基于大时空范围LBS匿名集的推理攻击及隐私保护方法研究

Technological advances in GIS，wireless communications and GPS receivers gave rise to Location Based Service(LBS) that exploits location data to offer services to the requester. As a typical application of context-aware services in pervasive computing environments, LBS can materialize the ambient intelligence. However, given that these services assume an in-depth knowledge of the mobile users' whereabouts，deployment of LBS may easily lead to abuse scenarios and pose a severe threat to users' privacy, which has attracted extensive interests of academia and the industry..During the last decade, LBS privacy protection techniques have been extensively studied. Spatial temporal k-anonymity approach pioneered in Gruteser and Grunwald's seminal paper in 2003 has been a prominent direction of LBS privacy protection, which also inspired a series of research publications. Supporting only anonymous uses of sporadic LBS, traditional approaches will sufferer multi-query attack based on linking a set of adjacent anonymous datasets. Revised approaches were proposed, which are based on reciprocity condition and anonymous datasets prediction. While, they still reveal LBS users' privacy because they will expose LBS users' habits to inference attacks..Inference methodologies based on spatial temporal association rules are mainly used for location prediction, which is focused by researchers of intelligent transportation systems and mobile computing applications. Correspondingly, blocking methodologies based on hiding sensitive spatial temporal association rules are popular with data sharing and publishing applications. Whereas, both these methods proposed so far are mainly focused on users' trajectory data but anonymous datasets, whose differentia will result in invalid simulation of attack scenarios and LBS privacy protection failure. Simultaneously, blocking methodologies should be provided with the capability of dynamic dealing with LBS anonymous queries..Therefore, we propose a new approach, which can resist inference attacks based on spatial temporal association rules of LBS anonymous query datasets and includes four phases: First, it generates anonymous datasets by spatial temporal k-anonymity methods. And then, defining probabilistic support and confidence, it mines historical anonymous datasets to identify frequent behavioral patterns. After that, it generates spatial temporal association rules from patterns and defines attack scenarios: inference attacks to sensitive spatial temporal region and identify inference attacks based on movement behaviors and request contents of anonymous users. Finally, it designs an optimal method to protect LBS privacy from aforementioned attacks by progressively hiding sensitive spatial temporal association rules. Research achievements are expected to promote developments of LBS and enrich GIS privacy preserving data mining.

目前，隐私保护成为LBS进一步发展亟待解决的关键问题。时空K-匿名方法以匿名数据真实可用、方法实现简洁灵活以及更适合LBS移动计算环境等特点，成为近年来研究的主流方向。传统的时空K-匿名及优化方法只适用于单次查询及时空临近的多次查询的隐私保护，并不能应对基于大时空范围匿名集的推理攻击。同时，依据LBS隐私保护对安全性与匿名集数据可用性的特殊需求，本项目提出攻守双方对等感知信息级别的LBS隐私保护机制。在分析移动对象数据的时空关联规则推理与防护方法以及移动对象数据与匿名集数据不同特性的基础上，结合LBS长期、连续、在线服务的特点，本项目研究基于匿名集时空关联规则动态隐藏，应对时空敏感区推理攻击与标识推理攻击的LBS隐私保护方法。成果预期将推动LBS的深入发展与广泛应用，并丰富地理数据挖掘以及地理信息安全等领域的研究内容与理论方法。

随着移动通讯技术、定位技术、地理信息等技术的发展和相互融合，LBS得到飞速发展。但是，一系列位置隐私泄露或非法使用等事件的发生，也使得隐私安全问题逐渐成为LBS进一步深入发展亟待解决的一项关键问题。时空K-匿名方法以数据真实可用、方法实现简洁灵活，适合LBS移动计算环境的特点，成为近年来LBS的隐私安全研究的主流方法。挖掘连续查询的时空K-匿名数据集，抽取出潜在的、有用的规律，可为众多行业应用提供预测分析功能，但同时也会产生基于敏感知识推理攻击用户隐私的威胁。传统的时空K-匿名及优化方法只适用于单次查询及时空临近的多次查询的隐私保护，不能适应LBS应用具有的长期、连续、在线服务的特点，不能应对基于大时空范围匿名集的推理攻击。依据LBS隐私保护对安全性与匿名集数据可用性的特殊需求，本项目提出并研究了基于攻守双方对等感知信息级别的LBS隐私保护机制。在分析移动对象数据的时空关联规则推理与防护方法以及移动对象数据与匿名集数据不同特性的基础上，结合LBS长期、连续、在线服务的特点，研究了时空关联规则的概率化挖掘与推理攻击方法、基于动态阻止推理攻击的渐进式隐私保护方法，以及阻止推理攻击的匿名保护模型的量化评估与优化方法。重点解决了“具有概率化、泛化特性的时空关联规则挖掘与推理分析”以及“动态感知敏感时空关联规则的匿名保护”的关键技术问题。探索建立了具有高安全性，匿名数据可用性以及动态服务特性的LBS隐私保护方法。本项目成果包含：申请了8项专利（其中2项已授权），发表了6篇科研论文（其中国家核心期刊3篇，国际期刊论文2篇，EI检索论文1篇），翻译出版了学术专著1部，培养了硕士研究生6名（其中1名获得国家奖学金，3名已毕业）。完成了合同规定的研究任务，达到了考核指标。成果对于推动LBS的深入发展与广泛应用、丰富地理数据挖掘以及地理信息安全等领域的研究内容与理论方法具有重要意义。