A noise generation method, device, terminal and medium based on differential privacy

Abstract

This application discloses a noise generation method, device, terminal, and medium based on differential privacy, which are applied to a data middle platform. The data middle platform includes a data file library and a server. The data file library is used to store original data files and file labels corresponding to the original data files. The method includes: determining the data sensitivity corresponding to the queried data based on the data characteristics of the queried data in the original data file; determining the privacy parameters corresponding to the queried data based on the queried data and the file labels; determining the scale parameters corresponding to the queried data based on the data sensitivity and privacy parameters; and obtaining target noise data based on the scale parameters and the file labels using a preset noise generation model. This application aims to generate high-quality noise data to protect the privacy of the data.

Description

Technical Field

[0001] The present application relates to the field of data privacy protection, and in particular to a noise generation method, device, terminal and medium based on differential privacy. Background Art

[0002] Differential privacy is a mathematical framework for protecting individual privacy. It aims to allow statistical analysis of databases while protecting the security and privacy of individual data. Even if an attacker possesses all copies of a particular record except one, they cannot determine the existence or specific content of that record.

[0003] However, for the data that needs to be protected in data files, the existing data privacy protection methods have the following shortcomings: they are unable to accurately provide more suitable noise data for the protected data based on the user's requirements for confidentiality levels and the specific actual scenarios in which the data is located, resulting in insufficient privacy protection of the protected data, or causing serious distortion of the protected data during the application process. Summary of the Invention

[0004] The main purpose of this application is to provide a noise generation method, device, terminal and medium based on differential privacy, aiming to generate high-quality noise data to protect the privacy of the data.

[0005] To achieve the above objectives, the present application provides a noise generation method based on differential privacy, which is applied to a data middle platform, wherein the data middle platform includes a data file library and a server, wherein the data file library is used to store original data files and file tags corresponding to the original data files, and the file tags are used to characterize the status of the data in the original data files. The data file library and the server are both in the same network;

[0006] The method comprises:

[0007] Determining data sensitivity corresponding to the queried data based on data features of the queried data in the original data file;

[0008] Determining privacy parameters corresponding to the queried data based on the queried data and the file tag;

[0009] Determine the scale parameters corresponding to the queried data based on data sensitivity and privacy parameters;

[0010] By presetting a noise generation model, target noise data is obtained according to the scale parameter and the file label to protect the privacy of the queried data.

[0011] Specifically, the file tag includes a compliance level tag, a confidentiality expectation tag, and an application scenario tag. The compliance level tag is used to indicate whether the content in the original data file is subject to industry-specific data protection standards. The confidentiality expectation tag is used to indicate the degree of expectation of the user corresponding to the original data file regarding the confidentiality level of the original data file. The application scenario tag is used to indicate the application scenario in which the content in the original data file is located.

[0012] The determining, based on the queried data and the file tag, a privacy parameter corresponding to the queried data includes:

[0013] By presetting the privacy parameter generation model, the privacy parameter is obtained according to the queried data, the compliance level label, the confidentiality expectation label and the application scenario label.

[0014] Specifically, the preset privacy parameter generation model includes a first input layer, a first embedding layer, a second embedding layer, a third embedding layer, a first feature fusion layer, and a fully connected layer;

[0015] The model is generated by presetting privacy parameters, and obtaining the privacy parameters according to the queried data, the compliance level label, the confidentiality expectation label, and the application scenario label, including:

[0016] Obtaining, through the first input layer, an input vector corresponding to the queried data according to the queried data;

[0017] Obtaining a compliance level label vector through the first embedding layer and the compliance level label;

[0018] Obtaining a confidentiality expected label vector through the second embedding layer and the confidentiality expected label;

[0019] Obtaining an application scenario label vector through the third embedding layer and the application scenario label;

[0020] Obtaining, through the first feature fusion layer, a first feature fusion vector according to the input vector corresponding to the queried data, the compliance level label vector, the confidentiality expectation label vector, and the application scenario label vector;

[0021] The privacy parameter is obtained through the fully connected layer according to the first feature fusion vector.

[0022] Specifically, the fully connected layer includes a first sub-fully connected layer, a second sub-fully connected layer, and a third sub-fully connected layer, the activation function of the first sub-fully connected layer and the activation function of the second sub-fully connected layer are both ReLU functions, and the activation function of the third sub-fully connected layer is a Sigmoid function;

[0023] Obtaining the privacy parameter according to the first feature fusion vector through the fully connected layer includes:

[0024] Obtaining a first intermediate vector according to the first feature fusion vector through the first sub-fully connected layer;

[0025] Obtaining a second intermediate vector according to the first intermediate vector through the second sub-fully connected layer;

[0026] The privacy parameter is obtained according to the second intermediate vector through the third sub-fully connected layer.

[0027] Specifically, determining the scale parameter corresponding to the queried data based on the data sensitivity and privacy parameters includes:

[0028] The scale parameter is calculated based on a preset scale parameter calculation formula according to the data sensitivity and the privacy parameter.

[0029] Specifically, the preset noise generation model includes a second input layer, a fourth embedding layer, a second feature fusion layer, a convolutional layer, an LSTM layer, a fourth sub-fully connected layer, a fifth sub-fully connected layer, and a sixth sub-fully connected layer, the activation function of the fourth sub-fully connected layer and the activation function of the fifth sub-fully connected layer are both ReLU functions, and the activation function of the sixth sub-fully connected layer is a linear activation function;

[0030] The method of obtaining target noise data by using a preset noise generation model according to the scale parameter and the file label to protect the privacy of the queried data includes:

[0031] Obtaining a scale parameter vector according to the scale parameter through the second input layer;

[0032] Obtaining a file label vector according to the file label through the fourth embedding layer;

[0033] Obtaining a second feature fusion vector according to the file label vector and the scale parameter vector through the second feature fusion layer;

[0034] Obtaining a convolution feature map through the convolution layer according to the second feature fusion vector;

[0035] Obtaining an LSTM feature vector according to the convolutional feature map through the LSTM layer;

[0036] Obtaining a third intermediate vector according to the LSTM feature vector through the fourth fully connected sub-layer;

[0037] Obtaining a fourth intermediate vector according to the third intermediate vector through the fifth fully connected sub-layer;

[0038] Target noise data is obtained according to the fourth intermediate vector through the sixth sub-fully connected layer.

[0039] Specifically, before determining the data sensitivity corresponding to the queried data based on the data features of the queried data in the original data file, the method further includes:

[0040] In response to the received data query instruction, the queried data is retrieved from the original data file in the data set file library.

[0041] To achieve the above objectives, the present application also provides a noise generation device based on differential privacy, which is applied to a data middle platform, wherein the data middle platform includes a data file library and a server, wherein the data file library is used to store original data files and file tags corresponding to the original data files, wherein the file tags are used to characterize the status of the data in the original data files, and the data file library and the server are both in the same network;

[0042] The device comprises:

[0043] The first unit is configured to determine the data sensitivity corresponding to the queried data according to the data characteristics of the queried data in the original data file;

[0044] A second unit is configured to determine a privacy parameter corresponding to the queried data based on the queried data and the file tag;

[0045] The third unit is used to determine the scale parameter corresponding to the queried data based on the data sensitivity and privacy parameters;

[0046] The fourth unit is configured to obtain target noise data according to the scale parameter and the file label by using a preset noise generation model to protect the privacy of the queried data.

[0047] To achieve the above objectives, the present application also provides a terminal, comprising a memory storing a plurality of instructions; the processor loads instructions from the memory to execute the steps in any one of the methods provided in the present application.

[0048] To achieve the above objectives, the present application also provides a medium, which stores multiple instructions, and the instructions are suitable for a processor to load to execute the steps of any method provided in the present application.

[0049] The present application provides a noise generation method, device, terminal, and medium based on differential privacy. First, the data sensitivity corresponding to the queried data can be determined based on the data characteristics of the queried data in the original data file; then, the privacy parameters corresponding to the queried data can be determined based on the queried data and the file label; then, the scale parameters corresponding to the queried data can be determined based on the data sensitivity and the privacy parameters; and finally, the target noise data can be obtained based on the scale parameters and the file label through a preset noise generation model to protect the privacy of the queried data.

[0050] Compared with existing data privacy protection methods, this application can accurately provide more suitable noise data for the protected data based on the user's requirements for confidentiality level and the specific actual scenario where the data is located, so that the protected data can be used for effective data analysis or application development while obtaining high-quality privacy protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a process flow provided for an embodiment of the present application;

[0052] Figure 2 A schematic diagram of the structure of a noise generation device based on differential privacy provided in an embodiment of the present application;

[0053] Figure 3 A schematic diagram of the structure of a terminal provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0055] Due to the data that needs to be protected in data files, the existing data privacy protection methods have the following shortcomings: they are unable to accurately provide more suitable noise data for the protected data based on the user's requirements for confidentiality levels and the specific actual scenarios where the data is located, resulting in insufficient privacy protection of the protected data, or causing serious distortion of the protected data during the application process.

[0056] Therefore, the embodiments of the present application provide a noise generation method, device, terminal and medium based on differential privacy to solve practical technical problems.

[0057] In some embodiments, the device may be integrated into an electronic device, which may be a terminal, a server, or other device.

[0058] In some embodiments, the server may also be implemented in the form of a terminal.

[0059] Among them, the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0060] The terminal may be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited thereto. The terminal and the server may be directly or indirectly connected via wired or wireless communication, and this application does not impose any restrictions thereon.

[0061] It should be noted that the serial numbers of the following embodiments are not intended to limit the preferred order of the embodiments.

[0062] An embodiment of the present application provides a noise generation method based on differential privacy, which can generate high-quality noise data to protect the privacy of the data.

[0063] A noise generation method based on differential privacy is applied to a data middle platform, which includes a data file library and a server. The data file library is used to store original data files and file tags corresponding to the original data files. The file tags are used to characterize the status of the data in the original data files. The data file library and the server are both in the same network.

[0064] The Data Middleware Platform (DMP) is an enterprise-level data management and application architecture model that plays a core role in the modern data-driven business environment. The goal of the DMP is to unify and optimize the management and use of data resources to support more efficient and flexible data analysis and business decision-making. Specifically, the DMP has the following features:

[0065] Data integration: The data center integrates and standardizes data from different sources and formats to form a unified data view, eliminating information silos and ensuring data consistency and accuracy.

[0066] Data governance: provides a complete set of data management processes, including data quality control, data security, and data lifecycle management, to ensure data compliance and availability;

[0067] Data analysis and services: The data center platform provides pre-processed data, models, and algorithms to support real-time analysis, predictive analysis, and intelligent decision-making, helping companies quickly respond to market changes.

[0068] Data sharing and reuse: Promotes cross-departmental and cross-system data sharing, reduces duplication of construction, increases data reuse rate, and reduces the cost of data acquisition and processing;

[0069] Flexibility and scalability: The data center adopts a microservices architecture, which can be easily expanded and adjusted to adapt to the rapid development and changes of the business;

[0070] Technology stack: The data center usually involves components such as big data processing technology, data warehouse, data lake, data API, and machine learning platform;

[0071] In some embodiments, the data file library is used to store original data files and file tags corresponding to the original data files, and the file tags are used to characterize the status of the data in the original data files. The data file library and the server are both in the same network to facilitate the management of original data files and data processing through the server.

[0072] like Figure 1 As shown, the specific process of the method can be as follows:

[0073] S110: Determine data sensitivity corresponding to the queried data according to data features of the queried data in the original data file.

[0074] In some embodiments, the queried data may refer to data accessed or retrieved by a specific query operation in a database or data set, and this data may be sensitive information such as user personal information, transaction records, medical data, etc.

[0075] In some embodiments, the data sensitivity of the queried data can refer to the maximum change in the query result when a record in the database changes. In differential privacy, data sensitivity is a key parameter, usually represented by Δf, which is used to determine the amount of noise added to protect the privacy of the data. Factors affecting data sensitivity include the following:

[0076] The nature or characteristics of the data being queried:

[0077] Different queried data has different data sensitivities. For example, the sensitivity of a count query is usually 1, because adding or deleting a record only changes the count by 1. The data sensitivity of a sum query depends on the maximum possible value of a single record. For numerical data, the data sensitivity depends on the range of the data. For example, if the value of the queried data of a record is between [0,100], the data sensitivity is 100.

[0078] Distribution of queried data:

[0079] If the values ​​of an attribute in the data are very sparse (i.e., most values ​​are zero or close to zero), the data sensitivity may be low; if the distribution of the data is very discrete, the data sensitivity may be high;

[0080] The structure of the queried data:

[0081] The number of records in the data affects data sensitivity. Generally speaking, the more records there are, the less impact a change in a single record has on the overall result, and the data sensitivity is likely to be lower. Sensitivity calculations for multidimensional data are more complex, as the impact of changes in each dimension on the result needs to be considered.

[0082] S120: Determine a privacy parameter corresponding to the queried data according to the queried data and the file tag.

[0083] In some embodiments, the file label includes a compliance level label, a confidentiality expectation label, and an application scenario label. The compliance level label is used to characterize the status of the content in the original data file being subject to industry-specific data protection standards. The confidentiality expectation label is used to characterize the degree of expectation of the user corresponding to the original data file for the confidentiality level of the original data file. The application scenario label is used to characterize the application scenario in which the content in the original data file is located.

[0084] Specifically, determining the privacy parameter corresponding to the queried data based on the queried data and the file tag includes the following specific implementation process:

[0085] By presetting the privacy parameter generation model, the privacy parameter is obtained according to the queried data, the compliance level label, the confidentiality expectation label and the application scenario label.

[0086] In some embodiments, the preset privacy parameter generation model includes a first input layer, a first embedding layer, a second embedding layer, a third embedding layer, a first feature fusion layer, and a fully connected layer.

[0087] Specifically, the model is generated by presetting privacy parameters, and obtaining the privacy parameters according to the queried data, the compliance level label, the confidentiality expectation label, and the application scenario label, including steps A1 to A6 as shown below:

[0088] A1. Obtain an input vector corresponding to the queried data according to the queried data through the first input layer.

[0089] In some embodiments, the queried data is input into the first input layer, and an input vector with a dimension of (1,) is output.

[0090] A2. Obtain a compliance level label vector through the first embedding layer and the compliance level label.

[0091] In some embodiments, the compliance level label may indicate whether the original data file complies with the data protection standards of a specific industry. Therefore, the label value of the compliance level label may be 0 (non-compliant), 1 (partially compliant) or 2 (fully compliant).

[0092] Specifically, the compliance level label is input into the first embedding layer, the embedding dimension of the first embedding layer is set to 3, and the embedding vector is [0.1, 0.2, 0.3], then a compliance level label vector with a dimension of (d1,) is output.

[0093] A3. Obtain a confidentiality expected label vector through the second embedding layer and the confidentiality expected label.

[0094] In some embodiments, the expected confidentiality tag may represent the user's expectation of the confidentiality level of the original data file. Therefore, the tag value of the expected confidentiality tag may be 0 (low), 1 (medium), or 2 (high).

[0095] Specifically, the confidentiality expected label is input into the second embedding layer, the embedding dimension of the second embedding layer is set to 2, and the embedding vector is [0.4, 0.5], then a confidentiality expected label vector with a dimension of (d2,) is output.

[0096] A4. Obtain an application scenario label vector through the third embedding layer and the application scenario label.

[0097] In some embodiments, the application scenario tag may represent the application scenario in which the content in the original data file is located. Therefore, the tag value of the application scenario tag may be 0 (research), 1 (application practice), or 2 (management).

[0098] Specifically, the application scenario label is input into the second embedding layer, the embedding dimension of the second embedding layer is set to 3, the embedding vector is [0.6, 0.7, 0.8], and an application scenario label vector with a dimension of (d3,) is output.

[0099] A5. Obtain a first feature fusion vector through the first feature fusion layer according to the input vector corresponding to the queried data, the compliance level label vector, the confidentiality expectation label vector, and the application scenario label vector.

[0100] Continuing with the above embodiment, the following vector is used as the input of the first feature fusion layer:

[0101] The input vector m corresponding to the queried data;

[0102] Compliance level label vector: [0.1, 0.2, 0.3];

[0103] Confidential expected label vector: [0.4, 0.5];

[0104] Application scenario label vector: [0.6, 0.7, 0.8].

[0105] Specifically, the output is the first feature fusion vector with a dimension of (1+d1+d2+d3,)=(1+3+2+3,)=(9,) and a shape of [m, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8].

[0106] A6. Obtain the privacy parameter according to the first feature fusion vector through the fully connected layer.

[0107] In some embodiments, the fully connected layer includes a first sub-fully connected layer, a second sub-fully connected layer, and a third sub-fully connected layer. The activation function of the first sub-fully connected layer and the activation function of the second sub-fully connected layer are both ReLU functions, and the activation function of the third sub-fully connected layer is a Sigmoid function.

[0108] A fully connected layer, also known as a dense layer, is one of the most fundamental components in a neural network model. In this layer, every node in the previous layer is connected to all nodes in this layer, forming a fully interconnected network structure. The primary function of a fully connected layer is to learn complex patterns in input data through a series of linear transformations coupled with nonlinear activation functions.

[0109] The ReLU function is a widely used activation function defined as follows: f(x) = max(0, x). This means that for all negative inputs, the output of the ReLU function is 0; for all positive inputs, the output is equal to the input itself. The advantages of the ReLU function include simple computation, the ability to alleviate the vanishing gradient problem, and the ability to accelerate the training process.

[0110] The Sigmoid function is another commonly used activation function. Its characteristic is its ability to compress any real value into the range (0, 1). This makes it particularly suitable for tasks that require outputting probability values, such as binary classification problems. Furthermore, the Sigmoid function's output curve is smooth and has good mathematical properties, facilitating gradient calculations.

[0111] Specifically, obtaining the privacy parameter according to the first feature fusion vector through the fully connected layer includes steps A61 to A63 as shown below:

[0112] A61. Obtain a first intermediate vector according to the first feature fusion vector through the first sub-fully connected layer.

[0113] Continuing with the above embodiment, the output dimension of the first sub-fully connected layer can be 16, and the input dimension of the first sub-fully connected layer is (9,). Assuming m = 1000, the ReLU function produces a first intermediate vector with a shape of [123.4, 567.8, 910.11, 1213.14, 1516.17, 1819.20, 2122.23, 2425.26, 2728.29, 3031.32, 3334.35, 3637.38, 3940.41, 4243.44, 4546.47, 4849.50].

[0114] A62. Obtain a second intermediate vector according to the first intermediate vector through the second sub-fully-connected layer.

[0115] Continuing with the above embodiment, the output dimension of the second sub-fully connected layer can be 8, and the input dimension of the second sub-fully connected layer is (16,). Through the action of the ReLU function, a second intermediate vector with a shape of [5152.53, 5455.56, 5758.59, 6061.62, 6364.65, 6667.68, 6970.71, 7273.74] is obtained.

[0116] A63. Obtain the privacy parameter according to the second intermediate vector through the third sub-fully connected layer.

[0117] Continuing with the above example, the output dimension of the third fully connected sub-layer is (1,), and the input dimension of the second fully connected sub-layer is (8,). A privacy parameter ε is obtained through the Sigmoid function, and its value is 0.1.

[0118] S130: Determine the scale parameter corresponding to the queried data according to the data sensitivity and privacy parameters.

[0119] In some embodiments, determining the scale parameter corresponding to the queried data based on the data sensitivity and privacy parameters includes the following specific implementation process:

[0120] The scale parameter is calculated based on a preset scale parameter calculation formula according to the data sensitivity and the privacy parameter.

[0121] In some embodiments, the scale parameter b may be a proportional factor of the standard deviation of the Laplace distribution, used to control the size of the added noise data. Therefore, the scale parameter b may be obtained by the following preset scale parameter calculation formula:

[0122] b=Δf / ε

[0123] In some embodiments, the scale parameter may be the standard deviation of noise generated by a Gaussian distribution, which may be represented by σ. The advantage of the Gaussian distribution is that it better conforms to the natural distribution characteristics of certain data in practical applications and also provides good privacy protection.

[0124] Specifically, to ensure that Gaussian noise satisfies (ε, δ)-differential privacy, the standard deviation σ can be obtained by the following preset scale parameter calculation formula:

[0125]

[0126] Wherein, Δf represents the data sensitivity, ε represents the privacy parameter, and δ represents the probability of failure of the preset privacy protection. Usually, δ is required to be very small. After the standard deviation σ is calculated, the standard deviation σ is determined as the scale parameter.

[0127] S140 , obtaining target noise data according to the scale parameter and the file label by using a preset noise generation model to protect the privacy of the queried data.

[0128] In some embodiments, the preset noise generation model includes a second input layer, a fourth embedding layer, a second feature fusion layer, a convolutional layer, an LSTM layer, a fourth sub-fully connected layer, a fifth sub-fully connected layer and a sixth sub-fully connected layer. The activation function of the fourth sub-fully connected layer and the activation function of the fifth sub-fully connected layer are both ReLU functions, and the activation function of the sixth sub-fully connected layer is a linear activation function.

[0129] Convolutional layers are the core component of convolutional neural networks (CNNs), primarily used to process data with a grid structure, such as images. Convolutional layers detect local features in the input data by applying a set of learnable filters (also called convolution kernels or kernels). These filters slide over the input data and perform dot multiplication operations, then sum the results to form a feature map.

[0130] The LSTM layer is a special recurrent neural network (RNN) layer designed to address the vanishing or exploding gradient problem of traditional RNNs when processing long sequences of data. LSTM controls the flow of information by introducing memory cells and three gating mechanisms: input gate, forget gate, and output gate.

[0131] The linear activation function is the simplest activation function, meaning that the output is linearly related to the input, without any nonlinear transformation. Linear activation functions are often used in the output layer of neural networks, especially in regression tasks, because the goal of regression tasks is to predict continuous values ​​and there is no need to constrain the output to a specific range.

[0132] In some embodiments, the method of obtaining target noise data based on the scale parameter and the file label by using a preset noise generation model to protect the privacy of the queried data includes steps B1 to B8 as shown below:

[0133] B1. Obtain a scale parameter vector according to the scale parameter through the second input layer.

[0134] In some embodiments, the scale parameter is input into the second input layer. Assuming that the input dimension and the output dimension are both (1,), and the scale parameter is 3777, the output scale parameter vector is represented as

[3777] .

[0135] B2. Obtain a file label vector according to the file label through the fourth embedding layer.

[0136] Continuing with the above embodiment, the output dimension corresponding to the fourth embedding layer is (8,), and the file label vector can be expressed as [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8].

[0137] B3. Obtain a second feature fusion vector according to the file label vector and the scale parameter vector through the second feature fusion layer.

[0138] Continuing with the above embodiment, the output dimension of the second feature fusion layer is (1+8,)=(9,), then the file label vector and the scale parameter vector are input into the second feature fusion layer, and the output second feature fusion vector can be expressed as [3777, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8].

[0139] B4. Obtain a convolution feature map through the convolution layer according to the second feature fusion vector.

[0140] Continuing with the above embodiment, the input dimension of the convolution layer is (9,) and the output dimension of the convolution layer is (H, W, C), wherein it can be assumed that the dimension of the feature map output by the convolution layer is (1, 1, 64), the convolution kernel size of the convolution layer is 3x3, the stride of the convolution layer is 1, and the sky supplement of the convolution layer is the same. Then, through the convolution layer, according to the second feature fusion vector, a convolution feature map with a dimension of (1, 1, 64) is obtained.

[0141] B5. Obtain an LSTM feature vector through the LSTM layer according to the convolutional feature map.

[0142] Continuing with the above embodiment, the output dimension of the LSTM layer can be set to (1, 64), and the convolutional feature map is input to the LSTM layer to obtain an LSTM feature vector represented as [123.4, 567.8, ..., 910.11] (length 64).

[0143] B6. Obtain a third intermediate vector based on the LSTM feature vector through the fourth sub-fully connected layer.

[0144] Continuing with the above embodiment, the input dimension of the fourth fully connected sub-layer is (64,), and the output dimension of the fourth fully connected sub-layer is (h1,), where h1 = 128. The LSTM feature vector is input to the fourth fully connected sub-layer, and a third intermediate vector represented as [1234.5, 6789.0, ..., 1112.3] (length 128) is output.

[0145] B7. Obtain a fourth intermediate vector according to the third intermediate vector through the fifth fully connected sub-layer.

[0146] Continuing with the above embodiment, the input dimension of the fifth fully-connected sub-layer is (128,), and the output dimension of the fourth fully-connected sub-layer is (h2,), where h2 = 64. The third intermediate vector is input to the fifth fully-connected sub-layer, and a fourth intermediate vector represented as [2345.6, 7890.1, ..., 2223.4] (length 64) is output.

[0147] B8. Obtain target noise data according to the fourth intermediate vector through the sixth sub-fully connected layer.

[0148] Continuing with the above embodiment, the input dimension of the fifth fully-connected sub-layer is (61,), and the output dimension of the fourth fully-connected sub-layer is (1,). The third intermediate vector is input to the fifth fully-connected sub-layer, and a linear activation function is applied to output a numerical value representing the target noise data.

[0149] In some embodiments, before determining the data sensitivity corresponding to the queried data based on the data features of the queried data in the original data file, the method further includes the following specific implementation process:

[0150] In response to the received data query instruction, the queried data is retrieved from the original data file in the data set file library.

[0151] Specifically, in response to the received data query instruction, retrieving the queried data from the original data file in the data set file library may include steps C1 to C4 as shown below:

[0152] C1. Receive data query instructions:

[0153] Data query commands are entered through the terminal, usually including query conditions and query types. Suppose a user sends a query command requesting the age and disease diagnosis results of a patient.

[0154] C2. Parsing data query instructions:

[0155] Parse the query instruction and extract the query conditions and query type. For example, extract the query condition as patient ID 12345 and the query type as obtaining age and disease diagnosis results.

[0156] C3. Locate the original data file based on the query conditions:

[0157] Based on the parsed query conditions, the corresponding raw data file is located in the data file library. For example, based on the patient ID 12345, the relevant file in the data file library is searched. Assume that the file found is named patient_12345.json.

[0158] C4. Retrieve the queried data:

[0159] Read the queried data from the located raw data file. For example, read the patient's age and disease diagnosis results from the file patient_12345.json.

[0160] According to the above steps, the queried data required by the user can be retrieved, and then the target noise data can be generated by the method to protect the privacy of the queried data.

[0161] As can be seen from the above, the embodiment of the present application can first determine the data sensitivity corresponding to the queried data based on the data characteristics of the queried data in the original data file; then, determine the privacy parameters corresponding to the queried data based on the queried data and the file label; then, determine the scale parameters corresponding to the queried data based on the data sensitivity and the privacy parameters; through a preset noise generation model, according to the scale parameters and the file label, finally obtain the target noise data to protect the privacy of the queried data.

[0162] In summary, compared with existing methods, the embodiments of the present application can accurately provide more suitable noise data for the protected data based on the user's requirements for confidentiality level and the specific actual scenario where the data is located, so that the protected data can be used for effective data analysis or application development while obtaining high-quality privacy protection.

[0163] To better implement the above method, the present application also provides a noise generation device based on differential privacy. The device can be integrated into an electronic device, such as a terminal or a server. The terminal can be a mobile phone, tablet computer, smart Bluetooth device, laptop computer, personal computer, etc. The server can be a single server or a server cluster consisting of multiple servers.

[0164] For example, in this embodiment, the method of the embodiment of the present application will be described in detail by taking the specific integration of the noise generation device based on differential privacy in the terminal as an example.

[0165] For example, Figure 2 As shown, the noise generation device based on differential privacy may include a first unit 201, a second unit 202, a third unit 203 and a fourth unit 204, which are applied to a data middle platform, wherein the data middle platform includes a data file library and a server, wherein the data file library is used to store original data files and file tags corresponding to the original data files, wherein the file tags are used to characterize the status of the data in the original data files, and the data file library and the server are both in the same network;

[0166] The device comprises:

[0167] The first unit 201 is configured to determine the data sensitivity corresponding to the queried data according to the data characteristics of the queried data in the original data file;

[0168] The second unit 202 is configured to determine a privacy parameter corresponding to the queried data based on the queried data and the file tag;

[0169] The third unit 203 is used to determine the scale parameter corresponding to the queried data according to the data sensitivity and the privacy parameter;

[0170] The fourth unit 204 is configured to obtain target noise data according to the scale parameter and the file label by using a preset noise generation model to protect the privacy of the queried data.

[0171] In specific implementation, the above units can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units can be found in the previous method embodiments and will not be repeated here.

[0172] From the above, it can be seen that the embodiment of the present application can use the data middle platform and the neural network model to accurately provide more suitable noise data for the protected data based on the user's requirements for the confidentiality level and the specific actual scenario where the data is located, so that the protected data can be used for effective data analysis or application development while obtaining high-quality privacy protection.

[0173] The present application also provides an electronic device, which may be a terminal, a server, or the like. The terminal may be a mobile phone, a tablet computer, a smart Bluetooth device, a laptop computer, a personal computer, or the like; the server may be a single server or a server cluster consisting of multiple servers, or the like.

[0174] In some embodiments, the noise generation device based on differential privacy can also be integrated into multiple electronic devices. For example, the noise generation device based on differential privacy can be integrated into multiple servers, and the noise generation method based on differential privacy of the present application is implemented by multiple servers.

[0175] In this embodiment, the electronic device of this embodiment is a terminal as an example for detailed description, for example, Figure 3 As shown, it shows a schematic diagram of the structure of the terminal involved in the embodiment of the present application, specifically:

[0176] The terminal may include one or more processing core processors 301, one or more media storages 302, a power supply 303, an input module 304, and a communication module 305. Those skilled in the art will appreciate that Figure 3 The terminal structure shown in the figure does not constitute a limitation to the terminal, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0177] in:

[0178] Processor 301 is the terminal's control center, connecting all components of the terminal using various interfaces and circuits. By running or executing software programs and / or modules stored in memory 302 and accessing data stored in memory 302, it performs various terminal functions and processes data, thereby providing overall terminal monitoring. In some embodiments, processor 301 may include one or more processing cores. In some embodiments, processor 301 may integrate an application processor and a modem processor. The application processor primarily handles the operating system, user interface, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 301.

[0179] The memory 302 can be used to store software programs and modules. The processor 301 executes various functional applications and data processing by running the software programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 302 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.

[0180] The terminal also includes a power supply 303 for supplying power to various components. In some embodiments, the power supply 303 can be logically connected to the processor 301 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 303 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0181] The terminal may further include an input module 304, which may be configured to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0182] The terminal may also include a communication module 305. In some embodiments, the communication module 305 may include a wireless module. The terminal may perform short-range wireless transmission via the wireless module of the communication module 305, thereby providing the user with wireless broadband Internet access. For example, the communication module 305 may be used to help the user send and receive emails, browse web pages, and access streaming media.

[0183] Although not shown, the terminal may further include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the terminal will load the executable files corresponding to the processes of one or more application programs into the memory 302 according to the following instructions, and the processor 301 will run the application programs stored in the memory 302 to implement various functions as follows:

[0184] Determining data sensitivity corresponding to the queried data based on data features of the queried data in the original data file;

[0185] Determining privacy parameters corresponding to the queried data based on the queried data and the file tag;

[0186] Determine the scale parameters corresponding to the queried data based on data sensitivity and privacy parameters;

[0187] By presetting a noise generation model, target noise data is obtained according to the scale parameter and the file label to protect the privacy of the queried data.

[0188] The specific implementation of the above operations can be found in the previous embodiments and will not be repeated here.

[0189] From the above, it can be seen that the embodiments of the present application can generate high-quality noise data to protect the privacy of the data.

[0190] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by instructions, or by controlling related hardware through instructions. The instructions may be stored in a medium and loaded and executed by a processor.

[0191] To this end, an embodiment of the present application provides a medium storing a plurality of instructions that can be loaded by a processor to execute the steps of any of the noise generation methods based on differential privacy provided in the embodiments of the present application. For example, the instructions can execute the following steps:

[0192] Determining data sensitivity corresponding to the queried data based on data features of the queried data in the original data file;

[0193] Determining privacy parameters corresponding to the queried data based on the queried data and the file tag;

[0194] Determine the scale parameters corresponding to the queried data based on data sensitivity and privacy parameters;

[0195] By presetting a noise generation model, target noise data is obtained according to the scale parameter and the file label to protect the privacy of the queried data.

[0196] The medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0197] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a medium. A processor of a computer device reads the computer instructions from the medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations provided in the above embodiments.

[0198] Since the instructions stored in the medium can execute the steps of any of the noise generation methods based on differential privacy provided in the embodiments of the present application, the beneficial effects that can be achieved by any of the noise generation methods based on differential privacy provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0199] The above is a detailed introduction to the noise generation method, device, terminal and medium based on differential privacy provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A noise generation method based on differential privacy, characterized in that: Applied to a data middle platform, the data middle platform includes a data file library and a server, the data file library is used to store original data files and file tags corresponding to the original data files, the file tags are used to characterize the status of the data in the original data files, and the data file library and the server are both in the same network; The method comprises: Determining data sensitivity corresponding to the queried data based on data features of the queried data in the original data file; Determining privacy parameters corresponding to the queried data based on the queried data and the file tag; Determine the scale parameters corresponding to the queried data based on data sensitivity and privacy parameters; Obtaining target noise data based on the scale parameter and the file label by using a preset noise generation model to protect the privacy of the queried data; The file tag includes a compliance level tag, a confidentiality expectation tag, and an application scenario tag. The compliance level tag is used to indicate whether the content in the original data file is applicable to industry-specific data protection standards. The confidentiality expectation tag is used to indicate the degree of expectation of the user corresponding to the original data file regarding the confidentiality level of the original data file. The application scenario tag is used to indicate the application scenario in which the content in the original data file is located. The determining, based on the queried data and the file tag, a privacy parameter corresponding to the queried data includes: Generate a model using preset privacy parameters to obtain the privacy parameters based on the queried data, the compliance level label, the confidentiality expectation label, and the application scenario label; The preset privacy parameter generation model includes a first input layer, a first embedding layer, a second embedding layer, a third embedding layer, a first feature fusion layer, and a fully connected layer; The model is generated by presetting privacy parameters, and obtaining the privacy parameters according to the queried data, the compliance level label, the confidentiality expectation label, and the application scenario label, including: Obtaining, through the first input layer, an input vector corresponding to the queried data according to the queried data; Obtaining a compliance level label vector through the first embedding layer and the compliance level label; Obtaining a confidentiality expected label vector through the second embedding layer and the confidentiality expected label; Obtaining an application scenario label vector through the third embedding layer and the application scenario label; Obtaining, through the first feature fusion layer, a first feature fusion vector according to the input vector corresponding to the queried data, the compliance level label vector, the confidentiality expectation label vector, and the application scenario label vector; Obtaining the privacy parameter according to the first feature fusion vector through the fully connected layer; The fully connected layer includes a first sub-fully connected layer, a second sub-fully connected layer and a third sub-fully connected layer, the activation function of the first sub-fully connected layer and the activation function of the second sub-fully connected layer are both ReLU functions, and the activation function of the third sub-fully connected layer is a Sigmoid function; Obtaining the privacy parameter according to the first feature fusion vector through the fully connected layer includes: Obtaining a first intermediate vector according to the first feature fusion vector through the first sub-fully connected layer; Obtaining a second intermediate vector according to the first intermediate vector through the second sub-fully connected layer; The privacy parameter is obtained according to the second intermediate vector through the third sub-fully connected layer.

2. The method according to claim 1, wherein Determining the scale parameters corresponding to the queried data based on the data sensitivity and privacy parameters includes: The scale parameter is calculated based on a preset scale parameter calculation formula according to the data sensitivity and the privacy parameter.

3. The method according to claim 1, wherein The preset noise generation model includes a second input layer, a fourth embedding layer, a second feature fusion layer, a convolutional layer, an LSTM layer, a fourth sub-fully connected layer, a fifth sub-fully connected layer, and a sixth sub-fully connected layer, wherein the activation function of the fourth sub-fully connected layer and the activation function of the fifth sub-fully connected layer are both ReLU functions, and the activation function of the sixth sub-fully connected layer is a linear activation function; The method of obtaining target noise data by using a preset noise generation model according to the scale parameter and the file label to protect the privacy of the queried data includes: Obtaining a scale parameter vector according to the scale parameter through the second input layer; Obtaining a file label vector according to the file label through the fourth embedding layer; Obtaining a second feature fusion vector according to the file label vector and the scale parameter vector through the second feature fusion layer; Obtaining a convolution feature map through the convolution layer according to the second feature fusion vector; Obtaining an LSTM feature vector according to the convolutional feature map through the LSTM layer; Obtaining a third intermediate vector according to the LSTM feature vector through the fourth fully connected sub-layer; Obtaining a fourth intermediate vector according to the third intermediate vector through the fifth fully connected sub-layer; Target noise data is obtained according to the fourth intermediate vector through the sixth sub-fully connected layer.

4. The method according to claim 1, wherein Before determining the data sensitivity corresponding to the queried data based on the data features of the queried data in the original data file, the method further includes: In response to the received data query instruction, the queried data is retrieved from the original data file in the data set file library.

5. A noise generation device based on differential privacy, characterized in that: Applied to a data middle platform, the data middle platform includes a data file library and a server, the data file library is used to store original data files and file tags corresponding to the original data files, the file tags are used to characterize the status of the data in the original data files, and the data file library and the server are both in the same network; The device comprises: The first unit is configured to determine the data sensitivity corresponding to the queried data according to the data characteristics of the queried data in the original data file; A second unit is configured to determine a privacy parameter corresponding to the queried data based on the queried data and the file tag; The third unit is used to determine the scale parameter corresponding to the queried data based on the data sensitivity and privacy parameters; A fourth unit is configured to obtain target noise data based on the scale parameter and the file label by using a preset noise generation model to protect the privacy of the queried data; The file tag includes a compliance level tag, a confidentiality expectation tag, and an application scenario tag. The compliance level tag is used to indicate whether the content in the original data file is applicable to industry-specific data protection standards. The confidentiality expectation tag is used to indicate the degree of expectation of the user corresponding to the original data file regarding the confidentiality level of the original data file. The application scenario tag is used to indicate the application scenario in which the content in the original data file is located. The determining, based on the queried data and the file tag, a privacy parameter corresponding to the queried data includes: Generate a model using preset privacy parameters to obtain the privacy parameters based on the queried data, the compliance level label, the confidentiality expectation label, and the application scenario label; The preset privacy parameter generation model includes a first input layer, a first embedding layer, a second embedding layer, a third embedding layer, a first feature fusion layer, and a fully connected layer; The model is generated by presetting privacy parameters, and obtaining the privacy parameters according to the queried data, the compliance level label, the confidentiality expectation label, and the application scenario label, including: Obtaining, through the first input layer, an input vector corresponding to the queried data according to the queried data; Obtaining a compliance level label vector through the first embedding layer and the compliance level label; Obtaining a confidentiality expected label vector through the second embedding layer and the confidentiality expected label; Obtaining an application scenario label vector through the third embedding layer and the application scenario label; Obtaining, through the first feature fusion layer, a first feature fusion vector according to the input vector corresponding to the queried data, the compliance level label vector, the confidentiality expectation label vector, and the application scenario label vector; Obtaining the privacy parameter according to the first feature fusion vector through the fully connected layer; The fully connected layer includes a first sub-fully connected layer, a second sub-fully connected layer and a third sub-fully connected layer, the activation function of the first sub-fully connected layer and the activation function of the second sub-fully connected layer are both ReLU functions, and the activation function of the third sub-fully connected layer is a Sigmoid function; Obtaining the privacy parameter according to the first feature fusion vector through the fully connected layer includes: Obtaining a first intermediate vector according to the first feature fusion vector through the first sub-fully connected layer; Obtaining a second intermediate vector according to the first intermediate vector through the second sub-fully connected layer; The privacy parameter is obtained according to the second intermediate vector through the third sub-fully connected layer.

6. A terminal, characterized in that: The method comprises a processor and a memory, wherein the memory stores a plurality of instructions; the processor loads instructions from the memory to execute the steps in the method according to any one of claims 1 to 4.

7. A medium, characterized in that The medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the method according to any one of claims 1 to 4.

Images