Research On Data Stream Query With Privacy Protection

Numberous application require continuous publication of statistics for monitoring purpose,such as real-time traffic analysis,timely disease outbreak discovery and social trends observation,and this data model is known as a data stream.These statistics may be derived from many users data,and these data may relate to the privacy of the user.Therefore,it is necessary to the protection of user sensitive information while publishing statistical data.A notable paradigm for offering strong privacy guarantees in statistics publishing is differential privacy and it has been used in the publishing of data streams.As we known,at present,differential privacy for data stream query methods are mainly concentrated in two aspects: one is to protect event-level privacy in infinite stream;the other is to protect user-level privacy in finite stream.However,these two methods have great disadvantages.The former releases data with low availability,while the latter can not protect the privacy data of continuous time.In addition to their respective shortcomings,these two methods are still some deficiencies:(1)Only the results of a single query protection.(2)For some related data,the correlation between the data can not be protected.(3)Low accuracy of the query results.For the above problems,the data flow can be divided into correlated data flow and uncorrelated data flow according to whether the data is relevant.Then two algorithms are proposed for these two kinds of data streams.Both algorithms are based on the sliding window model.The main contents are as follows:Firstly,in order to protect the privacy information of uncorrelated data flow,the algorithm DPQW is proposed.The algorithm is made up of two important algorithms,namely Sample and Restructure.DPQW algorithm can not only protect private information,but also improve the accuracy of query set results.Secondly,for correlated data streams,we need not only to protect the privacy information at every time,but also to protect the correlation between data at different time in continuous time.Therefore,we propose the?-window differential-privacy model.At the same time,we also propose a dynamic privacy budget allocation method DBD(Dynamic budget distribution)in the size window,which can improve the availability of data.Thirdly,through the real data set,we compare the DPQW algorithm with the basic noise adding algorithm Lap,and compare the DBD algorithm with other privacy budget allocation algorithms,and verify the effectiveness and usability of the two algorithms proposed in this paper.