A method, apparatus, device, medium, and product for multidimensional data privacy computing.

Abstract

This application discloses a method, apparatus, device, medium, and product for multidimensional data privacy computation, relating to the fields of data security and privacy protection. The method includes: a user terminal acquiring a user privacy level, allocating a personalized privacy budget based on the user privacy level, and determining the privacy budget allocation result; the user terminal adaptively selecting a privacy protection perturbation mechanism based on an adaptive privacy boundary and the privacy budget allocation result to perform local perturbation, obtaining a perturbation dataset formed by perturbing the user's multidimensional data; and an aggregation server using the perturbation dataset to perform multidimensional data joint distribution estimation, obtaining a joint probability distribution estimation result. This application balances user privacy protection, data utility accuracy control, and system scalability, achieving efficient and controllable multidimensional data privacy computation.

Description

Technical Field

[0001] This application relates to the field of data security and privacy protection, and in particular to a multidimensional data privacy computing method, apparatus, device, medium and product. Background Technology

[0002] With the widespread deployment of data-driven intelligent systems, the collection and analysis of massive amounts of multidimensional, discrete, and sensitive data pose a serious challenge to the leakage of personal privacy. To achieve localized privacy protection in a distributed environment, Local Differential Privacy (LDP), as one of the mainstream privacy protection models, has received widespread attention from academia and industry because it can achieve strict privacy protection without relying on trusted third parties. It has also been deployed and validated in commercial systems such as Google Chrome (Rappor), Apple iOS, and Windows Telemetry.

[0003] However, current local differential privacy mechanisms still face the following key technical bottlenecks when dealing with multidimensional discrete data application scenarios:

[0004] 1. Personalized privacy budget allocation lacks the ability to express differences at the dimensional level. Mainstream LDP mechanisms generally use a uniform privacy budget to perturb all data dimensions, ignoring the asymmetry of different dimensional attributes in terms of user privacy preferences and sensitivities. This results in the inability to effectively express personalized privacy needs, leading to uncontrollable loss of data utility.

[0005] 2. The perturbation mechanism configuration is rigid and lacks an accuracy-oriented adaptive mechanism selection strategy. Current LDP mechanisms often use fixed perturbation operators (such as GRR, OUE, or OLH), which lack the ability to dynamically switch mechanisms under diverse privacy budget conditions. This makes it difficult to optimize perturbation errors in a targeted manner, affecting the overall system stability and output accuracy.

[0006] 3. High-dimensional joint distribution modeling faces the dual challenges of the curse of dimensionality and estimation sparsity. In high-dimensional discrete spaces, attribute combinations grow exponentially, and data sparsity is exacerbated after perturbation, causing existing frequency recovery methods (such as frequent term estimation and marginal combination inference) to encounter significant bottlenecks in terms of modeling accuracy, computational complexity, and stability.

[0007] Although existing studies such as LoPub, LDPMiner, and PM have proposed some improved schemes, using strategies such as constructing a joint coding space through Bloom filters and hierarchical sensitivity control for optimization, the above methods still have the following technical shortcomings:

[0008] 1. Unable to dynamically adapt to users' diverse and heterogeneous privacy budget needs.

[0009] 2. It is difficult to achieve strategic guidance that minimizes errors in the selection of disturbance mechanisms.

[0010] 3. Lack of joint modeling mechanisms applicable to sparse estimation in high-dimensional attribute domains.

[0011] Therefore, it is crucial to achieve efficient and controllable privacy computing while balancing privacy protection, precise control of data utility, and system scalability without the need for trusted intermediaries. Summary of the Invention

[0012] The purpose of this application is to provide a multidimensional data privacy computing method, apparatus, device, medium, and product that balances privacy protection, data utility precision control, and system scalability, thereby achieving efficient and controllable multidimensional data privacy computing.

[0013] To achieve the above objectives, this application provides the following solution:

[0014] Firstly, this application provides a multidimensional data privacy computation method, which is implemented using a privacy processing system; the privacy processing system includes: a user terminal and an aggregation server; the user terminal and the aggregation server are connected via network communication; the multidimensional data privacy computation method includes:

[0015] The user's privacy level is obtained based on the user terminal, and a personalized privacy budget is allocated based on the user's privacy level to determine the privacy budget allocation result;

[0016] Based on the user terminal, the adaptive privacy boundary and privacy budget allocation results are calculated, and the privacy protection perturbation mechanism is adaptively selected to perform local perturbation, thereby obtaining a perturbation dataset formed by perturbing the user's multidimensional data.

[0017] The aggregation server performs multidimensional data joint distribution estimation based on the perturbation dataset to obtain a joint probability distribution estimation result. This joint probability distribution estimation result is used to achieve privacy-preserving computation and privacy protection in multidimensional data application scenarios. The joint probability distribution estimation result can be applied to privacy-preserving computation scenarios such as mobile crowdsourcing and collective sensing, achieving the realization of data value while protecting user privacy.

[0018] In one embodiment, the user's privacy level is obtained based on the user terminal, and a personalized privacy budget is allocated based on the user's privacy level to determine the privacy budget allocation result, specifically including:

[0019] Obtaining user privacy levels based on user terminals;

[0020] Personalized privacy budget allocation is performed based on the user's privacy level, and normalization is performed using a combination of linear scaling and interval projection to obtain the privacy budget allocation result.

[0021] In one implementation, personalized privacy budget allocation is performed based on the user's privacy level, using the following allocation rules:

[0022] ;

[0023] A normalization process using a combination of linear scaling and interval projection must satisfy the overall privacy budget constraint; the expression corresponding to the overall privacy budget constraint is:

[0024] ;

[0025] in, For users For attribute dimensions Privacy level; For average privacy budget; The budget results of the allocation; The number of attribute dimensions.

[0026] In one embodiment, based on the user terminal's adaptive privacy boundary and privacy budget allocation results, an adaptive privacy protection perturbation mechanism is selected to perform local perturbation, resulting in a perturbation dataset formed by perturbing the user's multidimensional data, specifically including:

[0027] The user terminal adaptively selects a privacy protection perturbation mechanism based on the calculated adaptive privacy boundary and the privacy budget allocation results; where, if If so, then the GRR mechanism is selected; if If so, then the OUE mechanism should be selected; The budget results of the allocation; To adapt to privacy boundaries;

[0028] Local perturbations are performed based on the selected perturbation mechanism, specifically including:

[0029] user The perturbation is performed using the GRR mechanism, with the perturbation probability... Preserve original values The corresponding position is 1, and the rest are 0; otherwise, the perturbation probability is used. from Randomly select one position and set it to 1, and set the rest to 0 to generate a perturbation vector. ; For attribute dimensions The number of candidate values; at this point, the attribute dimension superior mThe expression corresponding to the perturbation probability of a bit is:

[0030] ;

[0031] user The OUE mechanism is used to perturb the encoded vector. Each bit in the array is subjected to independent perturbation. If the bit is 1, then the perturbation probability is applied. Reserved as 1. Flip to 0; if the bit is 0, then perturb with probability. Flip to 1, Set to 0 to generate a perturbation vector. At this point, the attribute dimension superior m The expression corresponding to the perturbation probability of a bit is:

[0032] ;

[0033] Based on the user terminal, the user All attribute dimensions are perturbed, and the perturbed multidimensional data is concatenated to form a perturbation vector. And based on the multidimensional perturbation data of each user, a perturbation dataset is formed; the perturbation dataset is based on... The multidimensional perturbation data of each user were determined.

[0034] In one implementation, the adaptive privacy boundary is determined by calculating the minimum mean square error of a combination of a generalized random response mechanism and an optimal unary coding mechanism;

[0035] The expression for adaptive privacy boundaries is:

[0036] ;

[0037] in, To adapt to privacy boundaries; For attribute dimensions The number of candidate values.

[0038] In one embodiment, the joint probability distribution estimation result is obtained by performing multidimensional data joint distribution estimation based on the perturbation dataset on the aggregation server side, specifically including:

[0039] Based on the aggregation server, the frequency of each candidate value of each attribute in the perturbation dataset is calculated. ;

[0040] The candidate values ​​of different attributes are connected by a Cartesian product and then transposed to form a candidate value matrix. M ;

[0041] Based on frequency and candidate value matrix Constructing a linear system: ,in, P Let be the joint probability distribution vector to be estimated, and let the dimension correspond to all combinations of candidate values;

[0042] The linear system is solved using the LASSO regression method. The joint distribution of multidimensional data is estimated by minimizing the optimization objective function, and the joint probability distribution estimation result is obtained.

[0043] The expression corresponding to the optimization objective function is:

[0044] ;

[0045] in, To optimize the objective function; These are the joint probability estimation coefficients; The observation frequency of each candidate value in the perturbed data as statistically analyzed by the aggregation server; These are regularization control parameters; The mean square error between the observed frequency and the model predicted frequency; To The sparsity constraint causes most elements to be 0, thus suppressing noise.

[0046] Secondly, this application provides a multidimensional data privacy computing device, comprising:

[0047] The privacy budget allocation module is used to obtain the user's privacy level based on the user terminal, and to allocate a personalized privacy budget based on the user's privacy level, and to determine the privacy budget allocation result;

[0048] The perturbation module is used to adaptively select a privacy protection perturbation mechanism based on the user terminal's calculated adaptive privacy boundary and privacy budget allocation results to perform local perturbation and obtain a perturbation dataset formed by perturbing the user's multidimensional data.

[0049] The joint distribution estimation module is used to perform multidimensional data joint distribution estimation based on the perturbation dataset on the aggregation server side, and obtain the joint probability distribution estimation result. The joint probability distribution estimation result is used to realize privacy computing and privacy protection in multidimensional data application scenarios.

[0050] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multidimensional data privacy computing method described above.

[0051] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multidimensional data privacy computation method described above.

[0052] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the multidimensional data privacy computing method described above.

[0053] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0054] This application discloses a method, apparatus, device, medium, and product for multidimensional data privacy computing. The method includes: a user terminal acquiring a user privacy level, allocating a personalized privacy budget based on the user privacy level, and determining the privacy budget allocation result; the user terminal adaptively selecting a privacy protection perturbation mechanism based on the calculated adaptive privacy boundary and the privacy budget allocation result to perform local perturbation, obtaining a perturbation dataset formed by perturbing the user's multidimensional data; and an aggregation server using the perturbation dataset to perform multidimensional data joint distribution estimation, obtaining a joint probability distribution estimation result. The joint distribution estimation of multidimensional data can be applied to privacy computing scenarios such as mobile crowdsourcing and collective sensing, achieving the realization of data value through user privacy protection. The method in this application balances user privacy protection, data utility precision control, and system scalability, achieving efficient and controllable multidimensional data privacy computing. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart of a multidimensional data privacy computation method;

[0057] Figure 2 A schematic diagram of the overall module and key steps of a multidimensional data privacy computation method;

[0058] Figure 3 The experimental results comparing the method with the PLDP model on simulated data are presented.

[0059] Figure 4 The experimental results comparing the proposed method with the PLDP model for average variational distance (AVD) on a real dataset are presented.

[0060] Figure 5 A structural diagram of a multidimensional data privacy computing device;

[0061] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0062] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0063] This application introduces a user privacy level matrix to achieve dimensional-level differentiated budget mapping under a total budget constraint. It constructs a perturbation mechanism based on mean squared error theory to switch boundaries, enabling adaptive collaborative configuration of the Generalized Randomized Response (GRR) and Optimized Unary Encoding (OUE) mechanisms. Furthermore, it introduces the LASSO sparse regression algorithm on the server side for high-dimensional joint frequency recovery, effectively suppressing noise accumulation and highlighting core statistical patterns. This method balances privacy protection, accuracy control, and system scalability, making it suitable for various data-intensive privacy computing scenarios and demonstrating promising engineering application prospects.

[0064] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0065] In one exemplary embodiment, a multidimensional data privacy computation method is provided, which is implemented using a privacy processing system. The privacy processing system includes a user terminal and an aggregation server. The user terminal and the aggregation server are connected via network communication.

[0066] like Figure 1 As shown, the multidimensional data privacy computation method includes:

[0067] Step 100: Obtain the user's privacy level based on the user terminal, and allocate a personalized privacy budget based on the user's privacy level to determine the privacy budget allocation result.

[0068] In one embodiment, the user's privacy level is obtained based on the user terminal, and a personalized privacy budget is allocated based on the user's privacy level to determine the privacy budget allocation result, specifically including:

[0069] The user's privacy level is obtained based on the user terminal; a personalized privacy budget is allocated based on the user's privacy level, and a normalization process is performed using a combination of linear scaling and interval projection to obtain the privacy budget allocation result.

[0070] Personalized privacy budget allocation is based on user privacy levels, and the allocation rules are as follows:

[0071] .

[0072] A normalization process combining linear scaling and interval projection is employed, which must satisfy the overall privacy budget constraint; the expression corresponding to the overall privacy budget constraint is:

[0073] .

[0074] in, For users For attribute dimensions Privacy level; For average privacy budget; The budget results of the allocation; The number of attribute dimensions.

[0075] Step 200: Based on the user terminal's adaptive privacy boundary and privacy budget allocation results, adaptively select a privacy protection perturbation mechanism to perform local perturbation, and obtain a perturbation dataset formed by perturbing the user's multidimensional data.

[0076] The adaptive privacy boundary is determined by calculating the minimum mean square error of the generalized random response mechanism and the optimal unary coding mechanism (optimized unary coding mechanism); the expression corresponding to the adaptive privacy boundary is:

[0077] .

[0078] in, To adapt to privacy boundaries; For attribute dimensions The number of candidate values.

[0079] In one embodiment, based on the user terminal's adaptive privacy boundary and privacy budget allocation results, an adaptive privacy protection perturbation mechanism is selected to perform local perturbation, resulting in a perturbation dataset formed by perturbing the user's multidimensional data, specifically including:

[0080] The user terminal adaptively selects a privacy protection perturbation mechanism based on the calculated adaptive privacy boundary and the privacy budget allocation results; where, if If so, then the GRR mechanism is selected; if If so, then the OUE mechanism should be selected; The budget results of the allocation; To adapt to privacy boundaries.

[0081] Local perturbations are performed based on the selected perturbation mechanism, specifically including:

[0082] user The perturbation is performed using the GRR mechanism, with the perturbation probability... Preserve original values The corresponding position is 1, and the rest are 0; otherwise, the perturbation probability is used. from Randomly select one position and set it to 1, and set the rest to 0 to generate a perturbation vector. ; For attribute dimensions The number of candidate values; at this point, the attribute dimension superior m The expression corresponding to the perturbation probability of a bit is:

[0083] .

[0084] user The OUE mechanism is used to perturb the encoded vector. Each bit in the array is subjected to independent perturbation. If the bit is 1, then the perturbation probability is applied. Reserved as 1. Flip to 0; if the bit is 0, then perturb with probability. Flip to 1, Set to 0 to generate a perturbation vector. At this point, the attribute dimension superior m The expression corresponding to the perturbation probability of a bit is:

[0085] .

[0086] Based on the user terminal, the user All attribute dimensions are perturbed, and the perturbed multidimensional data is concatenated to form a perturbation vector. And based on the multidimensional perturbation data of each user, a perturbation dataset is formed; the perturbation dataset is based on... The multidimensional perturbation data of each user were determined.

[0087] Step 300: Based on the aggregation server, perform multidimensional data joint distribution estimation using the perturbation dataset to obtain the joint probability distribution estimation result. Multidimensional data joint distribution estimation can be applied to privacy-preserving computing scenarios such as mobile crowdsourcing and collective intelligence sensing, achieving the realization of data value while protecting user privacy.

[0088] In one embodiment, the joint probability distribution estimation result is obtained by performing multidimensional data joint distribution estimation based on the perturbation dataset on the aggregation server side, specifically including:

[0089] Based on the aggregation server, the frequency of each candidate value of each attribute in the perturbation dataset is calculated. The candidate values ​​of different attributes are connected by a Cartesian product and then transposed to form a candidate value matrix. Based on frequency and candidate value matrix Constructing a linear system: Where P is the joint probability distribution vector to be estimated, and the dimension corresponds to all candidate values; the linear system is solved by the LASSO regression method, and the joint distribution of multidimensional data is estimated by minimizing the optimization objective function to obtain the joint probability distribution estimation result.

[0090] The expression corresponding to the optimization objective function is:

[0091] .

[0092] in, To optimize the objective function; These are the joint probability estimation coefficients; The observation frequency of each candidate value in the perturbed data as statistically analyzed by the aggregation server; These are regularization control parameters; The mean square error between the observed frequency and the model predicted frequency; To The sparsity constraint causes most elements to be 0, thus suppressing noise.

[0093] This application proposes an adaptive personalized local differential privacy multidimensional data privacy computing method for privacy computing scenarios such as mobile crowdsourcing and collective sensing, including user local data perturbation terminals and aggregation computing server systems. This method combines user personalized privacy protection needs to design an adaptive personalized privacy computing method for multidimensional data, solving the problems of over-protection and under-protection, as well as reduced data utility, caused by the equivalent split privacy budget strategy in traditional multidimensional data privacy computing. The specific contents of this method include: (1) introducing the expression of user personalized privacy protection needs, and designing an adaptive privacy budget allocation mechanism under the premise of protecting user privacy under the overall privacy budget constraint. (2) constructing a minimum mean square error optimization model based on mean square error (MSE) to measure data utility, solving the privacy budget bound, and designing an adaptive privacy protection method with dynamic selection of perturbation mechanism. (3) designing a joint probability distribution estimation based on the LASSO method for the aggregation stage of privacy computing. The application of this method includes local random perturbation of user terminals and aggregation and distribution estimation on the aggregation server side. It is an efficient and controllable privacy computing method without the need for trusted intermediaries.

[0094] This application presents an adaptive, personalized, local differential privacy multidimensional data privacy computation method, primarily involving a user terminal and an aggregation server. It comprises three key modules, executed collaboratively by the user terminal and the aggregation server, and does not rely on a trusted third-party entity. The overall module structure and key steps of the designed method are as follows: Figure 2 As shown.

[0095] In practical applications, the specific execution process is as follows:

[0096] Module 1: The user terminal obtains the user's privacy level and completes the personalized privacy budget allocation.

[0097] S1: The system includes Each user owns [number] users. Dimensional attributes, user terminal obtains user For attributes Users are labeled with a privacy sensitivity level of 1-4 to form a privacy level vector. ,in User Privacy level standards, User For attributes Privacy levels, characterizing users Attribute dimension The higher the value, the greater the privacy sensitivity. The number of attribute dimensions.

[0098] S2: System Settings Average privacy budget for dimensional attributes ,user according to and For each attribute dimension Allocate a personalized privacy budget in accordance with the following formula (1):

[0099] (1)

[0100] S3: The user terminal normalizes the initial privacy budget allocation result using a combination of linear scaling and interval projection to ensure it meets the overall privacy budget constraint. This ensures that the privacy budget is allocated reasonably and that operations are efficient.

[0101] Module 2: The user terminal adaptively selects a perturbation mechanism based on the adaptive privacy boundary and performs local perturbation.

[0102] S1: User terminal for each attribute dimension Based on the number (quantity) of its candidate values. Calculate adaptive privacy boundaries This enables adaptive selection of GRR and OUE perturbation mechanisms. It is calculated based on the minimum mean squared error (MSE). The process is as follows:

[0103] Combining the implementation details of the Generalized Randomized Response (GRR) and Optimized Unary Encoding (OUE) mechanisms, the formula (2) is constructed respectively. and Measuring the two mechanisms in budget The mean squared error is as follows:

[0104] (2)

[0105] in, The mean squared error corresponding to GRR; The mean squared error corresponding to OUE; The number of candidate values ​​for a data dimension; For the first candidate values The actual probability of occurrence.

[0106] structure ,make , eliminate This yields a closed-form solution that is independent of the actual data distribution. .

[0107] S2: User of the user terminal Based on the allocated budget results With adaptive privacy boundaries The relationship is based on the attribute dimension. The strategy for adaptively selecting the optimal perturbation mechanism is:

[0108] like Choose the GRR mechanism; if Choose the OUE mechanism.

[0109] Specifically, the user terminal processes the original discrete data. Perform one-hot encoding to obtain the encoded vector. This encoding method ensures data structure consistency before and after the perturbation, facilitating subsequent processing. Then, the user terminal employs GRR and OUE mechanisms; the implementation details of the local random perturbation are as follows:

[0110] 1. User Perturbation using GRR mechanism The perturbation probability of m bits in the attribute dimension is given by formula (3):

[0111] (3)

[0112] Using the above perturbation probabilities, the user terminal first uses probability Preserve original values The position corresponding to [m] is 1, and the rest are 0; otherwise, it is determined by probability. from Randomly select one position and set it to 1, and set the rest to 0 to generate a perturbation vector. .

[0113] 2. User Perturbation using OUE mechanism The perturbation probability of m bits in the attribute dimension is given by formula (4):

[0114] (4)

[0115] The user terminal perturbation process is a process of encoding vector Each bit in the array is subjected to an independent perturbation; if the bit is 1, then the perturbation is performed with probability. Reserved as 1. Flip to 0; if the bit is 0, then proceed with probability. Flip to 1, Set to 0 to generate a perturbation vector. .

[0116] S3: The user terminal will connect to the user The result of perturbation of all attribute dimensions splicing together to form a perturbation-based multidimensional data record It is sent to the aggregation server.

[0117] Module 3: Joint distribution estimation on the aggregation server side.

[0118] S1: The aggregation server receives data from... Perturbation dataset for individual users Then, the frequency of each candidate value in the perturbation data is counted. ,in, ; For the first i The number of candidate values ​​for a dimension attribute.

[0119] S2: Connect the candidate values ​​of different attributes through Cartesian product and transpose them to form a candidate matrix. abbreviation M , It is the number of candidate value combinations. The candidate value frequency combination of dimensional data is represented as a vector. and according to the relationship Calculation .

[0120] S3: The aggregation server uses the LASSO regression method to estimate the joint probability distribution and constructs an optimization objective function. .in, ,and , representing the joint probability estimation coefficients, This is the regularization control parameter.

[0121] S4: Aggregate the estimated joint probability distribution results from the server-side output. The joint distribution estimation results of privacy computing can be applied to various scenarios such as mobile crowdsourcing, collective intelligence sensing, and personalized recommendation with privacy protection in multidimensional data privacy computing.

[0122] This application primarily addresses the shortcomings of existing multidimensional data privacy protection technologies, such as insufficient personalization of privacy protection, poor adaptability of perturbation mechanisms, and low accuracy of high-dimensional joint distribution estimation. Effective improvements and enhancements have been made to achieve the following technical effects:

[0123] (1) Enhance personalized privacy protection capabilities and availability of low-sensitivity data.

[0124] By introducing a budget allocation strategy based on user privacy level vectors in step S1 of module 1, the terminal device can allocate personalized local differential privacy budgets for each dimension according to the sensitivity differences of different attribute dimensions. This budget allocation mechanism not only realizes personalized expression of privacy protection under the constraint of total budget, but also significantly improves the usability of low-sensitivity dimensions, thereby improving the accuracy of global joint frequency estimation. This advantage directly stems from the privacy level mapping and budget normalization method described in step S1 of module 1 of this method.

[0125] (2) Achieve adaptive optimal switching of the disturbance mechanism to reduce estimation error.

[0126] In step S2, the perturbation mechanism switching boundary is obtained through theoretical derivation, and based on this boundary, the generalized random response mechanism (GRR) or the optimized unary coding mechanism (OUE) is adaptively selected to ensure that the perturbation method with the minimum mean square error is used under different privacy budget levels. This mechanism switching boundary is derived from the critical intersection of GRR and OUE in terms of mean square error, eliminating the error fluctuation problem caused by manual mechanism selection and improving the perturbation accuracy and consistency. This advantage stems from the mechanism selection boundary calculation and mechanism switching strategy described in modules S1 and S2 of the technical solution Module 2.

[0127] according to Figure 3 and Figure 4 It can be seen that, compared with the fixed mechanism (PLDP-GRR, PLDP-OUE) and the non-personalized perturbation method (GRR, OUE), the adaptive method (P-ADLDP) proposed in this application maintains a lower joint distribution estimation error under different privacy budget levels, which is particularly evident under low budget, proving that it is better in balancing privacy protection and accuracy. Figure 4 (1)-(5) in the diagram represent the relationship between data dimension and average variational distance under different privacy budget levels.

[0128] (3) Efficiently recover high-dimensional joint distribution and improve modeling accuracy and stability.

[0129] In step S3 of module 3, the server employs the Lasso regression method to recover the joint distribution of perturbation frequencies in the joint candidate space, effectively addressing the problem of the exponential growth in the number of candidate combinations in high-dimensional space. The Lasso method, by introducing an L1 regularization term, can adaptively model the sparse structure in the joint distribution, improving the estimation accuracy of high-frequency patterns and automatically suppressing low-frequency noise terms, thereby enhancing the model's stability and generalization ability.

[0130] While ensuring differential privacy constraints, this application significantly improves the flexibility of privacy protection, perturbation accuracy, and modeling usability in multidimensional data processing through innovative designs in key aspects such as budget allocation, perturbation mechanism selection, and joint distribution recovery. It has good practical value and engineering application prospects.

[0131] In one exemplary embodiment, such as Figure 5 As shown, a multidimensional data privacy computing device is provided, comprising:

[0132] The privacy budget allocation module is used to obtain the user's privacy level based on the user's terminal, and to allocate a personalized privacy budget based on the user's privacy level, thereby determining the privacy budget allocation result.

[0133] The perturbation module is used to adaptively select a privacy protection perturbation mechanism based on the user terminal's calculated adaptive privacy boundary and privacy budget allocation results to perform local perturbation and obtain a perturbation dataset formed by perturbing the user's multidimensional data.

[0134] The joint distribution estimation module is used to perform multidimensional data joint distribution estimation based on the perturbation dataset on the aggregation server side, and obtain the joint probability distribution estimation result. The joint probability distribution estimation is used to realize privacy computing and privacy protection in multidimensional data application scenarios.

[0135] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores multidimensional data privacy computation data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a multidimensional data privacy computation method.

[0136] Those skilled in the art will understand that Figure 6The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0137] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0138] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0139] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0140] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0141] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0142] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0143] 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.

[0144] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multidimensional data privacy computation method, characterized in that, A privacy processing system is used; the privacy processing system includes: a user terminal and an aggregation server; the user terminal and the aggregation server are connected via network communication. The multidimensional data privacy computation method includes: The user's privacy level is obtained based on the user terminal, and a personalized privacy budget is allocated based on the user's privacy level to determine the privacy budget allocation result; Based on the user terminal, the adaptive privacy boundary and privacy budget allocation results are calculated, and the privacy protection perturbation mechanism is adaptively selected to perform local perturbation, thereby obtaining a perturbation dataset formed by perturbing the user's multidimensional data. Based on the aggregation server, a multidimensional data joint distribution estimation is performed on the perturbation dataset to obtain the joint probability distribution estimation result; Based on the user terminal's adaptive privacy boundary and privacy budget allocation results, an adaptive privacy protection perturbation mechanism is selected to perform local perturbation, resulting in a perturbation dataset formed by perturbing the user's multidimensional data, specifically including: The user terminal adaptively selects a privacy protection perturbation mechanism based on the calculated adaptive privacy boundary and the privacy budget allocation results; where, if If so, then the GRR mechanism is selected; if If so, then the OUE mechanism should be selected; The budget results of the allocation; To adapt to privacy boundaries; Local perturbations are performed based on the selected perturbation mechanism, specifically including: user The perturbation is performed using the GRR mechanism, with the perturbation probability... Preserve original values The corresponding position is 1, and the rest are 0; otherwise, the perturbation probability is used. from Randomly select one position and set it to 1, and set the rest to 0 to generate a perturbation vector. ; For attribute dimensions The number of candidate values; at this point, the attribute dimension superior m The expression corresponding to the perturbation probability of a bit is: ； Encoding vector The mth position; To remove the m-th bit from 1 to A set of sequences; user The OUE mechanism is used to perturb the encoded vector. Each bit in the array is subjected to independent perturbation. If the bit is 1, then the perturbation probability is applied. Reserved as 1. Flip to 0; if the bit is 0, then perturb with probability. Flip to 1, Set to 0 to generate a perturbation vector. At this point, the attribute dimension superior m The expression corresponding to the perturbation probability of a bit is: ； Based on the user terminal, the user All attribute dimensions are perturbed, and the perturbed multidimensional data is concatenated to form a perturbation vector. And based on the multidimensional perturbation data of each user, a perturbation dataset is formed; the perturbation dataset is based on... Determined by multidimensional perturbation data of individual users; The adaptive privacy boundary is determined by calculating the minimum mean square error of the combination of the generalized random response mechanism and the optimal unary coding mechanism; The expression for adaptive privacy boundaries is: ； in, To adapt to privacy boundaries; For attribute dimensions The number of candidate values.

2. The multidimensional data privacy computation method according to claim 1, characterized in that, The user's privacy level is obtained based on the user terminal, and a personalized privacy budget is allocated based on the user's privacy level to determine the privacy budget allocation result, specifically including: Obtaining user privacy levels based on user terminals; Personalized privacy budget allocation is performed based on the user's privacy level, and normalization is performed using a combination of linear scaling and interval projection to obtain the privacy budget allocation result.

3. The multidimensional data privacy computation method according to claim 2, characterized in that, Personalized privacy budget allocation is performed based on the user's privacy level, using the following allocation rules: ； A normalization process using a combination of linear scaling and interval projection must satisfy the overall privacy budget constraint; the expression corresponding to the overall privacy budget constraint is: ； in, For users For attribute dimensions Privacy level; For average privacy budget; The budget results of the allocation; The number of attribute dimensions.

4. The multidimensional data privacy computation method according to claim 1, characterized in that, Based on the multidimensional joint distribution estimation of the perturbation dataset performed by the aggregation server, the joint probability distribution estimation results are obtained, specifically including: Based on the aggregation server, the frequency of each candidate value of each attribute in the perturbation dataset is calculated. ; The candidate values ​​of different attributes are connected by a Cartesian product and then transposed to form a candidate value matrix. ; Based on frequency and candidate value matrix Constructing a linear system: ,in, P Let be the joint probability distribution vector to be estimated, and let the dimension correspond to all combinations of candidate values; The linear system is solved using the LASSO regression method. The joint distribution of multidimensional data is estimated by minimizing the optimization objective function, and the joint probability distribution estimation result is obtained. The expression corresponding to the optimization objective function is: ； in, To optimize the objective function; These are the joint probability estimation coefficients; The observation frequency of each candidate value in the perturbed data as statistically analyzed by the aggregation server; These are regularization control parameters; The mean square error between the observed frequency and the predicted frequency; To Sparsity constraints.

5. A multidimensional data privacy computing device, characterized in that, include: The privacy budget allocation module is used to obtain the user's privacy level based on the user terminal, and to allocate a personalized privacy budget based on the user's privacy level, and to determine the privacy budget allocation result; The perturbation module is used to adaptively select a privacy protection perturbation mechanism based on the user terminal's calculated adaptive privacy boundary and privacy budget allocation results to perform local perturbation and obtain a perturbation dataset formed by perturbing the user's multidimensional data. The joint distribution estimation module is used to perform multidimensional data joint distribution estimation based on the perturbation dataset on the aggregation server side, and obtain the joint probability distribution estimation result; Based on the user terminal's adaptive privacy boundary and privacy budget allocation results, an adaptive privacy protection perturbation mechanism is selected to perform local perturbation, resulting in a perturbation dataset formed by perturbing the user's multidimensional data, specifically including: The user terminal adaptively selects a privacy protection perturbation mechanism based on the calculated adaptive privacy boundary and the privacy budget allocation results; where, if If so, then the GRR mechanism is selected; if If so, then the OUE mechanism should be selected; The budget results of the allocation; To adapt to privacy boundaries; Local perturbations are performed based on the selected perturbation mechanism, specifically including: user The perturbation is performed using the GRR mechanism, with the perturbation probability... Preserve original values The corresponding position is 1, and the rest are 0; otherwise, the perturbation probability is used. from Randomly select one position and set it to 1, and set the rest to 0 to generate a perturbation vector. ; For attribute dimensions The number of candidate values; at this point, the attribute dimension superior m The expression corresponding to the perturbation probability of a bit is: ； Encoding vector The mth position; To remove the m-th bit from 1 to A set of sequences; user The OUE mechanism is used to perturb the encoded vector. Each bit in the array is subjected to independent perturbation. If the bit is 1, then the perturbation probability is applied. Reserved as 1. Flip to 0; if the bit is 0, then perturb with probability. Flip to 1, Set to 0 to generate a perturbation vector. At this point, the attribute dimension superior m The expression corresponding to the perturbation probability of a bit is: ； Based on the user terminal, the user All attribute dimensions are perturbed, and the perturbed multidimensional data is concatenated to form a perturbation vector. And based on the multidimensional perturbation data of each user, a perturbation dataset is formed; the perturbation dataset is based on... Determined by multidimensional perturbation data of individual users; The adaptive privacy boundary is determined by calculating the minimum mean square error of the combination of the generalized random response mechanism and the optimal unary coding mechanism; The expression for adaptive privacy boundaries is: ； in, To adapt to privacy boundaries; For attribute dimensions The number of candidate values.

6. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the multidimensional data privacy computing method according to any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the multidimensional data privacy computation method as described in any one of claims 1-4.

8. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the multidimensional data privacy computation method as described in any one of claims 1-4.

Images