Personalized differential privacy data sharing method and system based on visitor attributes

Abstract

The invention discloses a personalized differential privacy data sharing method and system based on visitor attributes. The method comprises the following steps: a cloud service provider transmits a visitor attribute domain to a data owner; a quantifiable hierarchical access structure is constructed based on the visitor attribute domain and the data needing to be shared, the information entropy of the attribute of the data to be shared is calculated, and the information is transmitted to a cloud service provider for storage; the visitor generates a data query function and sends the data query function to the cloud service provider; generating a matching index based on the matching degree between the visitor and the access structure; generating a privacy budget calculation matrix based on the matching index and the information entropy of the data attribute; and calculating the privacy budget of the visitor based on the privacy budget calculation matrix, and adding noise into a query result for data sharing. Based on the technical scheme, an attacker can be prevented from accessing the data, and the availability and sharing efficiency of the shared data are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer science and technology, and particularly to a personalized differential privacy data sharing method and system based on visitor attributes. Background Art

[0002] The rapid development of cloud computing and Internet of Things technologies has promoted the growth of various types of data. By sharing data with device manufacturers or relevant researchers, users can enjoy various efficient services, such as remote diagnosis and treatment, condition monitoring, and disease prediction. However, the data shared by users includes a lot of private information, such as diagnosis and treatment records, health conditions, etc., which brings the risk of leakage of private information while enjoying these services. Once these sensitive data are leaked, it will have a serious negative impact on the user's life. Therefore, how to protect the privacy of users during the data sharing process while ensuring the service quality has become an urgent problem to be solved.

[0003] To protect users' personal privacy, currently, technologies such as access control, data encryption, and data perturbation are mainly adopted. Among them, differential privacy (DP), as the main method of data perturbation, has been widely used in data privacy protection due to its low computational overhead. DP protects users' private information by adding noise to the data, and its protection strength depends on the amount of noise. However, excessive noise will reduce the usability of the data and it is difficult to meet the data usage requirements of users.

[0004] To solve this problem, personalized differential privacy (PDP) has been proposed, which can meet the data privacy and usage requirements of users in practical applications. Currently, scholars have improved PDP from multiple perspectives such as data providers and data sensitivity. However, only meeting the needs of data providers will reduce the usability of the shared data; and the sensitivity of the data often varies due to different visitors, and the existing solutions still lack research on it in complex scenarios. Therefore, it is necessary to discuss the balance between data privacy and usage requirements based on personalized differential privacy from the perspective of visitors.

[0005] In addition, during the data sharing process, it is also crucial to ensure the efficiency of the sharing stage. Different from traditional differential privacy methods, personalized differential privacy must dynamically adjust the amount of noise when processing each user's query according to their privacy requirements, thus requiring additional computing resources to calculate and allocate the privacy budget for each visitor. However, in a cloud computing environment, this additional computational overhead will greatly increase the burden on the system and affect the efficiency of data sharing. Therefore, when faced with a large number of visitors or complex scenarios, how to reasonably manage and allocate the privacy budget to avoid performance bottlenecks caused by overly heavy budget calculations has become a major challenge currently. Summary of the Invention

[0006] The present invention discloses a personalized differential privacy data sharing method and system based on visitor attributes. The method constructs a hierarchical access structure associated with the data, realizes precise control over the access permissions of the accessed data, and adaptively determines the privacy level of the accessed data for different visitors by quantifying the matching degree between the visitor and the access structure, improving the usability of the shared data. In addition, the solution designs a lightweight privacy budget calculation matrix based on the matching degree to determine the privacy budget, reducing the computational overhead caused by calculating visitor attributes and improving the system efficiency.

[0007] For this reason, the present invention provides the following technical solutions:

[0008] On the one hand, a personalized differential privacy data sharing method based on visitor attributes provided by the present invention includes the following steps:

[0009] Step 1: Initialization operation, the cloud service provider CSP transmits the visitor attribute domain to the data owner DO.

[0010] Step 2: Construct an access structure, construct a quantifiable hierarchical access control structure based on the visitor attribute domain and the data to be shared, and calculate the information entropy of each column of data attributes in the data to be shared, and transmit the access structure, the data to be shared, and their information entropy set to the cloud service provider CSP for storage.

[0011] Step 3: Initiate a query, the visitor DV generates a data query function and sends it to the cloud service provider CSP.

[0012] Step 4: Quantification, after receiving the query, the cloud service provider CSP generates a series of matching indices according to the matching degree between the attribute values of the visitor DV and the access structure.

[0013] Step 5: Generate a matrix, generate a privacy budget calculation matrix based on the matching indices and the information entropy of the data attributes.

[0014] Step 6: Data sharing. Calculate the privacy budget of the visitor DV based on the privacy budget calculation matrix, and add noise corresponding to the privacy budget to the query result for data sharing.

[0015] Further optionally, the process of constructing the access structure is as follows:

[0016] First, the data owner DO selects a series of value domain subsets from the visitor attribute domain AD as leaf nodes lnode x , indicating that visitors with this attribute can obtain authorization for this node;

[0017] Then, the data owner DO corresponds each column D in the data i to the non - leaf nodes nnode of the access structure y and assigns an access threshold number r y to it, which is the minimum number of child nodes that a visitor needs to satisfy to obtain this column of data;

[0018] After that, the data owner DO calculates the information entropy Ent(D i ) of each column of data attributes, and normalizes the result to obtain the data column information entropy set CE for subsequent construction of the privacy budget calculation matrix;

[0019] The calculation process of the data attribute information entropy Ent(D i ) is as follows:

[0020]

[0021] In the formula, n is the number of sample classes in D, C j is the number of samples in the j - th class, is the probability of the j - th state occurring in the sample set;

[0022] The calculation process of CE is as follows:

[0023]

[0024] Among them, CE i refers to the information entropy of the i - th column data set;

[0025] Finally, the data owner DO takes {lnode, nnode, r} as the access structure Γ, and transmits the access structure Γ, the data D to be shared, and the information entropy CE of the data attributes to the cloud service provider CSP.

[0026] Further optionally, the quantization process is as follows:

[0027] First, the cloud service provider CSP traverses all the leaf nodes lnode of the access structure Γ using the attributes A of the visitor DV; if A and lnode x have an intersection, then mark this leaf node as TRUE, that is:

[0028]

[0029] Then, the cloud service provider CSP checks the access threshold r of all non-leaf nodes from bottom to top; if the number of its child nodes marked as TRUE is greater than or equal to r x , then mark this non-leaf node as TRUE:

[0030]

[0031] where, are the child nodes of node x;

[0032] Finally, a matching index MI is defined in each non-leaf node, representing the matching degree between the visitor attribute A and the access structure Γ:

[0033]

[0034] where, is the number of child nodes of node x; for non-leaf nodes with t = FALSE, MI unauth , it is easy to get MI unauth = 0.

[0035] Further optionally, based on the matching index and information entropy, the process of generating the matrix is as follows:

[0036] The cloud service provider CSP represents the matching index MI and the information entropy CE as a matrix P:

[0037]

[0038] Considering the problem of inconsistent element indices in the two columns of the matrix, the cloud service provider CSP unifies the two columns of elements in the matrix to the maximum index:

[0039]

[0040] where is the conversion result, and then the conversion matrix

[0041]

[0042] Let γ be the number after removing the unauthorized data columns, and BCM can be expressed as:

[0043]

[0044] The cloud service provider CSP removes the data column that is not authorized to be accessed, i.e., x i1 For the case of =0, the privacy budget calculation matrix BCM is obtained.

[0045] Further optionally, the process of data sharing is as follows:

[0046] First, the CSP solves the group utility value S:

[0047] BCM + ={1, max(y 12 , y 22 , …, y γ2 )}

[0048] BCM - ={0, min(y 12 , y 22 , …, y γ2 )}

[0049]

[0050] S i represents the group effect value of the i-th row of the matrix; w j is set by the CSP and is the weight of the j-th column; BCM + and BCM - are the positive ideal solution and the negative ideal solution of the matrix respectively, representing the optimal element and the worst element in each column;

[0051] Then, the CSP solves the individual regret value R. The individual regret value R focuses on the worst parameter in each row of the matrix. The smaller the R value, the closer the worst parameter in that row is to the ideal state:

[0052]

[0053] where R i represents the individual regret value of the i-th row;

[0054] After that, the CSP solves the compromise solution value C to find a compromise choice between the group utility value S and the individual regret value R, and generates the privacy budget matrix, which is defined as:

[0055]

[0056] where k is the decision coefficient used to balance the influence of the group utility value S and the individual regret value R on C, and is set by the CSP;

[0057] The CSP generates the privacy budget matrix BM according to ε i =exp(-C i )

[0058]

[0059] Finally, the CSP processes the query of the DV, outputs the query result, and runs the differential privacy algorithm to output the result Ans;

[0060]

[0061] Where S(f) is the sensitivity of the query function f. The CSP uses the Laplace mechanism to add noise to the query result according to the above formula and shares the output result Ans with the DV.

[0062] In a second aspect, the present invention provides a system for a personalized differential privacy data sharing method based on visitor attributes, which includes: three entities: a data owner DO, a visitor DV, and a cloud service provider CSP;

[0063] Wherein, the data owner DO and the data visitor DV are both communicatively connected to the cloud server CSP;

[0064] The cloud service provider CSP transmits a visitor attribute domain to the data owner DO, and the attribute domain includes visitor attribute categories and optional values;

[0065] The data owner DO constructs a quantifiable hierarchical access control structure based on the visitor attribute domain and the data to be shared, and calculates the information entropy of each column of data attributes in the data to be shared, and transmits the access structure, the data to be shared, and their information entropy set to the cloud service provider CSP for storage;

[0066] The visitor DV generates a data query function and sends it to the cloud service provider CSP;

[0067] After receiving the query, the cloud service provider CSP generates a series of matching indices according to the matching degree between the attribute values of the visitor DV and the access structure;

[0068] The cloud service provider CSP generates a privacy budget calculation matrix based on the matching indices and the information entropy of the data attributes;

[0069] The cloud service provider CSP calculates the privacy budget of the visitor DV based on the privacy budget calculation matrix and adds noise corresponding to the privacy budget to the query result for data sharing.

[0070] In a third aspect, the present invention provides a system for a personalized differential privacy data sharing method based on visitor attributes, including:

[0071] Access structure construction module: used to construct a quantifiable hierarchical access control structure, calculate the information entropy of each column of data attributes in the data to be shared, and transmit the access structure, the data to be shared, and its information entropy set to the cloud service provider CSP for storage;

[0072] Matching index quantization module: generate a series of matching indexes according to the matching degree between the attribute values of the visitor DV and the access structure;

[0073] Privacy budget matrix generation module: used to generate a privacy budget calculation matrix based on the matching index and the information entropy of the data attributes;

[0074] Data sharing module: used to calculate the privacy budget of the visitor DV based on the privacy budget calculation matrix, and add noise corresponding to the privacy budget to the query result for data sharing.

[0075] Fourthly, the present invention provides an electronic terminal, at least including:

[0076] One or more processors;

[0077] A memory storing one or more computer programs;

[0078] Wherein, the processor calls the computer program to execute:

[0079] The steps of a personalized differential privacy data sharing method based on visitor attributes.

[0080] Fifthly, the present invention provides a computer-readable storage medium storing a computer program, and the computer program is called by the processor to execute:

[0081] The steps of a personalized differential privacy data sharing method based on visitor attributes.

[0082] Beneficial effects

[0083] The technical solution of the present invention enables attackers not to obtain dataset information through the access control structure, realizing flexible and precise accessor identity control. In addition, the present invention further optimizes the solution, including: reducing the dimension of the calculation matrix, that is, removing inaccessible data attributes, so as to improve the efficiency of the algorithm; using the matrix to calculate the privacy budget, making the accessor attributes related to the data they access, so as to improve the pertinence of privacy protection; the privacy standard of the matrix can be adjusted through parameters, providing as accurate data query results as possible for the accessor while protecting privacy. Therefore, the usability of the shared data is improved. Description of the drawings

[0084] Figure 1 It is a schematic diagram of the system model of the present invention;

[0085] Figure 2 It is a schematic diagram of the access structure provided by the present invention;

[0086] Figure 3 It is the root mean square error comparison of shared data, where the query function in Figure (a) is sum, and the query function in Figure (b) is count;

[0087] Figure 4 It is the comparison of the privacy budget allocated for each data attribute in the scheme;

[0088] Figure 5 It is the computational overhead of the step of constructing the access structure, where Figure (a) has 15 visitor attributes and Figure (b) has 50 visitor attributes;

[0089] Figure 6 It is the computational overhead of the quantization step, where Figure (a) has 15 visitor attributes and Figure (b) has 50 visitor attributes;

[0090] Figure 7 It is the computational overhead of the step of generating the matrix and data sharing, where Figure (a) has 15 visitor attributes and Figure (b) has 50 visitor attributes. Detailed implementation manners

[0091] The technical solution of the present invention provides a personalized differential privacy data sharing method and system based on visitor attributes. This algorithm realizes flexible identity control of data sharing visitors; and improves the efficiency of the algorithm by reducing the dimension of the computational matrix, that is, removing inaccessible data attributes; and uses matrix calculation for the privacy budget to link visitor attributes with the data they access, improving the pertinence of privacy protection; and adjusts the privacy standard of the matrix through parameters to provide visitors with as accurate data query results as possible while protecting privacy, improving the availability of shared data. The present invention will be further described below in conjunction with embodiments.

[0092] Embodiment 1:

[0093] As Figure 1 shown is the system model framework provided by the embodiment of the present invention. This model mainly includes three entities: including a data owner (DO), a visitor (DV), and a cloud service provider (CSP). Specifically, the descriptions of each entity are as follows:

[0094] (1) Data owner DO: The DO aims to determine the privacy level of the shared data according to the visitor. First, the DO obtains the domain of the visitor attributes and selects its subdomain to generate the access structure. Then, the DO calculates the information entropy of the data attributes. Finally, the DO uploads the access structure, data, and entropy to the CSP for data sharing.

[0095] (2) Visitor DV: DV is the entity that queries data. It registers on the cloud service provider CSP, submits a query request, and obtains the perturbed result of the data query.

[0096] (3) Cloud service provider CSP: The CSP is responsible for providing cloud storage services for data owners and data query services for visitors. After receiving the query and the visitor's attributes, the CSP checks whether the visitor's attributes meet the access structure. Then, it outputs a privacy budget calculation matrix according to the matching degree between the visitor and the structure and the information entropy. Finally, the CSP calculates the privacy budget for the visitor based on this matrix and performs personalized information sharing.

[0097] As Figure 1 shown in the system model framework provided by the present invention, the specific implementation of the method mainly includes six steps: initialization operation, constructing an access structure, initiating a query, quantization, generating a matrix, and data sharing. The main working process is as follows:

[0098] Step 1: Initialization operation - Setup(AD): The cloud service provider CSP transmits the visitor attribute domain AD to the data owner DO.

[0099] Step 2: Constructing an access structure - Construct(AD, D): The data owner DO inputs the visitor attribute domain AD and the data D to be shared, and flexibly sets the access structure Γ according to its own needs. The data owner DO calculates the information entropy of each column of data attributes in the data to be shared. The data owner DO transmits the access structure Γ, the data D to be shared, and its information entropy set CE to the cloud service provider CSP for storage.

[0100] The specific execution process is as follows:

[0101] First, the data owner DO selects a series of value range subsets from the visitor attribute domain AD x as leaf nodes lnode

[0102] , indicating that a visitor with this attribute can obtain the authorization of this node. i Then, the data owner DO corresponds each column D y in the data to the non - leaf nodes nnode y in the access structure and assigns an access threshold number r

[0103] as the minimum number of child nodes that a visitor needs to satisfy to obtain this column of data. i After that, the data owner DO calculates the information entropy Ent(D

[0104] The calculation process of the information entropy Ent(D i ) is as follows:

[0105]

[0106] In the formula, n is the number of sample classes in D, and C j is the number of samples in the j-th class, is the probability of the j-th state occurring in the sample set;

[0107] The calculation process of CE is as follows:

[0108]

[0109] Among them, CE i refers to the information entropy of the i-th column data set.

[0110] Finally, the data owner DO takes {lnode, nnode, r} as the access structure Γ, and transmits the access structure Γ, the data D to be shared, and the information entropy CE of the data attributes to the cloud service provider CSP.

[0111] Step 3: Initiate query - Query(f, A): The visitor DV generates a data query function f and sends it together with its own visitor attribute A to the cloud service provider CSP.

[0112] Step 4: Quantify - Quantify(A, f, Γ): The CSP inputs the attribute A of the visitor DV and the query function f, then quantifies the matching degree between A and Γ, and finally outputs a matching index MI.

[0113] The specific execution process is as follows:

[0114] First, the cloud service provider CSP traverses all leaf nodes lnode of the access structure Γ using the attribute A of the visitor DV. If A and lnode x have an intersection, then mark this leaf node as TRUE, that is:

[0115]

[0116] Then, the cloud service provider CSP checks the access threshold r of all non-leaf nodes from bottom to top. If the number of its child nodes marked as TRUE is greater than or equal to r x , then mark this non-leaf node as TRUE, that is:

[0117]

[0118] Among them, are the child nodes of node x;

[0119] Finally, in each non - leaf node, a matching index MI is defined to represent the matching degree between the attribute A of the visitor DV and the access structure Γ:

[0120]

[0121] where is the number of children of node x. For non - leaf nodes with t = FALSE, MI unauth , it is easy to get MI unauth = 0.

[0122] The specific algorithm for this step is shown in Algorithm 1.

[0123]

[0124]

[0125] Step 5: Generate matrix - GenBCM(MI, CE): The cloud service provider CSP constructs an initial matrix by combining the aforementioned matching index MI and the information entropy CE, and unifies the dimension of the matrix through normalization. Then, the data attributes that do not meet the access requirements are removed to obtain the privacy budget calculation matrix BCM.

[0126] The specific execution process is as follows:

[0127] The cloud service provider CSP represents MI and the information entropy CE as a matrix P:

[0128]

[0129] Considering the problem of inconsistent element indices in the two columns of the matrix, the cloud service provider CSP unifies the two - column elements in the matrix to the maximum index:

[0130]

[0131] where is the conversion result, and then the conversion matrix

[0132]

[0133] Let γ be the number after removing the unauthorized data columns, and BCM can be expressed as:

[0134]

[0135] The cloud service provider CSP removes the data columns that are not authorized to be accessed, that is, the case where x i1 = 0, to obtain the privacy budget calculation matrix BCM.

[0136] Step 6: Data Sharing - Share(BCM, f, D, k, w): The cloud service provider inputs the privacy budget calculation matrix BCM of the visitor DV, calculates the privacy budget matrix BM of the visitor DV by combining the weight w and the decision coefficient k. Finally, the cloud service provider adds the corresponding noise to the query result according to BM and the query function f of the visitor DV and performs data sharing.

[0137] The specific execution process is as follows:

[0138] First, the CSP solves the group utility value S:

[0139] BCM + = {1, max(y 12 , y 22 , …, y γ2 )} (11)

[0140] BCM - = {0, min(y 12 , y 22 , …, y γ2 )} (12)

[0141]

[0142] where S i represents the group effect value of the i-th row of the matrix. w j is set by the CSP and is the weight of the j-th column. BCM + and BCM - are the positive ideal solution and the negative ideal solution of the matrix respectively, representing the optimal and the worst elements in each column.

[0143] Then, the CSP solves the individual regret value R. The individual regret value R focuses on the worst parameter in each row of the matrix. The smaller the R value, the closer the worst parameter in that row is to the ideal state:

[0144]

[0145] where, R i represents the individual regret value of the i-th row.

[0146] After that, the CSP solves the compromise solution value C to find a compromise choice between the group effect value S and the individual regret value R and generates the privacy budget matrix. Its definition is:

[0147]

[0148] where k is the decision coefficient used to balance the influence of the group effect value S and the individual regret value R on C, and is set by the CSP

[0149] The CSP determines according to εi = exp(-C i ), generate the privacy budget matrix BM, whose form is

[0150]

[0151] Finally, the CSP processes the query of the visitor DV, outputs the query result, and runs the differential privacy algorithm to output the result Ans.

[0152]

[0153] Among them, S(f) is the sensitivity of the query function f. The CSP adds noise to the query result using the above formula through the Laplace mechanism and shares the output result Ans with the DV.

[0154] The specific algorithm for this step is shown in Algorithm 2.

[0155]

[0156] It should be understood that the above embodiments are the best embodiments of the present invention, which are the most prominent in terms of security and performance. In other feasible embodiments, without departing from the concept of the present invention, that is, through the access control structure, attackers cannot obtain dataset information, realizing flexible and precise access control of visitors. And by reducing the dimension of the calculation matrix, that is, removing inaccessible data attributes, the efficiency of the algorithm is improved. And the privacy budget is calculated using the matrix, making the visitor attributes related to the data they access, improving the pertinence of privacy protection. And by adjusting the privacy standard of the matrix through parameters, while protecting privacy, providing as accurate data query results as possible for visitors, improving the availability of shared data. The deletion or replacement of some steps in other feasible embodiments also falls within the protection scope of the present invention.

[0157] The security and performance of the best embodiment of the present invention will be analyzed as follows:

[0158] (1) Security analysis

[0159] The following proves that the method satisfies the definition of differential privacy.

[0160] Suppose the privacy budget matrix is BM, the privacy budget for a certain query is ε. f(D) represents the query function, D′ is the adjacent dataset of D, h(x) is the probability density function, and x is the similar output of f(D′) and f(D). Then, it can be obtained that:

[0161]

[0162] Also because:

[0163] S(f) = max||f(D) - f(D′)|| (20)

[0164] So:

[0165]

[0166] where max(ε) is the maximum privacy budget in BM. Therefore, the method of this application satisfies the definition of ε-differential privacy and can protect the privacy of shared data.

[0167] (2) Performance analysis

[0168] Analyze the theoretical performance of the method for the calculation problems of the main steps in the method. Assume that the dimension of the dataset is row D ×col D , and the derivation is as follows:

[0169] In the step of constructing the access structure, the method needs to calculate the information entropy of each column in the data, and the time complexity of this step is O(row D ×col D ).

[0170] In the quantization step, the method first traverses all leaf nodes to determine t x . Assume that there are lnode leaf nodes in the tree, then its time complexity is O(lnode). Then, the method traverses all non-leaf nodes to determine s(x) and MI. Assume that there are nnode non-leaf nodes in the tree, then its time complexity is O(nnode). Let's assume that there are a total of N nodes in the tree, then the total time complexity of this stage is O(N).

[0171] In the step of generating the matrix, the method needs to unify the dimension of the P matrix, and its computational complexity depends on the number of rows of the matrix, that is, O(col D ).

[0172] In the data sharing step, the method operates on the matrix BCM. The dimension of BCM is γ×2, then its computational complexity is O(γ).

[0173] In summary, the time complexity of each run of the method is O(row D ×col D +N+γ). Among them, since the dimension of the matrix BCM is not greater than that of the matrix P, so γ≤col D . At the same time, the step of constructing the access structure is only run once for the same dataset. Therefore, in the actual data sharing application, the computational complexity of the method when facing a large number of visitors can be reduced to the first-order level, that is, O(col D +N).

[0174] Example 2:

[0175] An embodiment of the present invention further provides a system for implementing the above method, including a Data Owner (DO), a Data Visitor (DV), and a Cloud Service Provider (CSP).

[0176] As Figure 1 shown, both the Data Owner (DO) and the Data Visitor (DV) are communicatively connected to the Cloud Service Provider (CSP). The Cloud Service Provider (CSP) transmits a visitor attribute domain to the Data Owner (DO), and the attribute domain includes visitor attribute categories and optional values; the Data Owner (DO) constructs a quantifiable hierarchical access control structure based on the visitor attribute domain and the data to be shared, and calculates the information entropy of each column of data attributes in the data to be shared, and transmits the access structure, the data to be shared, and their information entropy set to the Cloud Service Provider (CSP) for storage; the Data Visitor (DV) generates a data query function and sends it to the Cloud Service Provider (CSP); after receiving the query, the Cloud Service Provider (CSP) generates a series of matching indices according to the matching degree between the attribute value of the Data Visitor (DV) and the access structure; the Cloud Service Provider (CSP) generates a privacy budget calculation matrix based on the matching indices and the information entropy of the data attributes; the Cloud Service Provider (CSP) calculates the privacy budget of the Data Visitor (DV) based on the privacy budget calculation matrix, and adds noise corresponding to the privacy budget to the query result for data sharing.

[0177] Embodiment 3

[0178] An embodiment of the present invention further provides a system for implementing the above method, including an initialization operation module, an access structure construction module, a query initiation module, a matching index quantification module, a privacy budget matrix generation module, and a data sharing module.

[0179] Among them, the initialization operation module is used to enable the Cloud Service Provider (CSP) to transmit a visitor attribute domain to the Data Owner (DO); the access structure construction module is used to construct a quantifiable hierarchical access control structure, and calculate the information entropy of each column of data attributes in the data to be shared, and transmit the access structure, the data to be shared, and their information entropy set to the Cloud Service Provider (CSP) for storage; the query initiation module is used to send the data query function generated by the data visitor to the Cloud Service Provider (CSP); the matching index quantification module generates a series of matching indices according to the matching degree between the attribute value of the Data Visitor (DV) and the access structure; the privacy budget matrix generation module is used to generate a privacy budget calculation matrix based on the matching indices and the information entropy of the data attributes; the data sharing module is used to calculate the privacy budget of the Data Visitor (DV) based on the privacy budget calculation matrix, and add noise corresponding to the privacy budget to the query result for data sharing.

[0180] It should be understood that the implementation process of each module can refer to the content description of the foregoing method. The division of the above functional modules is only a division of logical functions. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. At the same time, the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0181] Embodiment 4

[0182] An embodiment of the present invention provides an electronic terminal, which at least includes: one or more processors; a memory storing one or more computer programs; wherein, the processor calls the computer program to execute: the steps of a personalized differential privacy data sharing method based on visitor attributes.

[0183] Specifically execute:

[0184] Initialization operation, the cloud service provider CSP transmits the visitor attribute domain to the data owner DO;

[0185] Construct an access structure, construct a quantifiable hierarchical access control structure based on the visitor attribute domain and the data to be shared, and calculate the information entropy of each column of data attributes in the data to be shared, and transmit the access structure, the data to be shared and its information entropy set to the cloud service provider CSP for storage;

[0186] Initiate a query, the visitor DV generates a data query function and sends it to the cloud service provider CSP;

[0187] Quantification, after receiving the query, the cloud service provider CSP generates a series of matching indexes according to the matching degree between the attribute value of the visitor DV and the access structure;

[0188] Generate a matrix, generate a privacy budget calculation matrix based on the matching index and the information entropy of the data attributes;

[0189] Data sharing, calculate the privacy budget of the visitor DV based on the privacy budget calculation matrix, and add noise corresponding to the privacy budget to the query result for data sharing.

[0190] For the specific implementation process of each step, please refer to the description of the foregoing method.

[0191] It should be understood that in the embodiments of the present invention, the so-called processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0192] It should be emphasized that the examples described in the present invention are illustrative rather than restrictive. Therefore, the present invention is not limited to the examples described in the specific embodiments. Any other embodiments obtained by those skilled in the art according to the technical solutions of the present invention, whether modified or replaced, as long as they do not depart from the purpose and scope of the present invention, also belong to the protection scope of the present invention.

[0193] Embodiment 5

[0194] The embodiments of the present invention provide a computer-readable storage medium storing a computer program, and the computer program is called by a processor to execute the steps of a personalized differential privacy data sharing method based on visitor attributes.

[0195] Specifically execute:

[0196] Initialization operation: The cloud service provider CSP transmits the visitor attribute domain to the data owner DO.

[0197] Construct an access structure: Based on the visitor attribute domain and the data to be shared, construct a quantifiable hierarchical access control structure, and calculate the information entropy of each column of data attributes in the data to be shared, and transmit the access structure, the data to be shared and its information entropy set to the cloud service provider CSP for storage.

[0198] Initiate a query: The visitor generates a data query function and sends it to the cloud service provider CSP.

[0199] Quantification: After receiving the query, the cloud service provider CSP generates a series of matching indices according to the matching degree between the attribute values of the visitor DV and the access structure.

[0200] Generate a matrix, and generate a privacy budget calculation matrix based on the matching index and the information entropy of data attributes;

[0201] Data sharing, calculate the privacy budget of the visitor DV based on the privacy budget calculation matrix, and add noise corresponding to the privacy budget to the query result for data sharing.

[0202] For the specific implementation process of each step, please refer to the description of the foregoing method.

[0203] The readable storage medium is a computer-readable storage medium, which may be an internal storage unit of the controller described in any of the foregoing embodiments, such as the hard disk or memory of the controller. The readable storage medium may also be an external storage device of the controller, such as a plug-in hard disk equipped on the controller, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the readable storage medium may also include both the internal storage unit of the controller and the external storage device. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium may also be used to temporarily store data that has been output or will be output.

[0204] Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. And the foregoing readable storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0205] Experimental result analysis

[0206] (1) Preparation

[0207] Experimental environment: All experiments were conducted on a Windows 11 system equipped with an Intel i7@1.80GHz CPU and 16GB of RAM, and the code was written in Python.

[0208] Experimental dataset: The Adult dataset is a widely used personal information dataset that contains 14 attributes, including age, education, and occupation. In addition, this application uses a synthetic dataset to simulate visitor attributes.

[0209] To ensure the reliability of the experiment, multiple trials were conducted in this application. For each experimental data point, at least 50 repeated experiments were carried out and the average value was taken.

[0210] (2) Comparison of actual experimental performance

[0211] The comparison and analysis of the experiment in this application mainly focus on two key factors: data availability and efficiency.

[0212] Data availability: The root mean square error (RMSE) is used to evaluate the data availability of the scheme. The root mean square error is a metric that measures the magnitude of the difference between the evaluated shared result and the true result, and it can be expressed as:

[0213]

[0214] f(D) is the true query result, Ans i is the actual return result for the i-th time, and n is the number of queries. High data availability means that this method can provide more accurate data results for visitors, thus providing higher-quality services.

[0215] In Figure 3 , this application compared the RMSE under different Cs. Figure 3 (a) and Figure 3 (b) show that when processing the sum and count functions, the RMSE of the scheme in this application is significantly lower than that of [1]. This is because the scheme in this application can provide visitors with diverse privacy budgets, thereby reducing query errors and improving data utility. Therefore, the scheme in this application can provide higher data utility for most queries.

[0216] Figure 4 shows the comparison of the privacy budget with each data attribute. It shows that [2] allocates a fixed privacy budget for all data attributes D i while the privacy budget in the scheme of this application can be differentially allocated. In addition, for specific data attributes, the change of k will also lead to the change of the privacy budget. This is because k determines the weights of R and S, thus controlling the privacy policy. Therefore, this application not only provides differential privacy protection for various data attributes, but also can formulate personalized privacy strategies by adjusting k, thereby improving utility and preventing resource waste.

[0217] Efficiency: This application evaluated the scheme using the operation processing time of each step as the key performance indicator. The shorter the operation time, the higher the efficiency of the scheme and the stronger the real-time performance.

[0218] Figure 5Shows the computational overhead of the steps for constructing the access structure under different numbers of rows in the dataset. As the number of rows in the dataset increases, the time cost of the solution of this application grows linearly, but generally remains significantly lower than that of [2] and [3]. The reason is that [2] requires more time to construct the attribute matrix for the visitors, while [3] requires more time for data encryption. Therefore, the overall computational overhead of this application in the stage of constructing the index structure is better than that of [2] and [3].

[0219] Figure 6 Shows the comparison of the computational overhead of the quantization step with the number of visitors. Figure 6 (a) shows that as the number of visitors increases, the computational costs of the solution of this application and [2] both grow linearly, and the maximum difference between the two solutions does not exceed 20 ms. In addition, Figure 6 (b) shows that when the number of visitor attributes increases to 50, the computational overhead of this step remains almost unchanged. This is because the operations performed by both in this stage are similar, and the time complexity of each run only depends on the number of nodes in the access structure. Therefore, the computational overheads of the two solutions in this step are roughly the same.

[0220] Figure 7 Shows the comparison of the computational overhead when the generation matrix and the data sharing stage vary with the number of visitors. Figure 7 (a) The results show that the computational overheads of [2] and the solution of this application both increase linearly with the number of visitors, while the growth rate of the computational cost of the solution of this application is significantly lower. Figure 7 (b) shows that when the number of visitor attributes increases to 50, the solution of this application has more advantages. This is because the solution of this application only calculates the data attributes allowed to be accessed in the matrix, regardless of the number of visitor attributes. In contrast, the cost of [2] increases with the increase in the number of visitor attributes. Therefore, when dealing with a large number of visitor attributes, the solution of this application maintains better efficiency in this stage.

[0221] In summary, for the problems existing in data availability and sharing efficiency in privacy-preserving data sharing, the technical solution provided by the present invention, first of all, in order to ensure that the data is not accessed by attackers, this application constructs a quantifiable hierarchical access structure based on visitor attributes. In addition, the solution adaptively adjusts the privacy protection level according to the matching degree between the visitor and the access structure, improving data utility. Using matrix to calculate the privacy budget links the visitor attributes with the data they access, improving the pertinence of privacy protection. Then, the solution also designs a lightweight privacy budget calculation matrix to reduce the computational overhead and improve the sharing efficiency. Analysis proves that the solution proposed in this application can not only ensure privacy but also maintain reasonable efficiency. The main advantages of the present invention are as follows:

[0222] (1) Personalized Privacy Protection: This solution ensures personalized privacy protection for data access by constructing a hierarchical access structure based on the attributes of visitors. Each visitor is granted different data access permissions according to their attributes, and the privacy protection level is adaptively adjusted according to the matching degree between the visitor and the access structure. This flexible strategy guarantees that only authorized visitors can obtain the required data, reduces the risk of privacy leakage, and improves the security of data sharing.

[0223] (2) Improve Sharing Efficiency: The solution designs a lightweight privacy budget calculation matrix, which improves the efficiency of data sharing by reducing the computational overhead. By optimizing the dimensions of the matrix and removing inaccessible data attributes, the computational complexity is reduced. This not only alleviates the system burden but also significantly improves the computational speed of privacy budget allocation, ensuring efficient data processing and timely sharing responses.

[0224] (3) Balance Data Availability and Privacy: The solution allocates privacy budgets according to the queries of visitors and achieves a balance between privacy protection and data utility according to the adjustment of the decision coefficient. By adjusting the privacy standard through parameters, it ensures that while protecting privacy, accurate data query results are provided as much as possible. This flexible adjustment mechanism not only protects user privacy but also improves the availability of shared data, enabling the data to provide more valuable information for visitors while ensuring security.

[0225] It should be emphasized that the examples described in the present invention are illustrative rather than restrictive. Therefore, the present invention is not limited to the examples described in the specific embodiments. Any other embodiments obtained by those skilled in the art based on the technical solutions of the present invention, whether modified or replaced, as long as they do not depart from the spirit and scope of the present invention, also belong to the protection scope of the present invention.

[0226] [1] Dwork C, Naor M, Vadhan S. The Privacy of the Analyst and the Power of the State[C]. 2012 IEEE 53rd Annual Symposium on Foundations of Computer Science. New Brunswick, NJ, USA, 2012: 400 - 409.

[0227] [2]Zhang M, Zhou J, Zhang G, et al. APDP: Attribute-Based Personalized Differential Privacy Data Publishing Scheme for Social Networks[J]. IEEE Transactions on Network Science and Engineering, 2023, 10(2): 922-933.

[0228] [3]Han D, Pan N, Li K C. A Traceable and Revocable Ciphertext-Policy Attribute-based Encryption Scheme Based on Privacy Protection[J]. IEEE Transactions on Dependable and Secure Computing, 2022, 19(1): 316-327.

Claims

1. A personalized differential privacy data sharing method based on visitor attributes, characterized by: The following steps are involved: Step 1: Initialization operation, the cloud service provider CSP transmits the visitor attribute domain to the data owner DO; Step 2: Build an access structure, build a quantifiable hierarchical access control structure based on the visitor attribute domain and the data to be shared, and calculate the information entropy of each column of data attributes in the data to be shared, and transmit the access structure, the data to be shared and its information entropy set to the cloud service provider CSP for storage; Step 3: Initiate a query. The visitor DV generates a data query function and sends it to the cloud service provider CSP; Step 4: Quantification. After receiving the query, the cloud service provider CSP generates a series of matching indexes based on the matching degree between the attribute value of the visitor DV and the access structure; Step 5: Generate a privacy budget calculation matrix based on the matching index and the information entropy of the data attributes; Step 6: Data sharing. The privacy budget of the visitor DV is calculated based on the privacy budget calculation matrix, and noise corresponding to the privacy budget is added to the query results for data sharing.

2. The personalized differential privacy data sharing method based on visitor attributes according to claim 1 is characterized by: The process of building an access structure is as follows: First, the data owner DO selects a series of value domain subsets from the visitor attribute domain AD As a leaf node lnode x , indicating that the visitor with this attribute can obtain authorization for this node; Then, the data owner DO sets each column D in the data i Corresponding to the non-leaf node nnode of the access structure y And assign an access threshold number r to it y , as a visitor to obtain the minimum number of child nodes required for the column data; Afterwards, the data owner DO calculates the information entropy Ent(D i ), and normalize the result to obtain the data column information entropy set CE, which is used to construct the privacy budget calculation matrix later; Data attribute information entropy Ent(D i ) is calculated as follows: Where n is the number of sample classes in D, C j is the number of samples in the jth class, is the probability of the jth state occurring in the sample set; The calculation process of CE is as follows: Among them, CE i Refers to the information entropy of the i-th column data set; Finally, the data owner DO uses {lnode, nnode, r} as the access structure Γ, and transmits the access structure Γ, the data to be shared D, and the information entropy CE of the data attributes to the cloud service provider CSP.

3. According to claim 1, a personalized differential privacy data sharing method based on visitor attributes is characterized by: The visitor DV has been registered at the cloud service provider CSP, and the authenticity of the visitor DV attributes has been verified.

4. The personalized differential privacy data sharing method based on visitor attributes according to claim 1, characterized in that: The quantification process is as follows: First, the cloud service provider CSP uses the attribute A of the visitor DV to traverse all leaf nodes lnode of the visit structure Γ; if A is consistent with lnode x If there is an intersection, the leaf node is marked as TRUE, that is: Then, the cloud service provider CSP will check the access threshold r of all non-leaf nodes from bottom to top; if the number of its child nodes that are TRUE is greater than or equal to r x , then the record marks the non-leaf node as TRUE, that is: in, It is the child node of node x; Finally, a matching index MI is defined in each non-leaf node, which indicates the matching degree between the visitor attribute A and the access structure Γ: in, is the number of children of node x; for non-leaf nodes MI where t = FALSE unauth , easy to get MI unauth =0.

5. The personalized differential privacy data sharing method based on visitor attributes according to claim 1, characterized in that: The process of generating the matrix is ​​as follows: The cloud service provider CSP represents MI and information entropy CE as a matrix P: Considering the problem of inconsistent indices of the two columns of elements in the matrix, the cloud service provider CSP unifies the two columns of elements in the matrix into the maximum index: in is the transformation result, and then the transformation matrix is ​​obtained The cloud service provider CSP removes the data column that is not authorized to access, that is, x i1 = 0; let γ be the number after removing the unauthorized data column, which can be expressed as: BCM is the privacy budget calculation matrix.

6. The personalized differential privacy data sharing method based on visitor attributes according to claim 1, characterized in that: The process of data sharing is as follows: First, CSP solves the group utility value S: BCM + ={1,max(and 12 ,and 22 ,…,and γ2 )} BCM - ={0,min(and 12 ,and 22 ,…,and γ2 )} Where S i represents the group effect value of the i-th row of the matrix; w j Set by CSP, it is the weight of the jth column; Then, CSP solves the individual regret value R: Among them, R i represents the individual regret value of the i-th row; Afterwards, CSP solves the compromise solution value C and generates the privacy budget matrix: Where k is the decision coefficient, which is used to balance the influence of group effect value S and individual regret value R on C and is set by CSP; CSP according to ε i =exp(-C i ), generate the privacy budget matrix BM; Finally, CSP processes the query of DV, outputs the query result, and runs the differential privacy algorithm to generate the result Ans: Where S(f) is the sensitivity of the query function f. CSP uses the above formula to add noise to the query result through the Laplace mechanism and shares the output result Ans to DV.

7. A system based on the personalized differential privacy data sharing method based on visitor attributes according to any one of claims 1 to 6, characterized in that: include: Data owner DO, accessor DV, cloud service provider CSP; Among them, the data owner DO and the visitor DV are both connected to the cloud service provider CSP; The cloud service provider CSP transmits a visitor attribute domain to the data owner DO, wherein the attribute domain includes a visitor attribute category and an optional value; The data owner DO builds a quantifiable hierarchical access control structure based on the visitor attribute domain and the data to be shared, and calculates the information entropy of each column of data attributes in the data to be shared, and transmits the access structure, the data to be shared and its information entropy set to the cloud service provider CSP for storage; The visitor DV generates a data query function and sends it to the cloud service provider CSP; After receiving the query, the cloud service provider CSP generates a series of matching indexes based on the matching degree between the attribute values ​​of the visitor DV and the access structure; The cloud service provider CSP generates a privacy budget calculation matrix based on the matching index and the information entropy of the data attributes; The cloud service provider CSP calculates the privacy budget of the visitor DV based on the privacy budget calculation matrix, and adds noise corresponding to the privacy budget to the query results for data sharing.

8. A system based on the personalized differential privacy data sharing method based on visitor attributes according to any one of claims 1 to 6, characterized in that: include: Initialization operation module: used to enable the cloud service provider CSP to transmit the visitor attribute domain to the data owner DO; Access structure building module: used to build a quantifiable hierarchical access control structure and calculate the information entropy of each column of data attributes in the data to be shared, and transmit the access structure, the data to be shared and its information entropy set to the cloud service provider CSP for storage; Query initiation module: used to send the data query function generated by the visitor DV to the cloud service provider CSP; Matching index quantification module: generates a series of matching indexes based on the matching degree between the attribute value of the visitor DV and the access structure; Privacy budget matrix generation module: used to generate the privacy budget calculation matrix based on the matching index and the information entropy of data attributes; Data sharing module: used to calculate the privacy budget of the visitor DV based on the privacy budget calculation matrix, and add noise corresponding to the privacy budget to the query results for data sharing.

9. An electronic terminal, characterized in that: At least: one or more processors; a memory storing one or more computer programs; The processor calls a computer program to execute: The steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: A computer program is stored, which is called by a processor to execute: The steps of the method according to any one of claims 1 to 6.

Images