【作者】 张浩；

【导师】 俞能海；

【作者基本信息】 中国科学技术大学 ， 信息安全， 2014， 博士

【摘要】 移动互联网将信息技术的发展带入了一个新时代,对人类的发展有着极为深刻的意义,已经影响到了医疗、娱乐、金融、政治、教育等人类生产生活的各个领域。移动性是移动互联网最为重要的特性之一,与地理位置信息的结合,使得移动互联网与人们生活结合更加紧密。基于位置的服务(LBS:Location-Based Service)便是移动互联网中最为耀眼的服务模式之一,仅我国就已形成了数百亿规模的LBS市场。LBS应用已成为移动互联网中人们最为关注的应用服务。然而,LBS的发展并非一帆风顺,使用LBS应用时的隐私泄露问题便是制约LBS进一步发展的重要因素。目前已有大量关于LBS隐私保护的研究工作。LBS应用可分为“用户提问——服务器应答”模式与“服务器提问——用户应答”模式。对于LBS隐私保护的研究工作大都集中在第一类模式上,而对第二类模式关注较少。为此,本文主要关注第二类模式,特别是其典型应用——基于位置的信息统计应用中的隐私保护问题。在该应用中,参与者向服务器贡献位置信息和特定的个人数据,服务器从中计算相应信息的地理分布,在此过程中需要保护参与者的位置隐私和数据隐私。论文的主要研究工作和创新成果如下：1.提出了基于移动云计算的LBS隐私保护协议PPPL。该协议基于移动云计算中的“克隆”技术,结合P2P技术和具有同态性质的公钥加密算法,在“独立半可信”的安全模型下实现对用户个人位置信息和数据信息的隐私保护。PPPL协议既克服了采用中心可信代理的隐私保护方法中可信代理的瓶颈问题,又克服了去代理的隐私保护方法的本地资源消耗多、隐私保护强度不稳定等缺点。仿真实验结果表明,在大规模应用中,相比于使用中心代理的方法中代理的负载增长速率O(n),单个克隆体的负载增长速率仅为0(logn)。2.提出了可以抵御篡改攻击的多聚合协议SMAP与GMAP。SMAP与GMAP协议基于弱化了“半可信”安全假设的安全模型,通过多次计算的方式,使得LBS服务器在攻击者修改部分预处理结果的情况下,仍然能够有较大的概率得到正确的结果。理论证明和仿真实验结果均表明,在控制因子y∈(0,0.5)时,简单多聚合协议SMAP相比于PPPL协议具有更高的安全性；而广义多聚合协议GMAP通过增大安全因子h,进一步增强了抵御篡改攻击的能力。同时,为了能够实现安全性与性能的平衡,本文提出并证明了最优参数的选择方法,使得多聚合协议在能够实现所要求的安全性的同时,最大限度地减小协议开销。3.提出了抵御服务器与恶意用户共谋的多路聚合协议MPAP与SMPAP。 MPAP协议基于弱化了“独立”安全假设的安全模型,将数据分解为多个部分,并通过多条路径传输实现对用户数据的保护。理论证明结果表明,在控制因子γ相同的情况下,MPAP协议比PPPL协议具有更高的隐私保护能力。然而,将数据分解的方式使得攻击者对结果正确性的威胁(如阻塞攻击)进一步增加,任意数据分片的丢失即会使得服务器无法得到正确的结果。为此,本文进一步提出了基于Shamir门限的多路聚合协议SMPAP。理论证明与仿真实验结果均表明,SMPAP协议相比于MPAP协议,在不降低隐私保护强度的条件下,大大降低了阻塞攻击对结果正确性的威胁。4.提出了一种高效的针对LBS信息统计应用中信息隐私保护的噪声添加协议NAP。在该协议中,噪声是实现对用户数据隐私保护的关键因素,为此,本文对应用的结果准确性和数据隐私性进行量化,在此基础上构建了一个数学架构来寻找最优噪声,即保证结果偏差在可容忍范围内的情况下最大化隐私保护能力,并进一步得到最优噪声分布与用户原始数据分布的关系。在此基础上,针对给定原始数据分布为高斯分布、截断高斯分布和任意连续分布的情况下,研究最优噪声分布的特性,得到相应的最优或近似最优的噪声分布。仿真实验结果表明,在给定原始数据分布的情况下,NAP协议得到的噪声分布性能远好于均匀分布和拉普拉斯分布,已达到或接近理论最优噪声分布性能。更多还原

【Abstract】 As a brand new age of information technology, mobile Internet has great meanings for human development, which has affected our society in all aspects of our production and life such as health, entertainment, finance, politics, and education. One of the most important feature of mobile Internet is mobility. By introducing the location information, mobile Internet becomes more integrated with our daily lives. Location Based Service (LBS) is one of the hottest information services. Only in China, it has formed a huge marketing with tens of billions Yuan. LBS has been the most conspicuous service.However, the development of LBS is not always smooth. Privacy disclosure in the usage of LBS is one of the key factors that limits its development. Recently, there are plenty of researches for privacy protection in LBS. LBS applications can be divided into two categories:"user-ask and server-answer" model and "server-ask and user-answer" model. Most of the LBS privacy-preserving researches are based on the first model, while only a few researches pay attention to the second one. Therefore, in this dissertation, we focus on the privacy protection of the second model, especially the typical application, location-based information survey application (LB-ISA). In this application, the participants contribute their location information and individual data. The server calculates the geographic distribution of participants’information, while the location privacy and data privacy of the individual user should be protected.The main work and contributions are as follows:1. A privacy-preserving protocol for LS-ISA (PPPL) based on mobile cloud computing is proposed. PPPL is based on the "clone" technology in mobile cloud computing, and combines the P2P technology and homomorphic public key encryption algorithm. It protects the user’s location privacy and data privacy based on the "independent and semi-trusted" threat model. PPPL not only conquers the single point of failure in proxy-based methods, but also overcomes the disadvantages of the privacy-preserving methods without proxy such as consuming much local resources, providing unstable privacy-preserving strength. The evaluation verifies that in the large scale applications, the increase rate of the load on one clone is O(logn), which is far less than0(n) which is the increase rate of the load on the central proxy in proxy-based methods.2. Two multiple aggregation protocols SMAP and GMAP are proposed to defend the modification attack. Based on the threat model weakening the security assumption "semi-trusted", MAP guarantees that the LBS server can get the correct result with a larger probability when the attacker modifies parts of preprocessed results. Theoretical proof and evaluation verify that when the control factor y E (0,0.5), SMAP is safer than PPPL. Furthermore, GMAP strengthens the capability to defend the modification attack by enlarging the security factor h. Meanwhile, in order to balance the security and the performance, a method to choose the optimal parameters is proposed and proved, which minimizes the resource consumption of GMAP while guaranteeing the requested security protection strength.3. Two multi-path aggregation protocols MPAP and SMPAP are proposed to defend the collusion attack. Based on the threat model weakening the security assumption "independent", MPAP protects the user’s data privacy by dividing the user’s data into multiple parts, and transmitting them through multiple different paths. Theoretical proof verifies that under the same control factor y, MPAP provides better privacy protection than PPPL. However, by data segmentation, the server would not get the correct result when any part of the data is lost, which increases the threat to the correctness of results (e.g. blocking attack). Therefore, a Shamir threshold based multi-path aggregation protocol SMPAP is proposed. From the theoretical proof and evaluation, compared with MPAP, SMPAP can largely reduce the threat to the correctness of result under the blocking attack when providing the same privacy-preserving strength.4. An efficient noise addition protocol NAP for information privacy protection in LS-ISA is proposed. In this protocol, noise is the critical factor. Therefore, we quantify the accuracy of result and the privacy of individual data, and develop a mathematical framework to derive the optimal noise distribution, where the noise provides the best privacy protection while guaranteeing that the result has an acceptable deviation. Based on the framework, the relationship of the optimal noise distribution and the distribution of original individual data is investigated. Furthermore, in the situations that the original individual data satisfies Gaussian distribution, the truncated Gaussian distribution and arbitrary continuous distribution respectively, we deeply investigate the properties of the optimal noise distribution, and get the optimal noise distribution or asymptotically optimal one. Evaluation verifies that given the distribution of the original individual data, the performance of the noise distribution from NAP is much better than the performance of Homogeneous distribution and Laplace distribution, and achieves or is close to the performance of the theoretical optimal noise.更多还原