Pseudo data generating device, method and recording medium

Abstract

The pseudo-data generating device includes: a random number generating unit, which generates first pseudo-data in which the ratio of the frequency distribution of each attribute is approximately the same as the ratio of the frequency distribution of each attribute of the pseudo-object data; and a data shaping unit, which uses a matrix obtained by performing Cholesky decomposition on the variance-covariance matrix of the pseudo-object data or an enlarged and reduced matrix obtained by performing singular value decomposition on the variance-covariance matrix of the pseudo-object data to shape the first pseudo-data so that the average vector and correlation matrix of the first pseudo-data are consistent with the average vector and correlation matrix of the pseudo-object data, and the minimum value and maximum value of the first pseudo-data exist within the range of the minimum value and maximum value of the pseudo-object data, and the shaped first pseudo-data is set as the pseudo-data.

Description

Technical Field

[0001] The present invention relates to a pseudo data generating technology. Background Art

[0002] In the field of privacy protection, when original data containing personal information cannot be processed due to security issues, pseudo data is sometimes generated as a substitute for analysis. Here, we consider the case of creating pseudo data in tabular form from original data in tabular form. Examples of tabular data are Figure 1 shown.

[0003] As prior arts for creating pseudo data in a tabular form from original data in a tabular form, there are known non-patent documents 1 and 2. In these prior arts, in the case of a table in which the pseudo data to be created has only numerical attributes, pseudo data is generated by shaping a random array so that the properties between the attributes in the original data (variance covariance, correlation, mean vector, etc.) can be maintained.

[0004] Prior art literature

[0005] Non-patent literature

[0006] Non-patent document 1: Zhengli Huang, Wenliang Du, and Biao Chen. "Deriving private information from randomized data", In Proceedings of the 2005ACM SIGMODinternational conference on Management of data, pages 37-48. ACM, 2005.

[0007] Non-patent literature 2: Haoran Li, Li Xiong, and Xiaoqian Jiang. "Differentiallyprivate synthesization of multi-dimensional data using copula functions", InAdvances in database technology: procedures. International Conference onExtending Database Technology, Vol. 2014, p. 475. NIH Public Access, 2014. Summary of the invention

[0008] Problems to be solved by the invention

[0009] However, this conventional method can maintain the average vector or correlation, but does not maintain the frequency distribution or the maximum and minimum values ​​of each attribute. Therefore, there is a problem that when the data is visualized and analyzed, there is a large deviation between the original data and the pseudo data.

[0010] An object of the present invention is to provide a pseudo data generating device, a method thereof, and a program that can generate pseudo data that does not deviate greatly from original data even when the data is visualized and analyzed.

[0011] Means for solving problems

[0012] In order to solve the above-mentioned problems, according to one embodiment of the present invention, a pseudo data generating device includes: a random number generating unit, which generates first pseudo data in which the ratio of the frequency distribution of each attribute is approximately equal to the ratio of the frequency distribution of each attribute of the pseudo object data; and a data shaping unit, which uses a matrix obtained by performing Cholesky decomposition on the variance-covariance matrix of the pseudo object data or an enlarged and reduced matrix obtained by performing singular value decomposition on the variance-covariance matrix of the pseudo object data to shape the first pseudo data so that the mean vector and correlation matrix of the first pseudo data are consistent with the mean vector and correlation matrix of the pseudo object data, and the minimum value and maximum value of the first pseudo data exist within the range of the minimum value and maximum value of the pseudo object data, and the shaped first pseudo data is set as the pseudo data.

[0013] In order to solve the above-mentioned problems, according to another embodiment of the present invention, a pseudo data generation method is a pseudo data generation method performed by a pseudo data generation device, comprising: a random number generation step, generating first pseudo data in which the ratio of the frequency distribution of each attribute is approximately equal to the ratio of the frequency distribution of each attribute of the pseudo object data; and a data shaping step, using a matrix obtained by performing Cholesky decomposition on the variance-covariance matrix of the pseudo object data or an enlarged and reduced matrix obtained by performing singular value decomposition on the variance-covariance matrix of the pseudo object data, to shape the first pseudo data so that the average vector and correlation matrix of the first pseudo data are consistent with the average vector and correlation matrix of the pseudo object data, and the minimum value and maximum value of the first pseudo data exist within the range of the minimum value and maximum value of the pseudo object data, and the shaped first pseudo data is set as the pseudo data.

[0014] Effects of the Invention

[0015] According to the present invention, it is possible to generate pseudo data that does not deviate greatly from original data even when the data is visualized and analyzed. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1This is a diagram showing an example of data in a tabular format handled in the present specification.

[0017] Figure 2 This is a functional block diagram of the dummy data generating device according to the first embodiment.

[0018] Figure 3 This is a diagram showing an example of a process flow of the dummy data generating device according to the first embodiment. DETAILED DESCRIPTION

[0019] The following describes the embodiments of the present invention. In addition, in the drawings used in the following description, the same reference numerals are added to the components having the same functions or the steps performing the same processing, and repeated descriptions are omitted. In the following description, the processing performed on each element of a vector or a matrix is ​​assumed to be applied to all elements of the vector or the matrix unless otherwise specified.

[0020] <Key points of the first embodiment>

[0021] We focus on the matrix Q obtained by Cholesky decomposition of the variance-covariance matrix of the original data. By multiplying the matrix Q by the proportionality coefficient p, we can generate pseudo data that completely maintains the mean vector, correlation matrix, and frequency distribution approximation of the original data, and the data exists within the maximum / minimum range of each attribute.

[0022] <First embodiment>

[0023] Figure 2 A functional block diagram showing a pseudo data generating device according to the first embodiment is shown. Figure 3 Indicates its processing flow.

[0024] The pseudo data generating device according to the first embodiment includes a random number generating unit 210 and a data shaping unit 230 .

[0025] The pseudo data generating device is, for example, a special device that reads a special program into a known or dedicated computer having a central processing unit (CPU: Central Processing Unit), a main storage device (RAM: Random Access Memory, Random Access Memory), etc. The pseudo data generating device, for example, performs each process based on the control of the central processing unit. The data input to the pseudo data generating device or the data obtained in each process is, for example, stored in the main storage device, and the data stored in the main storage device is read out to the central processing unit as needed for other processes. Each processing unit of the pseudo data generating device can also be at least partially composed of hardware such as an integrated circuit. Each storage unit of the pseudo data generating device can be composed of, for example, a main storage device such as RAM (Random Access Memory), an auxiliary storage device composed of a semiconductor storage element such as a hard disk, an optical disk or a flash memory, or a middleware such as a relational database or a key value storage.

[0026] The pseudo data generating device according to the first embodiment generates and outputs pseudo data D' when the original data D and the number of records n' included in the generated pseudo data D' are input. n ' ×d Completely maintain the average vector μ of the original data D D , correlation matrix, the frequency distribution is approximate, and the data exist within the range of maximum / minimum values ​​of each attribute.

[0027] Will Figure 1 Such tabular data is regarded as a matrix, and the row vector is called a record, and the column vector is called an attribute. In addition, the number of rows of the original data D and the pseudo data D' is sometimes different, but the number of columns is the same. Let the number of rows of the original data D be n>0, let the number of rows of the pseudo data D' be n'>0, and let the number of columns of each be d>0. Let only the numerical values ​​that can be expressed in the real number space R be stored in the original data D and the pseudo data D'. In the following, let the list in which the average values ​​in each attribute are arranged in the same arrangement as the original data D be the average vector μ. For example, Figure 1 The average vector in the table data is μ = (43.66, 166, 60.66) T In addition, all vectors without a record of "row vector" are assumed to be "column vectors".

[0028] In addition, as numerical attributes applicable to this embodiment, date attributes are also included. When date attributes are applied to this embodiment, the date to be the object in the original database is converted into a continuous value such as m milliseconds before and m milliseconds after a specific date.

[0029] <Random number generation unit 210>

[0030] Input: original data D∈R n×d , the number of records generated n'

[0031] Output: The first pseudo data X∈R n ' ×d

[0032] The random number generator 210 generates and outputs first pseudo data X (S210) in which the frequency distribution ratio of each attribute is close to the frequency distribution ratio of each attribute of the original data D. In addition, the accuracy of the approximation is related to the number of records n' included in the pseudo data, and the larger n' is, the higher the accuracy of the approximation is likely to be.

[0033] For example, first, the random number generator 210 calculates the frequency distribution {h i} d i =1.

[0034] Next, the random number generator 210 approximates the frequency distribution of the i-th attribute of the first dummy data X to the frequency distribution h of the original data D. i The i-th column vector is randomly generated in a ratio of . This operation is repeated until i increases from 1 to d. In addition, various well-known techniques can be used in the method of generating column vectors. For example, the discard method or inverse function method known in reference 1 can be used.

[0035] (Reference 1) Wakimoto Kazumasa, "Knowledge of Random Numbers", Morihoku Publishing Co., Ltd., 1970, p.61-71

[0036] The random number generator 210 arranges the generated d column vectors in the same arrangement as the original data D, and outputs the first dummy data X in a table format.

[0037] <Data shaping unit 230>

[0038] Input: original data D∈R n×d , the first pseudo data X∈R n ' ×d

[0039] Output: pseudo data D'∈R n ' ×d

[0040] The data shaping unit 230 uses a matrix obtained by performing Cholesky decomposition on the variance-covariance matrix of the original data D to shape the first pseudo data X so that the mean vector μ and the correlation matrix of the first pseudo data X are equal to the mean vector μ of the original data D. D The correlation matrix is ​​consistent, and the minimum and maximum values ​​of the first dummy data X are within the range of the minimum and maximum values ​​of the original data D ( S230 ), and the shaped first dummy data is output as dummy data D′.

[0041] For example, the first dummy data X is shaped through the following processes 1. to 11.

[0042] 1. Let variable p = 1. Variable p is also called proportionality coefficient.

[0043] 2. Set the range of possible values ​​of each i-th attribute in the original data D to R (i) D This operation is repeated until i increases from 1 to d.

[0044] 3. Calculate the mean vector μ and variance-covariance matrix Σ of the first pseudo data X.

[0045] 4. Calculate the condition that satisfies Σ=QQ T In addition, as the calculation of Q, Q T A method such as Cholesky decomposition can be used.

[0046] 5. Calculate r = Q for all r ∈ X -1 (r-μ). That is, by r=Q -1 (r-μ) is used to update the record r included in the first dummy data X.

[0047] 6. All the records r∈R calculated above d In general, let Z'∈R n'×d .

[0048] 7. Calculate the average vector μ of the original data D D and the variance-covariance matrix Σ D .

[0049] 8 Calculate the value that satisfies Σ D =Q D Q D T Q D This process is equivalent to converting the variance-covariance matrix Σ of the original data D into D Perform Cholesky decomposition, Q D The variance-covariance matrix Σ D The matrix obtained by performing Cholesky decomposition.

[0050] 9. Calculate Y = X (p·Q D ) T +I diag(μ D ). Let I here be the identity matrix, diag(μ D ) is the mean vector μ in the diagonal component D The elements of p are all diagonal matrices with all other components equal to 0. 2 ·Σ D =p 2 Q D Q D T =(p·Q D )(p·Q D ) T It is called the correlation matrix, and p when p = 1 is also called the correlation matrix. 2 ·Σ D =p 2 Q D Q D T =(p·Q D )(p·Q D ) T It is called the variance-covariance matrix. By calculating X(p·Q D ) T At the same time, the pseudo data X can be shaped so that the correlation matrix of the first pseudo data X is consistent with the correlation matrix of the original data D. In addition, by calculating I diag (μ D ) and generates dummy data at the same time, the first dummy data X can be shaped so that the average vector of the first dummy data X is consistent with the average vector of the original data D.

[0051] 10. Set the range of possible values ​​of each i-th attribute in the above Y to R (i) Y This operation is repeated until i increases from 1 to d.

[0052] 11. If for all i, R (i) Y Not in R (i) D If the value is outside the range of R, Y is output as pseudo data D'. Y (i) In R D (i) When the range is within , Y is output as pseudo data D'. (i) Y In R (i) DIf it is out of the range, set p = p / 2 and return to the above 9. In addition, the update formula of the variable p can be smaller than the current p in the range of the entire set R+ of non-negative real numbers. In other words, set α to an arbitrary real number greater than 1 and set p = p / α.

[0053] <Effect>

[0054] According to the above structure, it is possible to generate pseudo data D' that completely maintains the average vector, correlation matrix, and frequency distribution approximation of the original data D, and the data exists in the range of the maximum / minimum value of each attribute. Thus, the generated pseudo data D' can completely maintain the average vector and correlation matrix of the original data D, so that a linear regression model that is exactly the same as the original data D can be obtained. In particular, when the range of possible values ​​of each attribute in the original data D is the same, the frequency distribution and the approximation of the maximum and minimum values ​​of each attribute of the original data D can be maintained, so when the data is visualized and analyzed, pseudo data D' that does not deviate greatly from the original data D can also be generated. For example, it is possible to approximate the frequency distribution of each attribute of the original data D without generating a record with a profile such as a height of -170cm.

[0055] <Modification>

[0056] In this embodiment, when there is no original data D but data to be reproduced (pseudo object data) exists, it is not limited to the original data D, and the statistical values ​​of the pseudo object data (frequency distribution of each attribute, mean vector, variance covariance matrix, range of possible values ​​of each attribute, etc.) may be input. It can also be said that the original data D is an example of pseudo object data.

[0057] In this embodiment, the frequency distribution, mean vector, variance-covariance matrix, and the range of possible values ​​(maximum value / minimum value) of each attribute of the original data D are calculated within each unit. However, they may be calculated in advance outside each unit and set as inputs to the random number generation unit 210 and the data shaping unit 230, thereby omitting the calculation within each unit.

[0058] In the present embodiment, the first pseudo data X is shaped using Cholesky decomposition in the data shaping unit 230, but the first pseudo data X may be shaped using singular value decomposition. This processing example is described. For example, the same processing as in the present embodiment is performed for processing 1. to 7. and processing 10. and 11., and the following processing is performed for processing 8. and 9.

[0059] 8. Calculate the value that satisfies Σ D =U D Λ D U T D U D ,ΛD In addition, U D is a matrix that collects eigenvectors. ΛD is a diagonal matrix that sets eigenvalues ​​as diagonal components and other components as 0, also known as an enlargement and reduction matrix. This process is equivalent to converting the variance-covariance matrix Σ of the original data D into D Perform singular value decomposition and find U D , expansion and reduction matrix Λ D Further, calculate Λ D 1 / 2 . D 1 / 2 is to expand and reduce the matrix Λ D A matrix in which all diagonal components of are squared. D Equivalent to U in this modification D Λ D 1 / 2 .

[0060] 9. Calculate Y = X (p·U D Λ D 1 / 2 ) T +I diag(μ D ). D =p 2 ·U D Λ D U D T=(p·U D Λ D 1 / 2 )(p·U D Λ D 1 / 2 ) T It is called the correlation matrix, and Σ = p when p = 1 2 ·U D Λ D U D T =(p·U D Λ D 1 / 2 )(p·U D Λ D 1 / 2 ) T It is called the variance-covariance matrix. By calculating X(p·U D Λ D 1 / 2 )T, and simultaneously generates pseudo data, and can reshape the first pseudo data X so that the correlation matrix of the first pseudo data X is consistent with the correlation matrix of the original data D.

[0061] In addition, in the process 4., similarly, singular value decomposition may be used instead of Cholesky decomposition. That is, the following process 4. is performed.

[0062] 4. Calculate to satisfy Σ=UΛU T U and Λ. In addition, U is a matrix that collects eigenvectors. Λ is a diagonal matrix that sets eigenvalues ​​to diagonal components and other components to zero, also called an enlargement and reduction matrix. This process is equivalent to performing singular value decomposition on the variance-covariance matrix Σ of the first pseudo data and calculating U and the enlargement and reduction matrix Λ. Further, calculate Λ 1 / 2 . 1 / 2 It is a matrix obtained by squaring all the diagonal components of the enlargement and reduction matrix Λ. After processing 4., let Q = UΛ 1 / 2 And process it.

[0063] Theoretically, the variance-covariance matrix is ​​a positive definite matrix, so it can be calculated by Cholesky decomposition to become Σ = QQ T ,Σ D =Q D Q D T Q, Q D However, in numerical calculations using computers, when the number of records n of the original data or the number of records n' of X is small, Q and Q D is unstable and often cannot be calculated. Therefore, by not directly using the Cholesky decomposition to find Q and Q D , but by singular value decomposition to calculate U and Λ, U D and Λ D , we can calculate Q = UΛ 1 / 2 , Q D =U D Λ D 1 / 2 .

[0064] Through the above-mentioned processing, the same effects as those of the first embodiment can be obtained.

[0065] <Other Modifications>

[0066] The present invention is not limited to the above-mentioned embodiments and variations. For example, the above-mentioned various processes are not limited to being performed sequentially according to the records, but can also be performed in parallel or individually according to the processing capacity or needs of the device performing the processing. In addition, appropriate changes can be made within the scope of the main purpose of the present invention.

[0067] <Program and recording medium>

[0068] In addition, the various processing functions in each device described in the above-mentioned embodiments and modified examples can also be realized by a computer. In this case, the processing content of the functions that each device should have is described by a program. Then, by executing the program by a computer, the various processing functions in the above-mentioned devices are realized on the computer.

[0069] The program describing the processing contents can be recorded in advance in a computer-readable recording medium, such as a magnetic recording device, an optical disk, a magneto-optical recording medium, or a semiconductor memory.

[0070] In addition, the program can be circulated, for example, by selling, transferring, or lending a removable recording medium such as a DVD or CD-ROM on which the program is recorded. Furthermore, the program can be circulated by pre-storing the program in a storage device of a server computer and forwarding the program from the server computer to other computers via a network.

[0071] The computer executing such a program, for example, first temporarily stores the program recorded in a portable recording medium or the program forwarded from a server computer in its own storage unit. Then, when the processing is executed, the computer reads the program stored in its own storage unit and executes the processing according to the read program. In addition, as another embodiment of the program, the computer can directly read the program from the portable recording medium and execute the processing according to the program. Further, it can also be that each time the program is forwarded from the server computer to the computer, the processing according to the received program is executed one by one. In addition, it can also be configured to perform the above-mentioned processing by not forwarding the program from the server computer to the computer, and only by its execution instructions and result acquisition to realize the processing function of the so-called ASP (Application Service Provider) type service. In addition, it is assumed that the program contains content based on the program (not a direct instruction to the computer but data with the nature of specifying the processing of the computer, etc.) provided as information for the processing of the electronic computer.

[0072] Furthermore, although each device is configured by executing a predetermined program on a computer, at least a part of the processing contents may be realized in the form of hardware.

Claims

1. A pseudo personal information data generating device, for the purpose of data security, generates pseudo personal information data as a substitute without processing original data containing personal information, the pseudo personal information data generating device comprising: a random number generating unit for generating first pseudo personal information data in which the frequency distribution ratio of each personal attribute is close to the frequency distribution ratio of each personal attribute of the pseudo target personal information data; and The data shaping unit, when the variance-covariance matrix of the pseudo-object personal information data is a positive definite matrix, uses a matrix obtained by performing Cholesky decomposition on the variance-covariance matrix to shape the first pseudo-personal information data so that the average vector and the correlation matrix of the first pseudo-personal information data are consistent with the average vector and the correlation matrix of the pseudo-object personal information data, and the minimum value and the maximum value of the first pseudo-personal information data exist within the range of the minimum value and the maximum value of each personal attribute of the pseudo-object personal information data, and sets the shaped first pseudo-personal information data as the pseudo-personal information data, When visual analysis is performed on the data based on the pseudo personal information data generated by the data shaping unit, a linear regression model identical to the original personal information data is obtained. In the data shaping part, let α be any real number greater than 1, let matrix I be a matrix of n' rows and d columns with all 1s, Q D Let μ be a matrix with d rows and columns. D Set it as a vector consisting of d elements, X as the first pseudo personal information data, variable p as the proportional coefficient, find the average vector μ and variance covariance matrix Σ of the first pseudo personal information data, use the matrix Q calculated based on the variance covariance matrix Σ, through r = Q -1 (r-μ) Update the record r included in the first pseudo personal information data, and set the average vector and variance covariance matrix of the pseudo object personal information data to μ respectively. D and Σ D , calculate the value that satisfies Σ D =Q D Q D T Q D , calculate Y = X (p·Q D ) T +I diag(μ D ), the range of the attributes of each i-th person in Y is R Y (i) The acceptable range R of each i-th personal attribute in the pseudo-object personal information data D (i) If it is out of the range, set p = p / α and recalculate Y = X (p·Q D ) T +I diag(μ D ), in all the ranges R Y (i) In the range R D (i) When Y is within the range of the pseudo personal information data, The data shaping unit calculates the value of Σ=QQ by Cholesky decomposition. T Q.

2. A pseudo-personal information data generating device, For data security, without processing the original data containing personal information, pseudo personal information data is generated as a substitute, and the pseudo personal information data generating device includes: a random number generating unit for generating first pseudo personal information data in which the frequency distribution ratio of each personal attribute is close to the frequency distribution ratio of each personal attribute of the pseudo target personal information data; and The data shaping unit, when the variance-covariance matrix of the pseudo-object personal information data is a non-positive definite matrix, uses the enlarged and reduced matrix obtained by performing singular value decomposition on the variance-covariance matrix to shape the first pseudo-personal information data so that the average vector and the correlation matrix of the first pseudo-personal information data are consistent with the average vector and the correlation matrix of the pseudo-object personal information data, and the minimum value and the maximum value of the first pseudo-personal information data exist within the range of the minimum value and the maximum value of each personal attribute of the pseudo-object personal information data, and the shaped first pseudo-personal information data is set as the pseudo-personal information data, When visual analysis is performed on the data based on the pseudo personal information data generated by the data shaping unit, a linear regression model identical to the original personal information data is obtained. In the data shaping part, let α be any real number greater than 1, let matrix I be a matrix of n' rows and d columns with all 1s, μ D Let U be a vector of d elements, D Let it be a matrix with d rows and d columns, Λ D Set it as a matrix with d rows and d columns, set X as the first pseudo personal information data, set the variable p as the proportional coefficient, find the average vector μ and variance covariance matrix Σ of the first pseudo personal information data, use the matrix Q calculated based on the variance covariance matrix Σ, and pass r = Q -1 (r-μ) Update the record r included in the first pseudo personal information data, and set the average vector and variance covariance matrix of the pseudo object personal information data to μ respectively. D and Σ D , calculate the value that satisfies Σ D =U D Λ D U D T U D And Λ D , calculate Y = X (p·U D Λ D 1 / 2 ) T +Idiag(μ D ), the range of the attributes of each i-th person in Y is R Y (i) The acceptable range R of each i-th personal attribute in the pseudo-object personal information data D (i) If it is out of the range, set p = p / α and recalculate Y = X (p·U D Λ D 1 / 2 ) T +I diag(μ D ), in all the ranges R Y (i) In the range R D (i) When Y is set to the pseudo personal information data, the data shaping unit calculates the value satisfying Σ=UΛU by singular value decomposition. T U, Λ, let Q = UΛ 1 / 2 .

3. A method for generating pseudo personal information data, which is performed by a pseudo personal information data generating device for generating pseudo personal information data as a substitute without processing original data containing personal information for data security, the method comprising: A random number generation step of generating first pseudo personal information data in which the frequency distribution ratio of each personal attribute is close to the frequency distribution ratio of each personal attribute of the pseudo object personal information data; and a data shaping step, in which, when the variance-covariance matrix of the pseudo-object personal information data is a positive definite matrix, the first pseudo-personal information data is shaped using a matrix obtained by performing Cholesky decomposition on the variance-covariance matrix, so that the average vector and the correlation matrix of the first pseudo-personal information data are consistent with the average vector and the correlation matrix of the pseudo-object personal information data, and the minimum value and the maximum value of the first pseudo-personal information data exist within the range of the minimum value and the maximum value of each personal attribute of the pseudo-object personal information data, and the shaped first pseudo-personal information data is set as the pseudo-personal information data, When the pseudo personal information data generated in the data shaping step is used to perform a visual analysis on the data, a linear regression model identical to the original personal information data is obtained. In the data shaping step, let α be any real number greater than 1, let matrix I be a matrix of n' rows and d columns with all 1s, Q D Let μ be a matrix with d rows and columns. D Set it as a vector consisting of d elements, X as the first pseudo personal information data, variable p as the proportional coefficient, find the average vector μ and variance covariance matrix Σ of the first pseudo personal information data, use the matrix Q calculated based on the variance covariance matrix Σ, through r = Q -1 (r-μ) Update the record r included in the first pseudo personal information data, and set the average vector and variance covariance matrix of the pseudo object personal information data to μ respectively. D and Σ D , calculate the value that satisfies Σ D =Q D Q D T Q D , calculate Y = X (p·Q D ) T +I diag(μ D ), the range of the attributes of each i-th person in Y is R Y (i) The acceptable range R of each i-th personal attribute in the pseudo-object personal information data D (i) If it is out of the range, set p = p / α and recalculate Y = X (p·Q D ) T +I diag(μ D ), in all the ranges R Y (i) In the range R D (i) When Y is within the range of the pseudo personal information data, The data shaping step satisfies Σ=QQ by calculating the Cholesky decomposition T Q.

4. A method for generating pseudo personal information data, which is performed by a pseudo personal information data generating device for generating pseudo personal information data as a substitute without processing original data containing personal information for data security, the method comprising: A random number generation step of generating first pseudo personal information data in which the frequency distribution ratio of each personal attribute is close to the frequency distribution ratio of each personal attribute of the pseudo object personal information data; and a data shaping step, in which, when the variance-covariance matrix of the pseudo-object personal information data is a non-positive definite matrix, the first pseudo-personal information data is shaped using an enlarged and reduced matrix obtained by performing singular value decomposition on the variance-covariance matrix, so that the average vector and the correlation matrix of the first pseudo-personal information data are consistent with the average vector and the correlation matrix of the pseudo-object personal information data, and the minimum value and the maximum value of the first pseudo-personal information data exist within the range of the minimum value and the maximum value of each personal attribute of the pseudo-object personal information data, and the shaped first pseudo-personal information data is set as the pseudo-personal information data, When the pseudo personal information data generated by the data shaping step is used to perform a visual analysis on the data, a linear regression model identical to the original personal information data is obtained. In the data shaping step, let α be any real number greater than 1, let matrix I be a matrix of n' rows and d columns with all 1s, μ D Let U be a vector of d elements, D Let it be a matrix with d rows and d columns, Λ D Set it as a matrix with d rows and d columns, set X as the first pseudo personal information data, set the variable p as the proportional coefficient, find the average vector μ and variance covariance matrix Σ of the first pseudo personal information data, use the matrix Q calculated based on the variance covariance matrix Σ, and pass r = Q -1 (r-μ) Update the record r included in the first pseudo personal information data, and set the average vector and variance covariance matrix of the pseudo object personal information data to μ respectively. D and Σ D , calculate the value that satisfies Σ D =U D Λ D U D T U D And Λ D , calculate Y = X (p·U D Λ D 1 / 2 ) T +Idiag(μ D ), the range of the attributes of each i-th person in Y is R Y (i) The acceptable range R of each i-th personal attribute in the pseudo-object personal information data D (i) If it is out of the range, set p = p / α and recalculate Y = X (p·U D Λ D 1 / 2 ) T +I diag(μ D ), in all the ranges R Y (i) In the range R D (i) When Y is within the range of the pseudo personal information data, The data shaping step calculates the value that satisfies Σ=UΛU by singular value decomposition. T U, Λ, let Q = UΛ 1 / 2 .

5. A computer-readable recording medium recording a program for causing a computer to function as the pseudo personal information data generating device according to claim 1.

6. A computer-readable recording medium recording a program for causing a computer to function as the pseudo personal information data generating device according to claim 2.

Images