Private data anonymization protection method

Abstract

The invention relates to the technical field of privacy protection, and discloses a privacy data anonymization protection method, which comprises the following steps: carrying out data desensitization processing on original privacy data to obtain desensitized privacy data, carrying out differential privacy noise addition processing on the desensitized privacy data to obtain noise-added privacy data, and carrying out differential privacy noise addition processing on the noise-added privacy data to obtain the privacy data. Adding dynamic noise to the desensitized privacy data by adopting a Laplace mechanism to obtain privacy data with a differential privacy protection characteristic; carrying out confusion on the privacy data attributes after noise addition to obtain confused privacy data; and performing data fuzzy processing on the confused privacy data obtained by querying the query statement, limiting the accuracy of the query result, performing pseudo data generation on the privacy data of which the privacy data items of the query result are less than the specified threshold, and returning the query result and the generated pseudo data as the query result together, thereby avoiding the leakage of the differential privacy data.

Description

Technical Field

[0001] The present invention relates to the technical field of privacy protection, and in particular to a method for anonymizing and protecting privacy data. Background Art

[0002] With the rapid development of the Internet and information technology, the collection and use of personal data has become the norm in modern society. From social media, e-commerce to medical health and other fields, an increasing amount of data involves sensitive data such as users' personal information and behavioral habits. However, this is accompanied by concerns about privacy leakage. Many users are worried that their personal privacy may be abused, and even lead to damage to personal interests and security risks. When personal sensitive information is leaked, it may cause a series of problems, such as identity theft, financial loss, damage to reputation, and threats to personal safety. In addition, privacy data leakage may also lead to large-scale data abuse, which will have a serious impact on the entire society. In response to this problem, the present invention proposes a privacy data anonymization protection method, which can reduce the risk of malicious use of personal sensitive information through privacy computing de-identification processing. Summary of the invention

[0003] In view of this, the present invention provides a privacy data anonymization protection method, the purpose of which is to: 1) propose a privacy data anonymization protection method, which combines multiple privacy protection technologies such as data desensitization, differential privacy, data obfuscation and data fuzzification, and improves the protection level of privacy data through multi-level processing means; 2) adopts an improved recursive segmentation K anonymization operation in data desensitization processing, which can more effectively ensure that the desensitized privacy data has certain privacy protection characteristics and improve the security of the data; 3) introduces differential privacy noise processing, and ensures the differential privacy protection characteristics of the privacy data by adding dynamic noise to the desensitized privacy data, thereby improving the privacy protection level of the data; 4) for the situation where the privacy data entries of the query result are less than the specified threshold, data fuzzification processing is adopted, and by generating pseudo data and returning it together with the query result, the leakage of differential privacy data is avoided, thereby improving the privacy protection capability of the data.

[0004] To achieve the above object, the present invention provides a privacy data anonymization protection method, comprising the following steps:

[0005] S1: Desensitizing the original private data to obtain desensitized private data, ensuring that any desensitized private data entry is indistinguishable from at least K sensitive private data entries in the desensitized private data, wherein the improved recursive segmentation K anonymization operation is the implementation method of the data desensitization processing;

[0006] S2: Perform differential privacy noise processing on the desensitized private data to obtain noisy private data. Use the Laplace mechanism to add dynamic noise to the desensitized private data to obtain private data with differential privacy protection characteristics.

[0007] S3: Obfuscating the noisy private data attributes to obtain obfuscated private data, wherein the data obfuscation technology is an implementation method of the obfuscation;

[0008] S4: Perform data fuzzification on the obfuscated private data obtained by querying the query statement to limit the accuracy of the query result. Generate pseudo data for the private data entries in the query result that are less than the specified threshold, and return the query result and the generated pseudo data together as the query result to avoid differential privacy data leakage.

[0009] As a further improvement method of the present invention:

[0010] Optionally, performing data desensitization processing on the original private data to obtain desensitized private data in step S1 includes:

[0011] S11: Clean the original private data, remove outliers and missing values, and determine the sensitive attributes and non-sensitive attributes in the original private data;

[0012] S12: Use a recursive partitioning method to partition the cleaned original privacy data into different clusters until each cluster contains at least K records. Each time the partition is performed, the attribute that can best improve anonymity is selected for partitioning. Information gain is the basis for determining the attribute that improves anonymity, and the calculation formula is:

[0013] Gain(A)=H(D)―H(D|A)

[0014] in:

[0015] H(D) represents the information entropy of data set D;

[0016] H(D|A) represents the conditional entropy of data set D under the condition of selecting attribute A;

[0017] Gain(A) represents the information gain of attribute A;

[0018] S13: Generalize the attributes of the divided data clusters to ensure the anonymity and privacy protection of the data. The attribute generalization should maintain the diversity of the data, that is, select attributes that do not affect the diversity of the data for generalization processing. The calculation formula for the data diversity is:

[0019]

[0020] Where: Div(Q) represents the diversity of attribute Q;

[0021] S(q) represents the data subset whose attribute Q has a sensitive value of q;

[0022] |Q| represents the number of values ​​of attribute Q;

[0023] n is the number of data records;

[0024] S14: Generalize the original private data according to the generalized attribute to obtain the desensitized original private data.

[0025] Optionally, in step S2, differential privacy noise addition processing is performed on the desensitized privacy data, including:

[0026] S21: Determine the privacy budget parameter ε in differential privacy according to the sensitivity of the attribute data, where the data sensitivity is divided into four levels: {extremely sensitive, sensitive, generally sensitive, insensitive}, corresponding to values ​​{0.001, 0.1, 0.5, 0.99} respectively;

[0027] S22: Noise processing is performed on each desensitized attribute privacy data according to the privacy budget parameter, wherein dynamic noise generation based on the Laplace mechanism is the main implementation method of the noise processing, and the dynamic noise calculation formula based on the Laplace mechanism is:

[0028]

[0029] in:

[0030] Δf represents the sensitivity of the query function, and its value range is {0.01, 0.99}, indicating sensitive and insensitive respectively;

[0031] ε represents the privacy budget;

[0032] Lap() represents Laplace distribution;

[0033] S23: Add the generated dynamic noise to the desensitized private data to obtain the noisy private data.

[0034] Optionally, in step S3, the attributes of the noisy private data are obfuscated to improve the anti-interference property of the data to obtain obfuscated private data, including:

[0035] S31: generating a random vector according to the privacy attribute data after adding noise, wherein generating the random vector includes calculating the sensitivity of the privacy attribute data, and generating a random vector according to the privacy attribute data after adding noise and the sensitivity information, wherein the random vector elements obey Laplace distribution;

[0036] S32: Establish a confusion matrix using a polynomial function according to the generated random vector;

[0037] S33: performing a confusion operation on the noisy privacy attribute data according to the generated random vector and confusion matrix to obtain the obfuscated privacy attribute data. The calculation formula of the confusion operation is:

[0038] x′=xVMV

[0039] in:

[0040] x′ represents the obfuscated privacy attribute data;

[0041] x′ represents the privacy attribute data after noise addition;

[0042] V represents the generated random vector;

[0043] M represents the confusion matrix;

[0044] T represents the transpose operation.

[0045] In the step S31, generating a random vector according to the privacy attribute data after adding noise includes:

[0046] S311: Calculate the sensitivity of the privacy attribute data. For the privacy attribute data x, the sensitivity calculation formula is:

[0047]

[0048] in:

[0049] f(x) represents the sensitivity function;

[0050] neighbors(x,x ′ ) represents the set of data points in the neighborhood of data x;

[0051] x ′ Represents any data point variable within the neighborhood;

[0052] S312: Perform Laplace sampling according to the calculated sensitivity to generate a random vector, wherein the random vector element sampling calculation formula is:

[0053]

[0054] in:

[0055] b represents the scale parameter of the Laplace distribution, which controls the peak degree of the distribution;

[0056] x represents privacy attribute data;

[0057] p(x) represents the sampling result.

[0058] In the step S32, a confusion matrix is ​​established using a polynomial function according to the generated random vector V, including:

[0059] S321: According to the preset polynomial function f(x)=a n x n +a n―1 x n―1 +...+a1x+a0, taking each element in the random vector as input and the corresponding privacy attribute data after noise addition as output, constructing a system of equations;

[0060] in:

[0061] a n 、a n―1 , a1, a0 represent polynomial coefficients;

[0062] x represents the input variable;

[0063] S322: jointly solving the constructed equation group to obtain optimal polynomial function coefficients, and concatenating the polynomial function coefficients into a coefficient vector P;

[0064] S323: Perform outer product calculation on the random vector V and the coefficient vector to obtain a confusion matrix.

[0065] Optionally, in step S4, generating pseudo data for the privacy data of the query result whose privacy data entries are less than a specified threshold and returning the query result and the generated pseudo data together as the query result includes:

[0066] S41: construct a privacy data generation model, the model includes a generation network and a discriminant network, the generation network is responsible for generating pseudo data, and the discriminant network is responsible for distinguishing true privacy data from pseudo data;

[0067] S42: Training the constructed private data generation model, wherein the generation network receives a random vector as input and outputs pseudo data similar to the real private data, and the discriminant network receives the real private data or the pseudo data generated by the generation network for distinction, and the optimal generation network and the optimal discriminant network are obtained by alternately training the objective function of the private data generation model. The objective function calculation formula of the private data generation model is:

[0068]

[0069] in:

[0070] G represents the generation network;

[0071] D represents the discriminant network;

[0072] x represents real private data;

[0073] z represents the value from the prior distribution p z Random noise sampled in (z);

[0074] p data (x) represents the real private data distribution;

[0075] S43: Generate pseudo data using the optimal generative network.

[0076] In order to solve the above problem, the present invention provides an electronic device, the electronic device comprising:

[0077] A memory storing at least one instruction;

[0078] Communication interface, enabling electronic equipment to communicate; and

[0079] The processor executes the instructions stored in the memory to implement the above-mentioned privacy data anonymization protection method.

[0080] In order to solve the above problem, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned privacy data anonymization protection method.

[0081] Compared with the prior art, the present invention proposes a privacy data anonymization protection method, which has the following advantages:

[0082] First, this scheme proposes a privacy data anonymization protection method, which combines multiple privacy protection technologies such as data desensitization, differential privacy, data obfuscation and data blurring, and improves the protection level of privacy data through multi-level processing methods; and adopts an improved recursive segmentation K anonymization operation in data desensitization processing, which can more effectively ensure that the desensitized privacy data has certain privacy protection characteristics and improve the security of the data;

[0083] At the same time, this solution introduces differential privacy noise processing, which ensures the differential privacy protection characteristics of private data by adding dynamic noise to the desensitized private data, improves the privacy protection level of the data, and adopts data fuzzification processing when the private data entries of the query results are less than the specified threshold. By generating pseudo data and returning it together with the query results, the leakage of differential privacy data is avoided and the privacy protection ability of the data is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A flowchart of a privacy data anonymization protection method provided by an embodiment of the present invention;

[0085] Figure 2 A schematic diagram of the structure of an electronic device for implementing a method for protecting privacy data anonymization provided by an embodiment of the present invention.

[0086] In the figure: 1 electronic device, 10 processor, 11 memory, 12 program, 13 communication interface.

[0087] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0088] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0089] The embodiment of the present application provides a privacy data anonymization protection method. The execution subject of the privacy data anonymization protection method includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the privacy data anonymization protection method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0090] Embodiment 1:

[0091] S1: Desensitize the original private data to obtain desensitized private data, ensuring that any desensitized private data entry is indistinguishable from at least K sensitive private data entries in the desensitized private data.

[0092] In step S1, the original private data is desensitized to obtain desensitized private data, including:

[0093] S11: Clean the original private data, remove outliers and missing values, and determine the sensitive attributes and non-sensitive attributes in the original private data;

[0094] S12: Use a recursive partitioning method to partition the cleaned original privacy data into different clusters until each cluster contains at least K records. Each time the partition is performed, the attribute that can best improve anonymity is selected for partitioning. Information gain is the basis for determining the attribute that improves anonymity, and the calculation formula is:

[0095] Gain(A)=H(D)―H(D|A)

[0096] in:

[0097] H(D) represents the information entropy of data set D;

[0098] H(D|A) represents the conditional entropy of data set D under the condition of selecting attribute A;

[0099] Gain(A) represents the information gain of attribute A;

[0100] S13: Generalize the attributes of the divided data clusters to ensure the anonymity and privacy protection of the data. The attribute generalization should maintain the diversity of the data, that is, select attributes that do not affect the diversity of the data for generalization processing. The calculation formula for the data diversity is:

[0101]

[0102] Where: Div(Q) represents the diversity of attribute Q;

[0103] S(q) represents the data subset whose attribute Q has a sensitive value of q;

[0104] |Q| represents the number of values ​​of attribute Q;

[0105] n is the number of data records;

[0106] S14: Generalize the original private data according to the generalized attribute to obtain the desensitized original private data.

[0107] S2: Perform differential privacy noise processing on the desensitized privacy data to obtain noisy privacy data, and use the Laplace mechanism to add dynamic noise to the desensitized privacy data to obtain privacy data with differential privacy protection characteristics.

[0108] In step S2, differential privacy noise addition processing is performed on the desensitized privacy data, including:

[0109] S21: Determine the privacy budget parameter ε in differential privacy according to the sensitivity of the attribute data, where the data sensitivity is divided into four levels: {extremely sensitive, sensitive, generally sensitive, insensitive}, corresponding to values ​​{0.001, 0.1, 0.5, 0.99} respectively;

[0110] S22: Noise processing is performed on each desensitized attribute privacy data according to the privacy budget parameter, wherein dynamic noise generation based on the Laplace mechanism is the main implementation method of the noise processing, and the dynamic noise calculation formula based on the Laplace mechanism is:

[0111]

[0112] in:

[0113] Δf represents the sensitivity of the query function, and its value range is {0.01, 0.99}, indicating sensitive and insensitive respectively;

[0114] ε represents the privacy budget;

[0115] Lap() represents Laplace distribution;

[0116] S23: Add the generated dynamic noise to the desensitized private data to obtain the noisy private data.

[0117] S3: Obfuscate the noisy private data attributes to obtain obfuscated private data, wherein the data scrambling technology is an implementation method of the obfuscation.

[0118] In the step S3, the attributes of the noisy private data are obfuscated to improve the anti-interference property of the data and obtain the obfuscated private data, including:

[0119] S31: generating a random vector according to the privacy attribute data after adding noise, wherein generating the random vector includes calculating the sensitivity of the privacy attribute data, and generating a random vector according to the privacy attribute data after adding noise and the sensitivity information, wherein the random vector elements obey Laplace distribution;

[0120] S32: Establish a confusion matrix using a polynomial function according to the generated random vector;

[0121] S33: performing a confusion operation on the noisy privacy attribute data according to the generated random vector and confusion matrix to obtain the obfuscated privacy attribute data. The calculation formula of the confusion operation is:

[0122] x′=xVMV T

[0123] in:

[0124] x′ represents the obfuscated privacy attribute data;

[0125] x′ represents the privacy attribute data after noise addition;

[0126] V represents the generated random vector;

[0127] M represents the confusion matrix;

[0128] T represents the transpose operation.

[0129] In the step S31, generating a random vector according to the privacy attribute data after adding noise includes:

[0130] S311: Calculate the sensitivity of the privacy attribute data. For the privacy attribute data x, the sensitivity calculation formula is:

[0131]

[0132] in:

[0133] f(x) represents the sensitivity function;

[0134] neighbors(x,x ′ ) represents the set of data points in the neighborhood of data x;

[0135] x ′ Represents any data point variable within the neighborhood;

[0136] S312: Perform Laplace sampling according to the calculated sensitivity to generate a random vector, wherein the random vector element sampling calculation formula is:

[0137]

[0138] in:

[0139] b represents the scale parameter of the Laplace distribution, which controls the peak degree of the distribution;

[0140] x represents privacy attribute data;

[0141] p(x) represents the sampling result.

[0142] In the step S32, a confusion matrix is ​​established using a polynomial function according to the generated random vector V, including:

[0143] S321: According to the preset polynomial function f(x)=a n x n +a n―1 x n―1 +...+a1x+a0, taking each element in the random vector as input and the corresponding privacy attribute data after noise addition as output, constructing a system of equations;

[0144] in:

[0145] a n 、a n―1 , a1, a0 represent polynomial coefficients;

[0146] x represents the input variable;

[0147] S322: jointly solving the constructed equation group to obtain optimal polynomial function coefficients, and concatenating the polynomial function coefficients into a coefficient vector P;

[0148] S323: Perform outer product calculation on the random vector V and the coefficient vector to obtain a confusion matrix.

[0149] S4: Perform data fuzzification on the obfuscated private data obtained by querying the query statement to limit the accuracy of the query result. Generate pseudo data for the private data entries in the query result that are less than the specified threshold, and return the query result and the generated pseudo data together as the query result to avoid differential privacy data leakage.

[0150] In the step S4, pseudo data is generated for the privacy data of the query result whose privacy data items are less than the specified threshold, and the query result and the generated pseudo data are returned together as the query result, including:

[0151] S41: construct a privacy data generation model, the model includes a generation network and a discriminant network, the generation network is responsible for generating pseudo data, and the discriminant network is responsible for distinguishing true privacy data from pseudo data;

[0152] S42: Training the constructed private data generation model, wherein the generation network receives a random vector as input and outputs pseudo data similar to the real private data, and the discriminant network receives the real private data or the pseudo data generated by the generation network for distinction, and the optimal generation network and the optimal discriminant network are obtained by alternately training the objective function of the private data generation model. The objective function calculation formula of the private data generation model is:

[0153]

[0154] in:

[0155] G represents the generation network;

[0156] D represents the discriminant network;

[0157] x represents real private data;

[0158] z represents the value from the prior distribution p z Random noise sampled in (z);

[0159] p data (x) represents the real private data distribution;

[0160] S43: Generate pseudo data using the optimal generative network.

[0161] Embodiment 2:

[0162] like Figure 2 , which is a schematic diagram of the structure of an electronic device for implementing a privacy data anonymization protection method provided by an embodiment of the present invention.

[0163] The electronic device 1 may include a processor 10 , a memory 11 , a communication interface 13 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10 , such as a program 12 .

[0164] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. The memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1. The memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 may be used not only to store application software and various types of data installed in the electronic device 1, such as the code of the program 12, etc., but also to temporarily store data that has been output or is to be output.

[0165] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as the program 12 for implementing privacy data anonymization protection), and calls data stored in the memory 11, so as to execute various functions of the electronic device 1 and process data.

[0166] The communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices, and to achieve connection and communication between internal components of the electronic device.

[0167] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0168] Figure 2 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 2 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0169] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0170] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0171] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0172] The program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0173] Perform data desensitization on the original private data to obtain desensitized private data, ensuring that any desensitized private data entry is indistinguishable from at least K sensitive private data entries in the desensitized private data;

[0174] Perform differential privacy noise processing on the desensitized privacy data to obtain noisy privacy data, and use the Laplace mechanism to add dynamic noise to the desensitized privacy data to obtain privacy data with differential privacy protection characteristics;

[0175] Obfuscate the noisy private data attributes to obtain obfuscated private data;

[0176] The obfuscated private data obtained by querying the query statement is subjected to data fuzziness processing to limit the accuracy of the query results. For the private data entries in the query results that are less than the specified threshold, pseudo data is generated and the query results and the generated pseudo data are returned together as the query results to avoid differential privacy data leakage.

[0177] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figure 1 to Figure 2 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0178] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0179] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0180] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A privacy data anonymization protection method, characterized in that: The method comprises: S1: Desensitizing the original private data to obtain desensitized private data, ensuring that any desensitized private data entry is indistinguishable from at least K sensitive private data entries in the desensitized private data, wherein the improved recursive segmentation K anonymization operation is the implementation method of the data desensitization processing; S2: Perform differential privacy noise processing on the desensitized private data to obtain noisy private data. Use the Laplace mechanism to add dynamic noise to the desensitized private data to obtain private data with differential privacy protection characteristics. S3: Obfuscating the noisy private data attributes to obtain obfuscated private data, wherein the data obfuscation technology is an implementation method of the obfuscation; S4: Perform data fuzzification on the obfuscated private data obtained by querying the query statement to limit the accuracy of the query result. Generate pseudo data for the private data entries in the query result that are less than the specified threshold, and return the query result and the generated pseudo data together as the query result to avoid differential privacy data leakage.

2. A privacy data anonymization protection method as claimed in claim 1, characterized in that: In step S1, the original private data is desensitized to obtain desensitized private data, including: S11: Clean the original private data, remove outliers and missing values, and determine the sensitive attributes and non-sensitive attributes in the original private data; S12: Use a recursive partitioning method to partition the cleaned original privacy data into different clusters until each cluster contains at least K records. Each time the partition is performed, the attribute that can best improve anonymity is selected for partitioning. Information gain is the basis for determining the attribute that improves anonymity, and the calculation formula is: Gain(A)=H(D)―H(D|A) in: H(D) represents the information entropy of data set D; H(D|A) represents the conditional entropy of data set D under the condition of selecting attribute A; Gain(A) represents the information gain of attribute A; S13: Generalize the attributes of the divided data clusters to ensure the anonymity and privacy protection of the data. The attribute generalization should maintain the diversity of the data, that is, select attributes that do not affect the diversity of the data for generalization processing. The calculation formula for the data diversity is: Where: Div(Q) represents the diversity of attribute Q; S(q) represents the data subset whose attribute Q has a sensitive value of q; |Q| represents the number of values ​​of attribute Q; n is the number of data records; S14: Generalize the original private data according to the generalized attribute to obtain the desensitized original private data.

3. A privacy data anonymization protection method as claimed in claim 1, characterized in that: In step S2, differential privacy noise addition processing is performed on the desensitized privacy data, including: S21: Determine the privacy budget parameter ε in differential privacy according to the sensitivity of the attribute data, where the data sensitivity is divided into four levels: {extremely sensitive, sensitive, generally sensitive, insensitive}, corresponding to values ​​{0.001, 0.1, 0.5, 0.99} respectively; S22: Noise processing is performed on each desensitized attribute privacy data according to the privacy budget parameter, wherein dynamic noise generation based on the Laplace mechanism is the main implementation method of the noise processing, and the dynamic noise calculation formula based on the Laplace mechanism is: in: Δf represents the sensitivity of the query function, and its value range is {0.01, 0.99}, indicating sensitive and insensitive respectively; ε represents the privacy budget; Lap() represents Laplace distribution; S23: Add the generated dynamic noise to the desensitized private data to obtain the noisy private data.

4. A privacy data anonymization protection method as claimed in claim 1, characterized in that: In the step S3, the attributes of the noisy private data are obfuscated to improve the anti-interference property of the data and obtain the obfuscated private data, including: S31: generating a random vector according to the privacy attribute data after adding noise, wherein generating the random vector includes calculating the sensitivity of the privacy attribute data, and generating a random vector according to the privacy attribute data after adding noise and the sensitivity information, wherein the random vector elements obey Laplace distribution; S32: Establish a confusion matrix using a polynomial function according to the generated random vector; S33: performing a confusion operation on the noisy privacy attribute data according to the generated random vector and confusion matrix to obtain the obfuscated privacy attribute data. The calculation formula of the confusion operation is: x′=xVMV T in: x′ represents the obfuscated privacy attribute data; x′ represents the privacy attribute data after noise addition; V represents the generated random vector; M represents the confusion matrix; T represents the transpose operation.

5. A privacy data anonymization protection method as claimed in claim 4, characterized in that: In the step S31, generating a random vector according to the privacy attribute data after adding noise includes: S311: Calculate the sensitivity of the privacy attribute data. For the privacy attribute data x, the sensitivity calculation formula is: in: f(x) represents the sensitivity function; neighbors(x,x′) represents the set of data points in the neighborhood of data x; x′ represents any data point variable within the neighborhood; S312: Perform Laplace sampling according to the calculated sensitivity to generate a random vector, wherein the random vector element sampling calculation formula is: in: b represents the scale parameter of the Laplace distribution, which controls the peak degree of the distribution; x represents privacy attribute data; p(x) represents the sampling result.

6. A privacy data anonymization protection method as claimed in claim 4, characterized in that: In the step S32, a confusion matrix is ​​established using a polynomial function according to the generated random vector V, including: S321: According to the preset polynomial function f(x)=a n x n +a n―1 x n―1 +...+a1x+a0, taking each element in the random vector as input and the corresponding privacy attribute data after noise addition as output, constructing a system of equations; in: a n 、a n―1 , a1, a0 represent polynomial coefficients; x represents the input variable; S322: jointly solving the constructed equation group to obtain optimal polynomial function coefficients, and concatenating the polynomial function coefficients into a coefficient vector P; S323: Perform outer product calculation on the random vector V and the coefficient vector to obtain a confusion matrix.

7. A privacy data anonymization protection method as claimed in claim 1, characterized in that: In the step S4, pseudo data is generated for the privacy data of the query result whose privacy data items are less than the specified threshold, and the query result and the generated pseudo data are returned together as the query result, including: S41: construct a privacy data generation model, the model includes a generation network and a discriminant network, the generation network is responsible for generating pseudo data, and the discriminant network is responsible for distinguishing true privacy data from pseudo data; S42: Training the constructed private data generation model, wherein the generation network receives a random vector as input and outputs pseudo data similar to the real private data, and the discriminant network receives the real private data or the pseudo data generated by the generation network for distinction, and the optimal generation network and the optimal discriminant network are obtained by alternately training the objective function of the private data generation model. The objective function calculation formula of the private data generation model is: in: G represents the generation network; D represents the discriminant network; x represents real private data; z represents the value from the prior distribution p z Random noise sampled in (z); p data (x) represents the real private data distribution; S43: Generate pseudo data using the optimal generative network.

Images