Data privacy protection method and device, electronic equipment and storage medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2022-07-21
- Publication Date
- 2026-05-29
Smart Images
Figure CN115168898B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of artificial intelligence technology, specifically relating to a data privacy protection method, apparatus, electronic device, and computer-readable storage medium. Background Technology
[0002] Graph neural network technology, due to its ability to fully utilize both homogeneous and heterogeneous information in graph-structured data for high-dimensional feature modeling, has been increasingly widely applied in various fields, such as transportation, advertising recommendation, finance, and healthcare. Taking the financial industry as an example, as a data-intensive sector, its data is characterized by high value and high sensitivity. In the current wave of fintech, the privacy protection of data held by financial institutions and related companies requires particular attention.
[0003] To protect edge information (i.e., relationships between nodes) in graph-structured data, existing methods typically involve randomly selecting elements from the adjacency matrix and performing random, discrete 0- and 1 transformations. This random, discrete perturbation of the graph-structured data's edge information disrupts the sparsity of the adjacency matrix, impairs the performance of graph neural network models, and increases computational resource consumption (memory and computing power). Randomly and discretely perturbing the edge information of large graph data with small privacy budgets creates enormous memory and computing power requirements, rendering the graph neural network model unusable. How to achieve data privacy protection while reducing computational power and avoiding a sharp increase in computing resource demands is a pressing technical problem that needs to be solved by those skilled in the art.
[0004] Those skilled in the art have been seeking solutions to the above problems.
[0005] The preceding description is intended to provide general background information and does not necessarily constitute prior art. Summary of the Invention
[0006] The technical problem addressed by this application is that existing privacy data protection methods subject large graph data edge information to random discrete input perturbations with small privacy budgets, resulting in huge memory and computing power requirements, thus making graph neural network models unable to run. To address this, this application provides a data privacy protection method, apparatus, electronic device, and computer-readable storage medium.
[0007] The technical problem solved by this application is achieved by the following technical solution:
[0008] This application provides a data privacy protection method, comprising the following steps: obtaining an adjacency matrix of graph structure data, and obtaining a noise matrix by adding mixed noise to the adjacency matrix; calculating the number of edges Q used to filter matrix elements in the noise matrix based on the number of edges Q between nodes in the graph structure data; marking the matrix elements in the noise matrix according to the element values and the number of edges Q; converting the noise matrix into a perturbation matrix for describing the edge information of the graph structure data according to the position of the marked matrix elements; and adjusting the edge information of the graph structure data according to the perturbation matrix to obtain privacy graph structure data.
[0009] This application also provides a data privacy protection device, comprising: a noise addition module, used to obtain an adjacency matrix of graph structure data and obtain a noise matrix by adding mixed noise to the adjacency matrix; an edge number variation module, used to calculate the number of edges Q of the matrix elements in the noise matrix based on the number of edges Q between nodes in the graph structure data; an element marking module, used to mark the matrix elements in the noise matrix based on the element values and the number of edges Q; a matrix perturbation module, used to convert the noise matrix into a perturbation matrix for describing the edge information of the graph structure data based on the position of the marked matrix elements; and a structure adjustment module, used to adjust the edge information of the graph structure data according to the perturbation matrix to obtain privacy graph structure data.
[0010] According to one embodiment of this application, the noise addition module includes a noise mixing module. The noise mixing module is used to: perform differential privacy processing on at least two preset noises to obtain mixed noise; and superimpose the mixed noise on each element of the adjacency matrix to obtain a noise matrix.
[0011] According to one embodiment of this application, the mixed noise module further includes a mixed noise calculation module. The mixed noise module is used to: superimpose mixed noise onto each element of the adjacency matrix using the following formula to obtain a noise matrix:
[0012]
[0013] Where A(D) is the noise matrix; f(D) A mapping function for a given dataset D f:D→R d , indicating dataset D To one d Mapping relationships in 2D space; , p Let 1 be the norm. D’ To and D Another dataset that differs from each other by at most one data point; H It is mixed noise; εThis is the preset privacy budget parameter.
[0014] According to one embodiment of this application, the noise addition module includes a preset noise mixing module. The preset noise mixing module is used to mix Laplace noise and Gaussian noise according to preset weight values to obtain mixed noise.
[0015] According to one embodiment of this application, the preset noise mixing module includes a weight configuration module. The weight configuration module is used to set each preset weight value to 0.5.
[0016] According to one embodiment of this application, the edge number variation module includes an edge number variation calculation module. The edge number variation calculation module is used to: calculate a noise interference value based on a preset privacy budget parameter; and calculate the edge number Q based on the edge number Q and the noise interference value.
[0017] According to one embodiment of this application, the element marking module includes a sorting marking module. The sorting marking module is used to: mark the first Q matrix elements in the upper triangle of the noise matrix and the first Q matrix elements in the lower triangle of the noise matrix in descending order of element values.
[0018] According to one embodiment of this application, the matrix perturbation module includes a reset and readjustment module. The reset and readjustment module is used to: reset the noise matrix to a zero matrix, and set the marked matrix elements to 1 according to the positions of the marked matrix elements, while keeping the unmarked matrix elements as 0, to obtain the perturbation matrix.
[0019] According to one embodiment of this application, the structure adjustment module includes a node connection / disconnection module. The node connection / disconnection module is used to: connect the two nodes through an edge in the graph structure data when the corresponding element of the two nodes in the perturbation matrix is 1; and disconnect the connection between the two nodes in the graph structure data when the corresponding element of the two nodes in the perturbation matrix is 0.
[0020] This application also provides an electronic device including a processor and a memory: the processor is used to execute a computer program stored in the memory to implement the method as described above.
[0021] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method as described above.
[0022] Therefore, this application obtains a perturbation matrix by adding mixed noise to the adjacency matrix of the graph structure data. This makes the original graph structure data and the privacy-preserving graph structure data corresponding to the perturbation matrix indistinguishable, thus protecting data privacy. Simultaneously, the perturbation matrix is adjusted from the edge information of the graph structure data, ensuring that the perturbation matrix and the adjacency matrix retain similar sparsity. This avoids problems such as memory errors or rapid increases in computational power requirements, saving computational power and time costs.
[0023] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it according to the contents of the specification, and to make the above and other objects, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit this application. Attached Figure Description
[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0025] Figure 1 This is a schematic flowchart of a privacy data protection method provided in an embodiment of this application.
[0026] Figure 2 This is a schematic diagram of the original structure provided for an embodiment of this application.
[0027] Figure 3 This is a schematic diagram of an adjacency matrix provided in an embodiment of this application.
[0028] Figure 4 This is a schematic diagram of a noise matrix provided in one embodiment of this application.
[0029] Figure 5 This is a schematic diagram of a noise matrix with labeled elements provided in an embodiment of this application.
[0030] Figure 6 This is a schematic diagram of a perturbation matrix provided in an embodiment of this application.
[0031] Figure 7 This is a schematic diagram of a privacy graph structure provided in an embodiment of this application.
[0032] Figure 8 This is a schematic diagram of the structure of a data privacy protection device provided in an embodiment of this application.
[0033] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0034] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.
[0035] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.
[0036] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0037] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0038] In the specific implementation of this application, data such as user information are involved. When the various embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0039] The embodiments of this application will be further described in detail below with reference to the accompanying drawings.
[0040] Figure 1 This is a schematic flowchart illustrating a data privacy protection method provided in an embodiment of this application. For a clearer description of the data privacy protection method provided in an embodiment of this application, please refer to [link to relevant documentation]. Figures 1-7 .
[0041] In the process of business interactions with other companies or internally, enterprises inevitably need to provide some confidential business data. Industry practice typically involves data anonymization or providing "fake data." However, the content and structure of this data exhibit obvious graph structure characteristics (including nodes and edge relationships), making it susceptible to graph adversarial attacks, leading to frequent data breaches that seriously impact enterprise operational security and customer privacy, especially for financial institutions. Furthermore, all commercial data or personal privacy data with similar characteristics are threatened with data privacy breaches. Existing technologies that randomly and discretely perturb the edge information of large graph data with small privacy budgets create enormous memory and computing power requirements, rendering graph neural network models unusable. Therefore, this application proposes a data privacy protection method, including steps S110-S150.
[0042] Step S110: Obtain the adjacency matrix of the graph structure data, and obtain the noise matrix by adding mixed noise to the adjacency matrix.
[0043] In one embodiment, for ease of understanding, an example is provided below; the original diagram structure data can be referenced. Figure 2 , Figure 2 This is a schematic diagram of the original structure provided in an embodiment of this application. V1 - V5 Representing each node, where E1 - E6 This represents the edge information between each node. Based on the connection relationships between nodes, the adjacency matrix of this graph structure data can be obtained, that is... Figure 3 , Figure 3 This is a schematic diagram of an adjacency matrix provided in an embodiment of this application. In this matrix, the element corresponding to two nodes connected by an edge is 1, and the element corresponding to two nodes not connected by an edge is 0.
[0044] In one embodiment, step S110 involves adding mixed noise to the adjacency matrix to obtain a noise matrix, including: performing differential privacy processing on at least two preset noises to obtain mixed noise; and superimposing the mixed noise on each element of the adjacency matrix to obtain a noise matrix.
[0045] In one embodiment, mixed noise refers to noise obtained by mixing two or more noise mechanisms according to a preset weight configuration. The noise mechanisms may specifically include, but are not limited to, Gaussian noise, Laplace noise, Poisson noise, and salt-and-pepper noise. Differential privacy processing is performed to obtain the mixed noise, meaning the mixed noise needs to meet certain conditions. ε- The conditions for differential privacy, and the specific process of proving their satisfaction, will be detailed later and will not be elaborated here. By superimposing mixed noise onto each element of the adjacency matrix, a noise matrix can be obtained.
[0046] In one embodiment, a noise matrix is obtained by superimposing mixed noise onto each element of the adjacency matrix using the following formula (1):
[0047] ... (1)
[0048] Where A(D) is the noise matrix; f(D) A mapping function for a given dataset D f:D→R d , indicating dataset D To one d Mapping relationships in 2D space; , p Let 1 be the norm. D’ To and D Another dataset that differs from each other by at most one data point; H It is mixed noise; ε This is the preset privacy budget parameter.
[0049] In one implementation, A(.) is the function for obtaining the noise matrix. Therefore, during the calculation process, D can be the adjacency matrix, and A(D) is the output value of the function, which is the noise matrix. yes f(D) and f(D') The methods for measuring and calculating the difference between them are as described above. It is important to note that... ε , ε The privacy budget parameter is a preset value, ranging from 0 to 0.1, and needs to be estimated using cross-validation. In a preferred implementation, the default value is generally 0.05.
[0050] In one embodiment, differential privacy processing is performed on the noise to obtain mixed noise, including: mixing Laplace noise and Gaussian noise according to preset weight values to obtain mixed noise H.
[0051] In one embodiment, the formula (2) for generating Laplace noise is:
[0052] ... (2)
[0053] The formula (3) for generating Gaussian noise is:
[0054] ... (3)
[0055] Therefore, by mixing the Laplace noise and Gaussian noise according to the preset weight values, we can obtain the formula (4) for calculating the mixed noise H:
[0056] ... (4)
[0057] Where n is the preset weight value for Laplace noise, and the corresponding preset weight value for Gaussian noise is 1-n. The preset weight value n ranges from (0,1), but the final mixed noise H needs to satisfy... ε- Conditions for defining differential privacy.
[0058] In one embodiment, the Laplace noise and Gaussian noise can be averaged using equal weights to generate the mixed noise H, that is, the preset weights for the Laplace noise and Gaussian noise are each 0.5. Therefore, in a preferred embodiment, the mixed noise H is finally determined by formula (5):
[0059] ... (5)
[0060] in, u and μ Take 0, b and σ for .
[0061] In one embodiment, for generating mixed noise H by equally weighting Laplace noise and Gaussian noise using formula (5), it satisfies ε- The conditions for differential privacy are defined, and semantic proof will follow.
[0062] Wherein, the mixed noise H satisfies ε- The conditions that must be met for differential privacy definition are:
[0063]
[0064] Among them, probability Pr [·] represents the probability of the algorithm. O For the A(D) The calculated value.
[0065] set up: ,
[0066] ,
[0067]
[0068] Without loss of generality, let x i If all values are 0, then:
[0069] ,
[0070] .
[0071] Let the output vector of A(D) be... Then we have:
[0072] ,
[0073] ,
[0074] .
[0075] Next, all that is needed is to prove This can prove
[0076]
[0077] For each The following inequalities exist:
[0078]
[0079] Then we have:
[0080]
[0081] Therefore:
[0082] Proof obtained.
[0083] In one embodiment, after the above proof process, it can be determined that the mixed noise H obtained by formula (5) satisfies ε- The conditions for differential privacy definition. It is understood that formula (5) is a preferred implementation proposed in this application, that is, Laplace noise and Gaussian noise are mixed by equal weight averaging. In other implementations, any two or more other noise mechanisms can be mixed according to their respective preset weight values. The above examples and proofs are illustrative of the method, not limitations on the solution. However, it is also understood that the difference in noise mechanisms or weights in the mixed noise will affect the privacy protection effect. Therefore, when using other mixed noise mechanisms or different weight values, it is necessary to meet the following conditions. ε- The data privacy protection methods provided in this application can only be used under the conditions defined by differential privacy.
[0084] In one embodiment, the following description will use the mixed noise H obtained by formula (5) as an example, and the mixed noise obtained by formula (1) as an example. This will be used as a reference in the following text, and will not be repeated hereafter. Figure 2 and Figure 3 Taking the provided graph structure data and adjacency matrix as an example, for Figure 3 After adding mixed noise to the provided adjacency matrix, we can obtain... Figure 4 The noise matrix shown is Figure 4This is a schematic diagram of a noise matrix provided in an embodiment of this application. The sum of the values of each element in the matrix represents the calculated value of the mixed noise.
[0085] Step S120: Calculate the number of edges Q between nodes in the graph structure data to filter matrix elements in the noise matrix.
[0086] In one embodiment, step 120 calculates the number of edges Q used to filter matrix elements in the noise matrix based on the number of edges Q between nodes in the graph structure data, including: calculating the noise interference value based on a preset privacy budget parameter; and calculating the number of edges Q based on the number of edges Q and the noise interference value.
[0087] In one embodiment, the noise interference value is calculated based on a preset privacy budget parameter, and the number of edges Q is calculated based on the number of edges Q and the noise interference value. The calculation process can be referred to formula (6):
[0088] ... (6)
[0089] in, This can be considered as a mixed noise, which is an equal-weighted average of the Gaussian noise and Laplace noise mentioned earlier. u and μ Take 0, b for , σ for Furthermore, during the calculation process, for The output value is rounded up to the nearest integer. Therefore, in this embodiment, the output value is... Figure 2 In E1 - E6 It can be seen that the number of edges Q between nodes in the graph structure data is 6. The output value is rounded up to 1, and the final value of the number of sides Q of the matrix element used to filter the noise matrix is 7 according to formula (6).
[0090] Step S130: Mark the matrix elements in the noise matrix according to the element values and the number of edges Q.
[0091] In one embodiment, the first Q matrix elements in the upper triangle of the noise matrix are marked in descending order of element values, and the first Q matrix elements in the lower triangle of the noise matrix are also marked.
[0092] Specifically, since the calculated Q value in the aforementioned example is 7, and the adjacency matrix of the graph structure is a symmetric matrix with diagonal elements of 0, the top Q elements in the upper and lower triangles of the noise matrix obtained from the adjacency matrix are labeled respectively. The labeling process can be referred to... Figure 5 , Figure 5 This is a schematic diagram of a noise matrix with labeled elements provided in one embodiment of this application. For example... Figure 5 As shown, the first 7 elements in the upper and lower triangles are marked with dashed boxes. The mixing matrix is adjusted based on these Q index positions, ensuring that the resulting perturbation matrix maintains a similar sparsity to the original adjacency matrix.
[0093] Step S140: Based on the position of the marked matrix elements, convert the noise matrix into a perturbation matrix used to describe the edge information of the graph structure data.
[0094] In one embodiment, the noise matrix is reset to a zero matrix, and the marked matrix elements are set to 1 according to their positions, while the unmarked matrix elements are left as 0, thus obtaining the perturbation matrix.
[0095] Specifically, the final perturbation matrix can be referenced. Figure 6 , Figure 6 This is a schematic diagram of a perturbation matrix provided in an embodiment of this application. As shown in the figure, the marked elements in the noise matrix are all set to 1, while the unmarked elements are all set to 0. Specifically, the process can be as follows: Let the noise matrix be... Let S be the position of the marked element. First, the noise matrix... Reset to a zero matrix, and then... and Reset to 1 (where ),other The element value is kept as 0, thus obtaining the following: Figure 6 The perturbation matrix is shown.
[0096] Step S150: Adjust the edge information of the graph structure data according to the perturbation matrix to obtain the privacy graph structure data.
[0097] In one embodiment, adjusting the edge information of the graph structure data according to the perturbation matrix in step S150 includes: when the corresponding element of two nodes in the perturbation matrix is 1, connecting the two nodes through an edge in the graph structure data; when the corresponding element of two nodes in the perturbation matrix is 0, disconnecting the connection between the two nodes in the graph structure data.
[0098] In one implementation, it is understood that the perturbation matrix also represents the connection relationships between nodes; therefore, the graph structure can be modified based on the adjusted perturbation matrix. Figure 6 The perturbation matrix shown can adjust the edge information of the graph structure: when the corresponding element of two nodes in the perturbation matrix is 1, the two nodes are connected by an edge in the graph structure data; when the corresponding element of two nodes in the perturbation matrix is 0, the connection between the two nodes is broken in the graph structure data, to finally obtain... Figure 7 The privacy graph structure shown is Figure 7 This is a schematic diagram of a privacy graph structure provided in an embodiment of this application. Figure 2 and Figure 7 In comparison, it can be seen that originally Figure 2 In E5 The edges have been broken, and new ones have been added, such as... Figure 7 The middle dashed line connects E7 and E5 The sparsity of edges and graph structures is similar, making it impossible for attackers to infer the actual dataset from the obtained data or to deduce the actual edge information from the returned node relationship data, thus achieving privacy protection. Furthermore, by simulating multiple attacks on the dataset using differential attack methods, and comparing this with random discrete perturbations, tests using the data privacy protection method provided in this application showed that after the simulated attacks, the probability of edge information leakage in graph structure data decreased by an average of 19.3%, and the computation time cost was reduced by an average of 21.6%, saving computational power and time while protecting data privacy. It is understood that the order of edge information is only used for naming distinction and there is no difference in their relative order; simultaneously... E7 and E5 The dashed lines connecting the edges are only for easy differentiation from the original structure; they are not related to the actual structure. E7 and E5 The edge information is not actually different from other edge information. Additionally, a note is needed regarding... Figures 2 to 7 The graph structure data and matrices listed are merely examples for ease of understanding. The specific matrices and graph structures are not limited to these examples, and the above examples do not constitute a limitation on the technical solution.
[0099] Therefore, the data privacy protection method provided in this application adds mixed noise to the adjacency matrix of the graph structure data to obtain a perturbation matrix, making it difficult to distinguish between the original graph structure data and the privacy graph structure data corresponding to the perturbation matrix, thereby achieving data privacy protection. Simultaneously, the perturbation matrix is adjusted from the edge information of the graph structure data, ensuring that the perturbation matrix and the adjacency matrix retain similar sparsity, thus avoiding problems such as memory errors or rapid increases in computational power requirements, saving computational power and time costs. Furthermore, the data privacy protection method provided in one embodiment of this application can also generate mixed noise by combining Gaussian noise and Laplace noise with equal weights, forming a differential privacy method based on a mixed noise mechanism. This makes two similar datasets indistinguishable, meaning that in privacy attacks, attackers cannot infer the actual dataset from the data obtained from the attack. At the same time, by adding the aforementioned mixed noise to the adjacency matrix and performing corresponding transformations on the graph structure data, attackers cannot infer the actual edge information from the returned node relationship data, thus achieving a good privacy protection effect. Furthermore, training graph neural networks can be effectively performed on the entire dataset while maintaining privacy, significantly reducing the probability of edge information leakage in graph-structured data and effectively compressing computational time costs. By pre-calculating the original graph density to adjust the adjacency matrix after mixing noise, the sparsity of the original adjacency matrix is preserved, avoiding a sharp increase in computational resource requirements. This ensures that it can be applied to graph neural networks with small privacy budgets or large graph-structured data, effectively protecting the edge information of graph-structured data without leading to memory errors or a rapid increase in computational power requirements.
[0100] Figure 8 This is a schematic diagram of the structure of a data privacy protection device provided in an embodiment of this application. For a clearer description of the electronic device 110 provided in an embodiment of this application, please refer to... Figure 8 .
[0101] This application also provides a data privacy protection device, comprising: a noise addition module A810, which is used to obtain the adjacency matrix of graph structure data and obtain a noise matrix by adding mixed noise to the adjacency matrix; an edge number variation module A820, which is used to calculate the edge number Q of the matrix elements in the noise matrix based on the edge number Q between nodes in the graph structure data; an element marking module A830, which is used to mark the matrix elements in the noise matrix based on the element values and the edge number Q; a matrix perturbation module A840, which is used to convert the noise matrix into a perturbation matrix for describing the edge information of the graph structure data based on the position of the marked matrix elements; and a structure adjustment module A850, which is used to adjust the edge information of the graph structure data according to the perturbation matrix to obtain privacy graph structure data.
[0102] In one embodiment, the noise addition module A810 includes a noise mixing module. The noise mixing module is used to: perform differential privacy processing on at least two preset noises to obtain mixed noise; and superimpose the mixed noise on each element of the adjacency matrix to obtain a noise matrix.
[0103] In one embodiment, the noise mixing module further includes a noise mixing calculation module. The noise mixing module is used to: superimpose mixed noise onto each element of the adjacency matrix using the following formula (1) to obtain a noise matrix:
[0104] ... (1)
[0105] Where A(D) is the noise matrix; f(D) A mapping function for a given dataset D f:D→R d , indicating dataset D To one d Mapping relationships in 2D space; , p Let 1 be the norm. D’ To and D Another dataset that differs from each other by at most one data point; H It is mixed noise; ε This is the preset privacy budget parameter.
[0106] In one embodiment, the noise addition module A810 includes a preset noise mixing module. The preset noise mixing module is used to mix Laplace noise and Gaussian noise according to preset weight values to obtain mixed noise.
[0107] In one embodiment, the preset noise mixing module includes a weight configuration module. The weight configuration module is used to set each preset weight value to 0.5.
[0108] In one embodiment, the edge number variation module A820 includes an edge number variation calculation module. The edge number variation calculation module is used to: calculate a noise interference value based on a preset privacy budget parameter; and calculate the edge number Q based on the edge number Q and the noise interference value.
[0109] In one embodiment, the element marking module A830 includes a sorting marking module. The sorting marking module is used to: mark the first Q matrix elements in the upper triangle of the noise matrix and the first Q matrix elements in the lower triangle of the noise matrix in descending order of element values.
[0110] In one embodiment, the matrix perturbation module A840 includes a reset and readjustment module. The reset and readjustment module is used to: reset the noise matrix to a zero matrix, and set the marked matrix elements to 1 according to the positions of the marked matrix elements, while keeping the unmarked matrix elements as 0, to obtain the perturbation matrix.
[0111] In one embodiment, the structure adjustment module A850 includes a node connection / disconnection module. The node connection / disconnection module is used to: connect the two nodes through an edge in the graph structure data when the corresponding element of the two nodes in the perturbation matrix is 1; and disconnect the connection between the two nodes in the graph structure data when the corresponding element of the two nodes in the perturbation matrix is 0.
[0112] One embodiment of this application provides a data privacy protection device 80 to implement the steps of the data privacy protection method described in the above embodiments. By adding mixed noise to the adjacency matrix of the graph structure data to obtain a perturbation matrix, the original graph structure data and the privacy-preserving graph structure data corresponding to the perturbation matrix become indistinguishable, thereby achieving data privacy protection. Simultaneously, the perturbation matrix is adjusted from the edge information of the graph structure data, ensuring that the perturbation matrix and the adjacency matrix retain similar sparsity. This avoids problems such as memory errors or rapid increases in computational power requirements, saving computational power and time costs. Furthermore, the data privacy protection device 80 provided in one embodiment of this application can also generate mixed noise by combining Gaussian noise and Laplace noise with equal weights, forming a differential privacy method based on the mixed noise mechanism. This makes two similar datasets indistinguishable, meaning that in privacy attacks, attackers cannot infer the actual dataset from the data obtained through the attack. Simultaneously, by adding the aforementioned mixed noise to the adjacency matrix and performing corresponding transformations on the graph structure data, attackers cannot infer the actual edge information from the returned node relationship data, thus achieving a good privacy protection effect. Moreover, under the privacy protection of the entire dataset, graph neural network training can be effectively performed, significantly reducing the probability of edge information leakage from the graph structure data and effectively compressing computational time costs. By pre-calculating the original graph density to adjust the adjacency matrix after mixing noise, the sparsity of the original adjacency matrix is preserved, avoiding a sharp increase in computational resource requirements. This ensures that it can be applied to graph neural networks with small privacy budgets or large graph structure data, effectively protecting the edge information of the graph structure data without causing memory errors or a rapid increase in computational power requirements.
[0113] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For a clearer description of the electronic device 110 provided in an embodiment of this application, please refer to... Figure 1 and Figure 9 .
[0114] An embodiment of this application provides an electronic device 110, including a processor A101 and a memory A201, wherein the processor A101 is used to execute a computer program A6 stored in the memory A201 to implement the steps of the data privacy protection method as described in the embodiments of this application.
[0115] In one embodiment, the electronic device 110 provided in this embodiment includes at least one processor A101 and at least one memory A201. The at least one processor A101 may be referred to as a processing unit A1, and the at least one memory A201 may be referred to as a storage unit A2. Specifically, the storage unit A2 stores a computer program A6, which, when executed by the processing unit A1, causes the electronic device 110 provided in this embodiment to implement the steps of the data privacy protection method described in the various embodiments of this application. For example, Figure 1 The steps shown are as follows: Step S110: Obtain the adjacency matrix of the graph structure data, and obtain the noise matrix by adding mixed noise to the adjacency matrix; Step S120: Calculate the number of edges Q used to filter the matrix elements in the noise matrix based on the number of edges Q between nodes in the graph structure data; Step S130: Mark the first Q elements in the upper triangle of the noise matrix and the first Q elements in the lower triangle of the noise matrix in descending order of element values; Step S140: After resetting the noise matrix to a zero matrix, set the matrix elements at the marked positions to 1, and leave the matrix elements at the remaining positions as 0 to obtain the perturbation matrix; Step S150: Adjust the edge information of the graph structure data according to the perturbation matrix to obtain the privacy graph structure data.
[0116] In one embodiment, the electronic device 110 provided in this embodiment may include a plurality of memories A201 (referred to as memory units A2).
[0117] Storage unit A2 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The storage unit A2 described in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.
[0118] An embodiment of this application provides an electronic device 110, including a memory A101 and a processor A201. The processor A101 executes a computer program A6 stored in the memory A201 to implement the steps of the data privacy protection method described in the embodiments of this application. Therefore, the electronic device 110 provided in this embodiment obtains a perturbation matrix by adding mixed noise to the adjacency matrix of the graph structure data, making it difficult to distinguish between the original graph structure data and the privacy graph structure data corresponding to the perturbation matrix, thereby achieving data privacy protection. At the same time, the perturbation matrix is obtained by adjusting the edge information of the graph structure data, so that the perturbation matrix and the adjacency matrix retain similar sparsity, thereby avoiding problems such as memory errors or rapid growth in computing power requirements, and saving computing power and computation time costs. Furthermore, the electronic device 110 provided in one embodiment of this application can also generate mixed noise by combining Gaussian noise and Laplace noise with equal weights, forming a differential privacy method based on the mixed noise mechanism. This makes two similar datasets indistinguishable, meaning that in privacy attacks, attackers cannot infer the actual dataset from the data obtained through the attack. Simultaneously, by adding the aforementioned mixed noise to the adjacency matrix and performing corresponding transformations on the graph structure data, attackers cannot infer the actual edge information from the returned node relationship data, thus achieving a good privacy protection effect. Moreover, under the privacy protection of the entire dataset, graph neural network training can be effectively performed, significantly reducing the probability of edge information leakage from the graph structure data and effectively compressing computational time costs. By pre-calculating the original graph density to adjust the adjacency matrix after mixing noise, the sparsity of the original adjacency matrix is preserved, avoiding a sharp increase in computational resource requirements. This ensures that it can be applied to graph neural networks with small privacy budgets or large graph structure data, effectively protecting the edge information of the graph structure data without causing memory errors or a rapid increase in computational power requirements.
[0119] An embodiment of this application also provides a computer-readable storage medium storing a computer program A6, which, when executed by a processor A101, implements the steps of the data privacy protection method as described in the embodiments of this application.
[0120] In one embodiment, the computer-readable storage medium provided in this embodiment is a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.
[0121] The technical effects that can be achieved when the computer program A6 stored in the computer-readable storage medium provided in one embodiment of this application is executed by the processor A101 have been described in detail above, and will not be repeated here.
[0122] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above 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.
[0123] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment. In this document, unless otherwise stated, "a plurality of" or "several" means two or more.
[0124] It should be understood that although the steps in the flowcharts of this application's embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.
[0125] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium, and when executed, the program performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0126] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A data privacy protection method, characterized in that, include: Obtain the adjacency matrix of the graph structure data, and obtain a noise matrix by adding mixed noise to the adjacency matrix; wherein, differential privacy processing is performed on at least two preset noises to obtain mixed noise; the mixed noise is superimposed on each element of the adjacency matrix to obtain the noise matrix; Based on the number of edges t between nodes in the graph structure data, the number of edges Q used to label the matrix elements in the noise matrix is calculated; wherein, the number of edges Q is the sum of the number of edges t and the mixed noise; In descending order of element value, mark the first Q matrix elements in the upper triangle of the noise matrix and mark the first Q matrix elements in the lower triangle of the noise matrix. The noise matrix is reset to a zero matrix, and the marked matrix elements are set to 1 according to their positions, while the unmarked matrix elements are left as 0, thus obtaining the perturbation matrix. Adjust the edge information of the graph structure data according to the perturbation matrix to obtain privacy graph structure data.
2. The data privacy protection method according to claim 1, characterized in that, The step of calculating the number of edges Q used to filter matrix elements in the noise matrix based on the number of edges t between nodes in the graph structure data includes: Based on preset privacy budget parameters, the noise interference value is calculated; The number of edges Q is calculated based on the number of edges t and the noise interference value.
3. The data privacy protection method according to claim 1, characterized in that, The step of adjusting the edge information of the graph structure data according to the perturbation matrix includes: When the corresponding element of two nodes in the perturbation matrix is 1, the two nodes are connected by an edge in the graph structure data. When the corresponding element of the two nodes in the perturbation matrix is 0, the connection between the two nodes is disconnected in the graph structure data.
4. A data privacy protection device, characterized in that, include: A noise addition module is used to obtain the adjacency matrix of graph structure data and to obtain a noise matrix by adding mixed noise to the adjacency matrix; wherein, differential privacy processing is performed on at least two preset noises to obtain mixed noise; and the mixed noise is superimposed on each element of the adjacency matrix to obtain the noise matrix. An edge number variation module is used to calculate the edge number Q for filtering matrix elements in the noise matrix based on the edge number t between nodes in the graph structure data; wherein, the edge number Q is the sum of the edge number t and the mixed noise. An element marking module is used to mark the first Q matrix elements in the upper triangle of the noise matrix and the first Q matrix elements in the lower triangle of the noise matrix in descending order of element values. A matrix perturbation module is used to reset the noise matrix to a zero matrix, and set the marked matrix elements to 1 according to the position of the marked matrix elements, while keeping the unmarked matrix elements as 0, to obtain a perturbation matrix; A structure adjustment module is used to adjust the edge information of the graph structure data according to the perturbation matrix to obtain privacy graph structure data.
5. An electronic device, characterized in that, Including processor and memory: The processor is used to execute a computer program stored in the memory to implement the method as described in any one of claims 1 to 4.
6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 4.