【作者】 李晔；

【导师】 王轩；

【作者基本信息】 哈尔滨工业大学 ， 计算机应用， 2018， 博士

【摘要】 随着大数据的快速发展,如何有效地保护用户隐私成为亟待解决的重要问题。基于数据加密的具有隐私保护的数据挖掘技术是解决这一问题的有效手段之一。因此,本文以数据分类算法为研究对象,结合云计算技术,从安全模型(半诚实、恶意)、云计算服务(计算外包、存储及计算外包)及数据分布(水平分布、垂直分布和任意分布)等方面对具有隐私保护的数据分类方法展开研究,主要包括以下三点:首先,对半诚实模型下计算外包的隐私保护分类方法进行了研究。半诚实模型是指支持存在半诚实参与方的安全模型,半诚实参与方在协议执行中,完全遵循协议执行,但是可以通过协议执行过程中的中间信息推断其他参与方的隐私信息。本文将主要研究的三类数据分类方法(ID3、C4.5及随机决策树)在不同分布的数据集(水平、垂直、任意)中的隐私保护问题分别归纳为平均数计算问题、交集计算问题、频繁项集计算问题、安全和问题及安全投票问题等。(1)本文将两方安全平均数计算协议改进为基于单密钥半同态加密的多方外包安全平均数计算协议。在此基础上,提出了基于水平分布数据集的多方隐私保护ID3方法。(2)将两方安全平均数计算协议、安全交集计算协议、安全频繁项集计算协议改进为基于多密钥半同态加密的多密钥外包安全平均数计算协议、多密钥外包安全交集协议及多密钥外包安全频繁项集计算协议等,并在此基础上提出了基于水平、垂直及任意分布数据集的多方隐私保护C4.5方法。(3)将安全和计算协议及安全投票协议,改进为基于多密钥半同态加密的外包安全和计算协议以及多方外包安全投票协议,并提出了基于水平分布数据集的多方计算外包隐私保护随机决策树方法。通过安全性分析,各个安全子协议在半诚实模型下是安全的,所提出的各类计算外包隐私保护数据分类方法(ID3、C4.5及随机决策树)在半诚实模型下也是安全的。通过性能分析表明,所提出的方法也大大降低了用户端的负载。其次,对半诚实模型下存储及计算外包的隐私保护分类方法进行了研究。提出了两方及多方存储及计算外包的隐私保护ID3决策树方法。首先,将该存储及计算外包的ID3方法分解为密文数据比较、ln计算、密文最小值选择等运算。然后,基于单密钥同态加密方法提出了安全比较协议、两方安全ln计算协议及安全最小值选择算法,提出了存储及计算外包的两方隐私保护ID3方法。但由于单密钥同态加密方法难以有效支持多参与方的数据分类方法,本文进一步采用了EPOM多密钥同态加密方法,并提出了多方安全ln计算协议,从而提出了存储及计算外包的多方隐私保护ID3方法。通过安全性分析表明上述各类安全子协议在半诚实模型下是安全的,所提出的两方及多方存储及计算外包的隐私保护ID3决策树方法也是安全的。通过性能分析表明,所提出的方法也大幅降低了用户端与云端的数据通信。最后,对恶意模型下存储及计算外包的隐私保护分类方法进行了研究。恶意模型是指支持恶意参与方存在的安全模型,恶意参与方在协议执行中,不仅可以做任何事情来推断其他参与方的隐私信息,而且可以更改原始输入、篡改交互信息及中断协议执行。提出了恶意模型下两方外包存储及计算的隐私保护数据ID3决策树方法。针对恶意参与方存在的情况,提出了基于混淆电路的安全平均数计算协议,并结合两方安全ln计算协议、安全最小值选择算法及安全比特位分解协议等安全协议,提出了数据恶意模型下两方存储及计算外包隐私保护ID3方法。通过安全分析,所提出的方法可以在云端计算服务器恶意情况下,基于混淆电路的安全平均数计算协议是安全的。同时,在参与方和云端存储服务器是半诚实的情况下,所提出的恶意模型下两方外包存储及计算的隐私保护数据ID3决策树方法是安全的。更多还原

【Abstract】 With the rapid development of big data,how to effectively protect user privacy has become an important problem to be solved urgently.Privacy preserving data mining technology based on data encryption is an effective way to solve this problem.Therefore,this paper mainly aims at this problem,and combines cloud computing technology to research privacy protection data classification method from different security models(semi honest,malicious),different cloud computing methods(Computing outsourced,storage and computation outsourced)and different data distribution(horizontal distribution,vertical distribution and arbitrary distribution).The research is carried out,mainly including the following three points:First,the privacy preserving classification method for computing outsourced under semi honest model is studied.In this paper,three kinds of data classification methods(ID3,C4.5 and random decision tree)are mainly studied in the data sets of different distribution(horizontal,vertical and arbitrary),including the average computing problem,the intersection calculation problem,the frequent item set calculation problem,the security and the problem and the security voting problem.(1)in this paper,the two-party safety average computation protocol is improved to a multi party outsourced average average computing protocol based on single key semi homomorphic encryption.On this basis,a multi-party privacy preserving ID3 method based on horizontal distributed data sets is proposed.(2)to improve the two-party security computing protocol,to multi key outsourcing security average computing protocol,multi-key outsourcing security intersection protocol and multi-key outsourcing security frequent itemset calculation protocol.A multi-party privacy preserving C4.5 method based on horizontal,vertical and arbitrarily distributed datasets is presented.(3)the security and computing protocol and the security voting protocol are improved to the outsourcing security and computing protocol based on multi key semi homomorphism and the multi-party outsourcing security voting protocol,and a multi party computing outsourcing privacy protection random decision tree method based on the horizontal distributed data set is proposed.Through security analysis,each security sub protocol is safe under the semi honest model.The proposed classification method of various computing outsourcing privacy preserving classification method(ID3,C4.5 and random decision tree)is also safe under the semi honest model.The performance analysis shows that the proposed method also greatly reduces the load on the user side.Secondly,a storage and computation outsourced privacy preserving classification method under semi honest model is proposed.A privacy preserving ID3 decision tree method for two party and multi-party outsourced storage and computation is proposed.First,the storage and computation outsourced privacy preserving ID3 method,is further decomposed into operations such as ciphertext data comparison,ln computation,and ciphertext minimum value selection.Then,based on the single key homomorphic encryption method,this paper proposes the security comparison protocol,the two-party secure ln computing protocol and the security minimum value selection algorithm,and the two-party privacy protection method for the outsourced storage and calculation is proposed.However,because the single key homomorphic encryption method is difficult to effectively support the multi-party data classification method,this paper further adopts the EPOM multi key homomorphic encryption method,and proposes a multi-party secure ln computing protocol,thus proposing the multi-party privacy protection ID3 method for outsourcing storage and computing.The security analysis shows that the all kinds of security sub protocols are safe under the semi honest model.The proposed ID3 decision tree method is also safe for the two party and multi party storage and computing outsourcing.The performance analysis shows that the proposed method also greatly reduces the data communication between the user side and the cloud.Finally,the storage and computation outsourced privacy preserving classification method under the malicious model is studied.A privacy preserving data ID3 decision tree method for two party storage and computation outsourced under malicious models is proposed.Firstly,in view of the presence of malicious participants,a security averaging protocol based on garbled circuit is proposed,and the two-party security ln protocol,the security minimum selection algorithm and the security bit decomposition protocol are combined,and the two-party storage and the computing outsourcing privacy protection ID3 party under the data malicious model are proposed.By means of security analysis,the proposed method can calculate the security averaging protocol based on garbled circuit in cloud computing server malicious case.At the same time,under the semi honest case of the participant and the cloud storage server,the ID3 decision tree method of the two-party outsourcing storage and computing privacy protection data under the malicious model is safe.更多还原