the anonymous process.Meanwhile,the existing trajectory anonymous algorithms are still not fully consider theinternal and external trajectory feature information,while calculating the trajectory similarity.To solve these problems,we propose the concept of personalized trajectory k-anonymity and trajectory structure similarity measure model,and propose the sparse minimum spanning tree clustering based personalized trajectory privacy protection algorithm.It generates an approximate optimal trajectory k-anonymity set by greedy strategy,and increases the availability of trajectory data significantly.Finally,while calculating trajectory similarity,existing trajectory anonymous algorithm only consider the location proximity of sample point,and it belongs to static proximity research of moving object.During the process of forming tajectory k-anonymous set of moving object,we often encounter such a situation: as the objects moving,the distance of the adjacent objects at the start time will be increasing,and the distance of the objects which are far away at the start time will be decreasing.Therefore,we present the concept of neighborhood similarity and neighborhood distortion density to fully consider the dynamics proximity of locations in the trajectory,and then propose two algorithms-Trajectory Anonymity Algorithm based on Neighborhood Similarity and Trajectory Anonymity Algorithm based on Trajectory Neighborhood Distortion Density to solve the problem.更多还原