A method, system and storage medium for personalized privacy protection of multidimensional data

Abstract

The present invention provides a method, system and storage medium for personalized privacy protection of multidimensional data, the method comprising: selecting a target privacy budget level from multiple privacy budget levels in a privacy budget list according to user settings; performing biased sampling on multidimensional mixed data according to the target privacy budget level to obtain a dimension index list, and judging the data type of each index in the dimension index list, the data type including categorical data and numerical data; performing corresponding perturbations on the categorical data and the numerical data respectively according to data perturbation rules, and obtaining perturbation vectors of the categorical data and perturbation values ​​of the numerical data after perturbation; performing aggregation operations on the categorical data perturbation vector and the numerical data perturbation value respectively. The method solves the technical problem that the prior art cannot meet the requirements of personalized privacy protection for multidimensional mixed data in the environment of localized differential privacy.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, system and storage medium for personalized privacy protection of multidimensional data. Background Art

[0002] Currently, most of the research on differential privacy protection methods is conducted in a centralized differential privacy environment, which requires a trusted third party to collect data and protect privacy. However, in real life, third parties may still be malicious or vulnerable to malicious attacks.

[0003] At the same time, since the privacy protection budget is set by the data collector as a unified value, it cannot meet the user's personalized privacy needs. In addition, there are still certain gaps in the current research on differential privacy protection for multidimensional mixed data. Therefore, for multidimensional mixed data, how to meet the user's personalized privacy needs in the environment of localized differential privacy is a problem that needs to be solved. Summary of the invention

[0004] Based on this, the purpose of the present invention is to provide a method, system and storage medium for personalized privacy protection of multidimensional data, which is used to solve the technical problem in the prior art that personalized privacy protection of multidimensional mixed data cannot be satisfied in the environment of localized differential privacy.

[0005] On one hand, the present invention provides a method for personalized privacy protection of multidimensional data, comprising:

[0006] Selecting a target privacy budget level from a plurality of privacy budget levels in a privacy budget list according to user settings;

[0007] Perform biased sampling on the multidimensional mixed data according to the target privacy budget level to obtain a dimension index list, and determine the data type of each index in the dimension index list, where the data type includes categorical data and numerical data;

[0008] According to the data perturbation rule, the categorical data and the numerical data are respectively perturbed accordingly, and a perturbation vector of the categorical data and a perturbation value of the numerical data after the perturbation are obtained;

[0009] Aggregation operations are performed on the categorical data disturbance vector and the numerical data disturbance value respectively.

[0010] As a further preferred embodiment of the present invention, the step of performing biased sampling on the multi-dimensional mixed data according to the target privacy budget level to obtain a dimensional index list specifically includes:

[0011] Dividing the multi-dimensional mixed data into a first dimension index group, a second dimension index group and a third dimension index group according to a preset classification rule;

[0012] The dimensional index list is obtained by performing biased sampling on the first dimensional index group and the second dimensional index group according to the target privacy budget level.

[0013] As a further preferred solution of the present invention, the first dimensional index group is used to store the first important index, the second dimensional index group is used to store the second important index, and the third dimensional index group is used to store meaningless or repeated indexes.

[0014] As a further preferred embodiment of the present invention, the steps of performing corresponding perturbations on the categorical data and the numerical data according to the data perturbation rule and obtaining the perturbed categorical data perturbation vector and the numerical data perturbation value specifically include:

[0015] Calculating the numerical data disturbance value according to the first disturbance rule for the numerical data, and correcting the numerical data with a multiple of the numerical data disturbance value;

[0016] The classified data disturbance vector is calculated for the classified data according to the second disturbance rule, and the classified data is corrected by a multiple of the classified data disturbance vector.

[0017] As a further preferred embodiment of the present invention, the step of performing aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively comprises:

[0018] Directly summing up each of the numerical data disturbance values ​​and then calculating the mean value to obtain an unbiased estimate of the statistical mean of the original numerical data;

[0019] After eliminating the noise of each group of the categorized data perturbation vectors corresponding to the target privacy budget level, the frequency values ​​of each group of the categorized data perturbation vectors are directly added to obtain an unbiased estimate of the original categorized data frequency.

[0020] On the other hand, the present invention also proposes a multi-dimensional data personalized privacy protection system, which is applicable to the above-mentioned multi-dimensional data personalized privacy protection method and includes: a parameter selection module for selecting a target privacy budget level from a plurality of privacy budget levels in a privacy budget list according to user settings;

[0021] A data sampling module, used to perform biased sampling on multi-dimensional mixed data according to the target privacy budget level to obtain a dimension index list, and determine the data type of each index in the dimension index list, wherein the data type includes categorical data and numerical data;

[0022] A privacy protection module, used to perform corresponding perturbations on the categorized data and the numerical data according to a data perturbation rule, and obtain a perturbation vector of the categorized data and a perturbation value of the numerical data after the perturbation;

[0023] The aggregation module is used to perform aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively.

[0024] As a further preferred embodiment of the present invention, the data sampling module comprises:

[0025] A data grouping unit, used for dividing the multi-dimensional mixed data into a first dimension index group, a second dimension index group and a third dimension index group according to a preset classification rule;

[0026] A data sampling unit is used to perform biased sampling on the first dimensional index group and the second dimensional index group according to the target privacy budget level to obtain the dimensional index list.

[0027] As a further preferred embodiment of the present invention, the privacy protection module includes:

[0028] A numerical data perturbation unit, used for calculating a numerical data perturbation value for the numerical data according to a first perturbation rule, and correcting the numerical data by a multiple of the numerical data perturbation value;

[0029] The classification data perturbation unit is used to calculate the classification data perturbation vector for the classification data according to the second perturbation rule, and correct it with a multiple of the classification data perturbation vector.

[0030] As a further preferred embodiment of the present invention, the aggregation module comprises:

[0031] A numerical data aggregation unit, used for directly adding up each of the numerical data disturbance values ​​and then calculating the average value, so as to obtain an unbiased estimate of the statistical mean of the original numerical data;

[0032] The categorized data aggregation unit is used to eliminate the noise of each group of categorized data disturbance vectors corresponding to the target privacy budget level, and directly add the frequency values ​​of each group of categorized data disturbance vectors to obtain an unbiased estimate of the original categorized data frequency.

[0033] The present invention also provides a storage medium, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-described method when executing the program.

[0034] According to a multidimensional data personalized privacy protection method, system and storage medium provided by the present invention, excessive noise addition is avoided by allowing users to select a privacy budget level based on their own privacy needs, thereby improving the practicality of the data. The server side can pre-process the multidimensional mixed data and perform biased sampling on the dimensions of the user data to obtain statistical data with higher accuracy. In addition, the sampling dimensions can also be used to make attribute judgments and execute corresponding personalized privacy protection algorithms for numerical data and categorical data respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flowchart of a method for personalized privacy protection of multi-dimensional data in a first embodiment of the present invention;

[0036] Figure 2 It is a structural block diagram of a multi-dimensional data personalized privacy protection system in a third embodiment of the present invention;

[0037] Figure 3 A comparison diagram of the utility loss of numerical data in the present invention;

[0038] Figure 4 This is a comparison diagram of utility loss of data of different types in the present invention;

[0039] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0041] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0042] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0043] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0044] The embodiment of the present invention aims to propose a personalized privacy protection method for multidimensional data, which is applied to the computer field. Specifically, most of the current research on differential privacy protection methods is conducted in a centralized differential privacy environment, that is, a trusted third party is required for data collection and privacy protection, but in real life, third parties are still malicious or vulnerable to malicious attacks. At the same time, since the privacy protection budget is set to a unified value by the data collector, it cannot meet the personalized privacy needs of users, and there are still certain gaps in the current differential privacy protection research for multidimensional mixed data. Therefore, for multidimensional mixed data, how to meet the personalized privacy needs of users in a localized differential privacy environment is a problem that needs to be solved.

[0045] To solve the above problems, an embodiment of the present invention aims to provide a personalized privacy protection method for multidimensional data. The user side selects the privacy budget level according to the user's own privacy needs, and performs personalized mean estimation methods and frequency estimation methods for numerical data and categorical data respectively, and proposes a personalized sampling scheme. The scheme pre-processes attribute labels according to the preferences of the server side, and performs biased sampling on data dimensions according to its collection preferences, thereby providing a better balance between privacy protection and data availability.

[0046] Specifically, the multidimensional data personalized privacy protection method of the embodiment of the present invention includes: selecting a target privacy budget level from multiple privacy budget levels in a privacy budget list according to user settings; performing biased sampling on multidimensional mixed data according to the target privacy budget level to obtain a dimension index list, and judging the data type of each index in the dimension index list, wherein the data type includes categorical data and numerical data; performing corresponding perturbations on the categorical data and the numerical data according to data perturbation rules, and obtaining perturbation vectors of categorical data and perturbation values ​​of numerical data after perturbation; and performing aggregation operations on the categorical data perturbation vectors and the numerical data perturbation values. The technical problem that the prior art cannot meet the requirements of personalized privacy protection for multidimensional mixed data in a localized differential privacy environment is solved.

[0047] For the convenience of subsequent description, we first briefly explain the relevant concepts of localized differential privacy and multidimensional mixed data.

[0048] Localized differential privacy is a privacy protection mechanism with a strict mathematical definition. It guarantees the privacy of user data without trusting any third party. In the LDP algorithm, users inject noise (such as Laplace noise) into their data according to the privacy budget value and send the perturbed data to the server. Therefore, an attacker with any background knowledge cannot infer the original data information of a user.

[0049] A user database U with a size of |U| is composed of a number of multi-dimensional mixed data of users, that is, U = {T1, T2, ..., T |U|}. The user's multidimensional mixed data is an ordered list containing d attribute values, represented by T i =[A1,A2,…,A d ], where A i is the value of the ith attribute. Attributes can be either numeric or categorical, where the range of each numeric attribute is [-1,1], and each categorical attribute has w different classification values, that is, its discrete domain is {1,2,…,w}.

[0050] Embodiment 1

[0051] like Figure 1 As shown, a method flow chart of a first embodiment of the method for personalized privacy protection of multidimensional data of the present invention is provided, and the method comprises the following steps:

[0052] Step S101, selecting a target privacy budget level from a plurality of privacy budget levels in a privacy budget list according to user settings;

[0053] Specifically, to achieve personalized privacy protection, the server provides m privacy budget levels, expressed as ε = ∪ m τ＝1 {ε τ}, where Δ∈=ε τ+1 -ε τ >0, that is, the smaller τ is, the greater ε is τ The smaller the value of , the higher the degree of privacy protection, but the lower the data availability. Users select the privacy budget ε that meets their privacy needs in the privacy budget list according to their own privacy needs. τ .

[0054] Step S102, performing biased sampling on the multidimensional mixed data according to the target privacy budget level to obtain a dimension index list, and determining the data type of each index in the dimension index list, wherein the data type includes categorical data and numerical data;

[0055] Specifically, according to the target privacy budget level selected by the user from the client, biased sampling is performed on the d-dimensional data, and each dimension in the obtained dimension index list index is determined to be numerical data or categorical data.

[0056] Step S103, performing corresponding perturbations on the categorical data and the numerical data respectively according to the data perturbation rule, and obtaining a perturbation vector of the categorical data and a perturbation value of the numerical data after the perturbation;

[0057] Specifically, the corresponding mechanism is used on the client to perturb its value, and the perturbed numerical data result R * i And the results of the classified data <C i * [j],τ> is sent to the server.

[0058] Step S104, performing aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively.

[0059] Specifically, the server directly sums the results of numerical data. Get the mean estimate of each attribute, and first group and count the results of the categorized data according to τ Finally, frequency estimates for each attribute are aggregated.

[0060] According to the above-mentioned multi-dimensional data personalized privacy protection method, users can choose the privacy budget level according to their own privacy needs, thereby avoiding excessive noise addition and improving the practicality of the data. The server side can pre-process the multi-dimensional mixed data and perform biased sampling on the dimensions of user data to obtain statistical data with higher accuracy. In addition, it can also perform attribute judgment on the sampling dimensions and execute corresponding personalized privacy protection algorithms for numerical data and categorical data respectively.

[0061] Embodiment 2

[0062] The multi-dimensional data personalized privacy protection method in the second embodiment of the present invention includes the following steps:

[0063] Step S11, selecting a target privacy budget level from a plurality of privacy budget levels in a privacy budget list according to user settings;

[0064] Step S12, performing biased sampling on the multidimensional mixed data according to the target privacy budget level to obtain a dimension index list, and determining the data type of each index in the dimension index list, wherein the data type includes categorical data and numerical data;

[0065] Step S13, performing corresponding perturbations on the categorical data and the numerical data respectively according to the data perturbation rule, and obtaining a perturbation vector of the categorical data and a perturbation value of the numerical data after the perturbation;

[0066] Step S14, performing aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively.

[0067] It should be explained that, for the parts not described in step S11, step S12, step S13 and step S14 in this embodiment, reference may be made to the first embodiment.

[0068] Furthermore, the step of performing biased sampling on the multi-dimensional mixed data according to the target privacy budget level to obtain a dimension index list specifically includes:

[0069] Dividing the multi-dimensional mixed data into a first dimension index group, a second dimension index group and a third dimension index group according to a preset classification rule;

[0070] The dimensional index list is obtained by performing biased sampling on the first dimensional index group and the second dimensional index group according to the target privacy budget level.

[0071] It should be explained that in existing work, in order to reduce the communication cost and computational cost between the client and the server, the server requires the user to uniformly sample its d-dimensional data. However, in actual applications, there may be repeated, statistically insignificant or insignificant attributes in the d-dimensional data, which will affect the availability of the collected data.

[0072] Therefore, the present invention designs a personalized sampling method PTop-k. The following process illustrates the personalized sampling process: First, the server divides the attribute labels into three groups, namely, Group1, Group2, and Group3, based on its own statistical preferences: the first dimension index group, the second dimension index group, and the third dimension index group. Group1 stores the attribute indexes that the server most wants to count, Group2 stores the attribute indexes that need to be counted, and Group3 stores the indexes that have no statistical significance or have repeated attribute meanings. The degree of biased sampling is controlled by setting the value of probability p in a personalized manner. Then, the indexes in Group1 are sampled k times without repetition with a probability of p, and the indexes in Group2 are sampled with a probability of 1-p, and finally the list Index of dimension indexes is returned.

[0073] Furthermore, the step of performing corresponding perturbations on the categorical data and the numerical data according to the data perturbation rule and obtaining the perturbed categorical data perturbation vector and the numerical data perturbation value specifically includes:

[0074] Calculating the numerical data disturbance value according to the first disturbance rule for the numerical data, and correcting the numerical data with a multiple of the numerical data disturbance value;

[0075] The classified data disturbance vector is calculated for the classified data according to the second disturbance rule, and the classified data is corrected by a multiple of the classified data disturbance vector.

[0076] Specifically, for multi-dimensional numerical data, a personalized perturbation method PPM is designed to satisfy LDP. For each numerical data index j, the PM method calculates the perturbation value according to its corresponding data T i [A j ] and the privacy budget divided by this dimension Calculate the disturbance value x j However, due to the personalized dimension sampling method, the

[0077]

[0078] x jCorrect the deviation caused by sampling, where len(Group) is the length of the group (Group1 or Group2) where the dimension index is located, pro is the probability that the index is sampled in its group (the corresponding probability of the index in Group1 is p, and the corresponding probability of the index in Group2 is 1-p), and k is the number of sampling dimensions.

[0079] For multi-dimensional categorical data, a personalized perturbation method POUE is designed to satisfy LDP. For each categorical data index j, the OUE method calculates the perturbation value according to its corresponding data T i [A j ] and the privacy budget divided by this dimension The perturbation vector Z is calculated. However, due to the use of a personalized dimensional sampling method, the deviation caused by sampling is corrected for the perturbation vector by the same multiple as in the numerical data.

[0080] Optionally, the personalized numerical data perturbation method PPM based on PM can also be based on other numerical data perturbation methods, such as Laplace mechanism or Gaussian mechanism, and the personalized classified data perturbation method POUE based on OUE can also be based on other classified data perturbation methods, such as RAPPOR mechanism or random matrix projection.

[0081] Furthermore, the step of performing aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively specifically includes:

[0082] Directly summing up each of the numerical data disturbance values ​​and then calculating the mean value to obtain an unbiased estimate of the statistical mean of the original numerical data;

[0083] After eliminating the noise of each group of the categorized data perturbation vectors corresponding to the target privacy budget level, the frequency values ​​of each group of the categorized data perturbation vectors are directly added to obtain an unbiased estimate of the original categorized data frequency.

[0084] Specifically, for numerical data, the goal of the server is to count the mean of each dimension. The server can directly add up the perturbation values ​​uploaded by all users and calculate the mean to obtain an unbiased estimate of the statistical mean of the original data. For categorical data, the goal of the server is to count the frequency of each category in each dimension. After eliminating the noise of the perturbation method based on the privacy budget value of each group, the server can directly add up the frequency values ​​of each group to obtain an unbiased estimate of the frequency of the original data.

[0085] For practical applications, please refer to Figure 3It shows the utility loss of the proposed model and other models in collecting numerical data under two different data sets. The two data sets are the Brazilian demographic data set (BR) and the Irish demographic data set (IR). It can be observed that the accuracy of the PPM mechanism results is higher than that of the PM mechanism results in all cases, especially when the privacy budget value is small. It can also be seen that Δ ε The larger the value of , the higher the accuracy of the PPM mechanism compared to the PM mechanism, that is, by providing users with a larger privacy budget that meets their privacy needs, higher data practicality can be obtained.

[0086] Figure 4 It represents the utility loss of the proposed model and other models in collecting classified data under two different data sets. It can be observed that the results of the POUE mechanism are more accurate than those of the OUE mechanism in all cases, especially when the privacy budget value is small. It can also be seen that Δ ε The larger the value of , the higher the accuracy of the POUE mechanism compared to the OUE mechanism, that is, by providing users with a larger privacy budget that meets their privacy needs, higher data practicality can be obtained.

[0087] Table 1 below shows two datasets: Brazilian demographic dataset (BR) and Irish demographic dataset (IR).

[0088]

[0089] Table 1

[0090] According to the above-mentioned multi-dimensional data personalized privacy protection method, users can choose the privacy budget level according to their own privacy needs, thereby avoiding excessive noise addition and improving the practicality of the data. The server side can pre-process the multi-dimensional mixed data and perform biased sampling on the dimensions of user data to obtain statistical data with higher accuracy. In addition, it can also perform attribute judgment on the sampling dimensions and execute corresponding personalized privacy protection algorithms for numerical data and categorical data respectively.

[0091] Embodiment 3

[0092] The present invention also provides a multi-dimensional data personalized privacy protection system, such as Figure 2 The system comprises:

[0093] A parameter selection module 10, configured to select a target privacy budget level from a plurality of privacy budget levels in a privacy budget list according to user settings;

[0094] A data sampling module 20, configured to perform biased sampling on multidimensional mixed data according to the target privacy budget level to obtain a dimension index list, and determine the data type of each index in the dimension index list, wherein the data type includes categorical data and numerical data;

[0095] The privacy protection module 30 is used to perform corresponding perturbations on the categorized data and the numerical data according to the data perturbation rule, and obtain the perturbation vector of the categorized data and the perturbation value of the numerical data after the perturbation;

[0096] The aggregation module 40 is used to perform aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively.

[0097] Preferably, the data sampling module 20 includes:

[0098] A data grouping unit, used for dividing the multi-dimensional mixed data into a first dimension index group, a second dimension index group and a third dimension index group according to a preset classification rule;

[0099] A data sampling unit is used to perform biased sampling on the first dimensional index group and the second dimensional index group according to the target privacy budget level to obtain the dimensional index list.

[0100] Preferably, the privacy protection module 30 includes:

[0101] A numerical data perturbation unit, used for calculating a numerical data perturbation value for the numerical data according to a first perturbation rule, and correcting the numerical data by a multiple of the numerical data perturbation value;

[0102] The classification data perturbation unit is used to calculate the classification data perturbation vector for the classification data according to the second perturbation rule, and correct it with a multiple of the classification data perturbation vector.

[0103] Preferably, the aggregation module 40 includes:

[0104] A numerical data aggregation unit, used for directly adding up each of the numerical data disturbance values ​​and then calculating the average value, so as to obtain an unbiased estimate of the statistical mean of the original numerical data;

[0105] The categorized data aggregation unit is used to eliminate the noise of each group of categorized data disturbance vectors corresponding to the target privacy budget level, and directly add the frequency values ​​of each group of categorized data disturbance vectors to obtain an unbiased estimate of the original categorized data frequency.

[0106] According to the above-mentioned multidimensional data personalized privacy protection system, users can choose the privacy budget level according to their own privacy needs, thereby avoiding excessive noise addition and improving the practicality of the data. The server side can pre-process the multidimensional mixed data and perform biased sampling on the dimensions of user data to obtain statistical data with higher accuracy. In addition, it can also perform attribute judgment on the sampling dimensions and execute corresponding personalized privacy protection algorithms for numerical data and categorical data respectively.

[0107] The present invention also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for personalized privacy protection of multi-dimensional data as described above is implemented.

[0108] The present invention also proposes a server, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned multi-dimensional data personalized privacy protection method when executing the program.

[0109] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0110] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for personalized privacy protection of multidimensional data, characterized in that: include: Selecting a target privacy budget level from a plurality of privacy budget levels in a privacy budget list according to user settings; According to the target privacy budget level, biased sampling is performed on the multidimensional mixed data to obtain a dimension index list, and the data type of each index in the dimension index list is determined, wherein the data type includes categorical data and numerical data. The biased sampling process is as follows: first, the server divides the attribute labels into three groups, namely, the first dimension index group, the second dimension index group and the third dimension index group, according to its own statistical preference. The first dimension index group stores the attribute indexes that the server most wants to count, the second dimension index group stores the attribute indexes that need to be counted, and the third dimension index group stores the indexes that have no statistical significance or have repeated attribute meanings. The biased sampling degree is controlled by setting the value of the probability p in a personalized manner, and then the indexes in the first dimension index group are sampled k times without repetition with a probability of p, and the indexes in the second dimension index group are sampled with a probability of 1-p. Finally, a list of dimension indexes is returned. According to the data perturbation rule, the categorical data and the numerical data are respectively perturbed accordingly, and a perturbation vector of the categorical data and a perturbation value of the numerical data after the perturbation are obtained; Performing aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively; The steps of performing corresponding perturbations on the categorical data and the numerical data according to the data perturbation rule and obtaining the perturbation vector of the categorical data and the perturbation value of the numerical data after the perturbation specifically include: The numerical data perturbation value is calculated for the numerical data according to the first perturbation rule, and the numerical data perturbation value is corrected by a multiple of the numerical data perturbation value. Specifically, for multidimensional numerical data, for each index of numerical data, its perturbation value x is calculated according to its corresponding data and the privacy budget divided by the dimension. j , in multiples x j Correct the deviation caused by sampling, where len(Group) is the length of the group in which the dimension index is located, and the group in which it is located is the first dimension index group or the second dimension index group, and pro is the probability that the index is sampled in its group, where the corresponding probability of the index in the first dimension index group is p, and the corresponding probability of the index in the second dimension index group is 1-p, and k is the number of sampling dimensions; The perturbation vector of the classified data is calculated for the classified data according to the second perturbation rule, and is corrected by multiples of the perturbation vector of the classified data. Specifically, for multi-dimensional classified data, for each index of the classified data, its perturbation vector is calculated according to its corresponding data and the privacy budget of the dimension, and the perturbation vector is corrected for deviations caused by sampling by the same multiples as in the numerical data.

2. The method for personalized privacy protection of multidimensional data according to claim 1, characterized in that: The step of performing aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively specifically includes: Directly summing up each of the numerical data disturbance values ​​and then calculating the mean value to obtain an unbiased estimate of the statistical mean of the original numerical data; After eliminating the noise of each group of the categorized data perturbation vectors corresponding to the target privacy budget level, the frequency values ​​of each group of the categorized data perturbation vectors are directly added to obtain an unbiased estimate of the original categorized data frequency.

3. A multidimensional data personalized privacy protection system, applicable to the multidimensional data personalized privacy protection method according to any one of claims 1-2, characterized in that: include: A parameter selection module, used for selecting a target privacy budget level from a plurality of privacy budget levels in a privacy budget list according to user settings; A data sampling module, used to perform biased sampling on multi-dimensional mixed data according to the target privacy budget level to obtain a dimension index list, and determine the data type of each index in the dimension index list, wherein the data type includes categorical data and numerical data; A privacy protection module, used to perform corresponding perturbations on the categorized data and the numerical data according to a data perturbation rule, and obtain a perturbation vector of the categorized data and a perturbation value of the numerical data after the perturbation; The aggregation module is used to perform aggregation operations on the categorical data disturbance vector and the numerical data disturbance value respectively.

4. The multi-dimensional data personalized privacy protection system according to claim 3, characterized in that: The data sampling module comprises: A data grouping unit, used for dividing the multi-dimensional mixed data into a first dimension index group, a second dimension index group and a third dimension index group according to a preset classification rule; A data sampling unit is used to perform biased sampling on the first dimensional index group and the second dimensional index group according to the target privacy budget level to obtain the dimensional index list.

5. The multi-dimensional data personalized privacy protection system according to claim 3, characterized in that: The privacy protection module includes: A numerical data perturbation unit, used for calculating a numerical data perturbation value for the numerical data according to a first perturbation rule, and correcting the numerical data by a multiple of the numerical data perturbation value; The classification data perturbation unit is used to calculate the classification data perturbation vector for the classification data according to the second perturbation rule, and correct it with a multiple of the classification data perturbation vector.

6. The multi-dimensional data personalized privacy protection system according to claim 3, characterized in that: The aggregation module includes: A numerical data aggregation unit, used for directly adding up each of the numerical data disturbance values ​​and then calculating the average value, so as to obtain an unbiased estimate of the statistical mean of the original numerical data; The categorized data aggregation unit is used to eliminate the noise of each group of categorized data disturbance vectors corresponding to the target privacy budget level, and directly add the frequency values ​​of each group of categorized data disturbance vectors to obtain an unbiased estimate of the original categorized data frequency.

7. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 2 is implemented.

Images