【作者】 王旭；

【导师】 万长胜； 陈海涛；

【作者基本信息】 东南大学 ， 电子与通信工程（专业学位）， 2021， 硕士

【摘要】 随着信息应用技术的高速发展,完备的数据成为各行各业发展的必要前提,在此背景下,数据共享也成为云存储技术的热门应用之一。然而由于数据本身蕴藏的巨大价值,数据在共享过程中的安全问题愈加严重。恶意用户、恶意云存储服务器以及黑客会通过各种方法窥探用户的隐私,常见的两种情况分别是通过用户的数据访问模式来推测用户访问数据的内容以及通过发布的数据来挖掘用户敏感信息。针对用户数据访问模式的保护,通常利用不经意的随机访问技术ORAM(Oblivious Random Access Memory)来隐藏用户访问模式,然而这些技术无法应用在多用户场景下,且无法实现数据访问权限的细粒度管理。针对上述问题,本文提出了三种基于属性的广播加密的多用户ORAM数据共享方案,主要工作内容如下:1.研究了目前较为经典的ORAM框架原理,分析其性能和特点,结合多用户数据共享场景,选取path ORAM作为方案的基本框架。2.提出基于属性的广播加密的多用户数据共享ORAM基本方案的思路,搭建了由第三方可信代理、用户以及云存储服务器组成的系统模型,以实现多用户共享数据访问模式保护和可撤销的用户访问权限控制。面向两个不同的数据访问应用场景,在基本方案的基础上提出了两个读写权限可控的多用户共享ORAM方案。经分析讨论,证明了上述方案满足选择明文安全,且具有良好的通信复杂度。最后经实验仿真证明,方案在处理多用户同时访问时,具有较优的用户平均访问效率,具有一定的实用性。针对发布数据所包含的敏感信息保护,最常用的方法之一是数据匿名化,然而将部分特征隐去后会导致发布的数据后续挖掘价值大打折扣,违背了数据共享的初衷。针对上述问题,本文提出了一种基于极端梯度提升特征重要性的K-匿名特征选择算法,主要工作内容如下:1.分析了最经典的3类数据匿名化技术,并结合关系型数据表最常遭遇链接攻击这一背景,选取K-匿名技术作为本次方案的特征选择算法。研究了极端梯度提升算法以及特征重要性指标的基本原理和特点。2.提出基于极端梯度提升特征重要性的K-匿名特征选择算法,该算法利用特征重要性衡量特征价值,并结合K匿名特征选择算法保护隐私信息,从而同时实现了数据敏感信息和数据价值的保护。在真实数据集上进行上述方案的仿真实验,结果证明了本方案的正确性与高效性。更多还原

【Abstract】 With the rapid development of information application technology,complete data has become a necessary prerequisite for the development of various industries.In this context,data sharing has also become one of the popular applications of cloud storage technology.However,due to the huge value inherent in the data,the security problems in the data sharing process have become more and more serious.Malicious users,malicious cloud storage servers,and hackers will spy on users’ privacy through various methods.The two common situations are inferring the content of user access data through user data access patterns and mining sensitive user information through published data.Regarding the protection of user data access patterns,inadvertent random access technologies are usually used to hide user access patterns.However,these technologies cannot be applied in multi-user scenarios and cannot achieve fine-grained management of data access permissions.In response to the above problems,this article proposes three attribute-based broadcast-encrypted multi-user ORAM data sharing schemes.The main work content is as follows:1.Research the principle of the classic ORAM framework so far,analyze its performance and characteristics,combine the multi-user data sharing scenario,and select path ORAM as the basic framework of the program.Understand the basic principles of attribute-based broadcast encryption algorithms.2.Propose the idea of multi-user ORAM data sharing basic scheme based on attributebased broadcast encryption,and build a system model composed of third-party trusted agents,users,and cloud storage servers to realize multi-user shared data access mode protection and revocable users Access control.Facing two different data access application scenarios,two multi-user shared ORAM solutions with controllable read and write permissions are proposed on the basis of the basic solution.After analysis and discussion,it is proved that the above scheme satisfies CPA security and has good communication complexity.Finally,it is proved by experimental simulation that this scheme has better average user access efficiency when dealing with simultaneous access by multiple users,and has certain practicability.Regarding the protection of sensitive information contained in published data,one of the most commonly used methods is data anonymization.However,concealing some features will cause the subsequent mining value of the published data to be greatly reduced,which violates the original intention of data sharing.In response to the above problems,this paper proposes a K-anonymous feature selection algorithm based on extreme gradient enhancement of feature importance.The main work content is as follows:1.Analyze the most classic three types of data anonymization technologies,and combined with the background that relational data tables are most frequently encountered with link attacks,select K-anonymity technology as the feature selection algorithm for this project.Research the basic principles and features of extreme gradient boosting algorithms and feature importance indicators.2.Propose a K-anonymous feature selection algorithm based on extreme gradients to enhance the importance of features.The algorithm uses feature importance to measure feature value,and combines K anonymity feature selection algorithm to protect private information,thereby simultaneously achieving data sensitive information and data value protection.After a comparative analysis and discussion of the simulation experiment results on the real data set,the correctness and efficiency of the scheme is proved.更多还原