Privacy information protection method and device based on homomorphic encryption and unsupervised feature selection

Abstract

The application discloses a privacy information protection method and device based on homomorphic encryption and unsupervised feature selection. The method comprises the following steps: when a user outsources privacy information, the privacy information is encrypted by using a preset homomorphic encryption algorithm to perform a feature selection operation, and then the privacy information is uploaded to a server to perform a preset data processing operation; and / or when the user performs feature selection processing on the privacy information, the privacy information is protected based on the preset homomorphic encryption algorithm and an unsupervised feature selection model during the feature selection processing. The application solves the technical problem that a feature set used for modeling has a high dimension and an information leakage is easily caused in a processing process. The application solves the feature selection problem under a label scarcity condition and realizes privacy protection of sensitive feature information.

Description

Technical Field

[0001] This application relates to the field of privacy information protection, and more specifically, to a privacy information protection method and apparatus based on homomorphic encryption and unsupervised feature selection. Background Technology

[0002] In network environments, it is often necessary to build intrusion detection models based on real-time collected traffic data.

[0003] The inventors discovered that outsourcing or feature selection of traffic data can easily lead to the leakage of sensitive information. Furthermore, the high dimensionality of traffic data datasets can also affect the computational efficiency and time of classifiers.

[0004] There is currently no effective solution to the problem that the feature sets used for modeling in related technologies have high dimensionality and the processing can easily lead to information leakage. Summary of the Invention

[0005] The main objective of this application is to provide a privacy information protection method and apparatus based on homomorphic encryption and unsupervised feature selection, in order to solve the problem that the feature set used for modeling has high dimensionality and the processing is prone to information leakage.

[0006] To achieve the above objectives, according to one aspect of this application, a privacy information protection method based on homomorphic encryption and unsupervised feature selection is provided.

[0007] The privacy information protection method based on homomorphic encryption and unsupervised feature selection according to this application includes: when a user outsources the processing of privacy information, the privacy information is encrypted using a preset homomorphic encryption algorithm for feature selection operations before being uploaded to a server for preset data processing operations; and / or, when the user performs feature selection processing on the privacy information, the privacy information is protected while performing the feature selection processing based on the preset homomorphic encryption algorithm and using an unsupervised feature selection model.

[0008] Furthermore, the unsupervised feature selection model also includes: self-representational selection of target feature data based on homomorphic encryption features.

[0009] Furthermore, when the user performs feature selection processing on the privacy information, the privacy information is protected simultaneously based on the preset homomorphic encryption algorithm and the unsupervised feature selection model, including:

[0010] Based on the unsupervised feature selection model, a feature matrix X is obtained; using the feature matrix X and projection matrix A obtained based on the preset homomorphic encryption algorithm, the importance of different features is determined, wherein ai Let a represent the i-th row of matrix A. i =(a i1 ,a i2 ,…,a iM ), i∈[1,D].

[0011] Further, the projection matrix A includes:

[0012] Based on the column vectors of matrix A reflecting the importance of different features, when only k columns of matrix A are non-zero, then k features in the corresponding feature matrix X are selected, and the rest are not selected. The reconstruction loss term is then... If a i If the vector is 0, then the corresponding i-th dimension feature x i Its contribution is 0.

[0013] Furthermore, after the feature selection process, the problem further includes: simultaneously satisfying the condition that the error between the reconstruction matrix AX and the matrix X is minimized, and that only k columns of the projection matrix A are non-zero, the constrained optimization problem of the reconstruction loss term can be expressed in the following form:

[0014]

[0015] Introducing a square matrix r of dimension m, where the elements on the diagonal are either 0 or 1, the number of 1s on the diagonal is k, and the row / column indices and all other elements in the non-zero columns of matrix A are 0, while introducing a regularization term to balance the algorithm's complexity and the fit of parameter optimization, the above constrained optimization problem is transformed into the following form:

[0016]

[0017] When the i-th row element of X is selected, r ii =1, otherwise r ii =0,λ A Used to balance algorithm complexity with the fit of parameter optimization.

[0018] Furthermore, when a user outsources the processing of privacy information, the privacy information is encrypted using a preset homomorphic encryption algorithm for feature selection before being uploaded to the server for preset data processing. This also includes: mapping the actual data of the privacy information to a plaintext space M using a preset preprocessing algorithm; converting floating-point data into binary data within the plaintext space, and letting D′ be the matrix after transforming the data feature matrix D, where all elements of the transformed feature matrix D′ belong to the plaintext space M. Then, each element of matrix D′ can be represented as...

[0019] d j =(d j1 ,dj2 ,L,d jN ),d ij ∈{0,1}.

[0020] Furthermore, the user outsources the processing of privacy information. After encrypting the feature selection operation using a preset homomorphic encryption algorithm, the privacy information is uploaded to the server for preset data processing operations. This also includes: when encrypting the feature selection operation using the preset homomorphic encryption algorithm, the same plaintext can be encrypted multiple times to obtain multiple different ciphertext results.

[0021] To achieve the above objectives, according to another aspect of this application, a privacy information protection device based on homomorphic encryption and unsupervised feature selection is provided.

[0022] The privacy information protection device based on homomorphic encryption and unsupervised feature selection according to this application includes:

[0023] The homomorphic encryption module is used to encrypt the feature selection operation of the privacy information using a preset homomorphic encryption algorithm when the user outsources the processing of privacy information, and then upload it to the server for preset data processing operations.

[0024] An unsupervised feature selection module is used to protect the privacy information when the user performs feature selection processing based on the preset homomorphic encryption algorithm and an unsupervised feature selection model.

[0025] To achieve the above objectives, according to another aspect of this application, a computer-readable storage medium is provided, wherein a computer program is stored therein, and the computer program is configured to execute the method at runtime.

[0026] To achieve the above objectives, according to another aspect of this application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method.

[0027] In this application embodiment, a privacy information protection method and apparatus based on homomorphic encryption and unsupervised feature selection are provided. When a user outsources the processing of their privacy information, the feature selection operation is encrypted using a preset homomorphic encryption algorithm before being uploaded to a server for preset data processing. Alternatively, when the user performs feature selection processing on their privacy information, the privacy information is protected simultaneously based on the preset homomorphic encryption algorithm and an unsupervised feature selection model. By using the preset homomorphic encryption algorithm and the unsupervised feature selection model, the aim of homomorphic encryption of sensitive information and unsupervised feature selection of feature data is achieved, thereby realizing the technical effect of protecting user privacy data and solving the technical problem of high-dimensional feature sets used for modeling and the ease with which information leakage can occur during processing. Attached Figure Description

[0028] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings:

[0029] Figure 1 This is a schematic diagram of the hardware structure of a privacy information protection method based on homomorphic encryption and unsupervised feature selection according to an embodiment of this application;

[0030] Figure 2 This is a flowchart illustrating a privacy information protection method based on homomorphic encryption and unsupervised feature selection according to an embodiment of this application.

[0031] Figure 3 This is a schematic diagram of a privacy information protection device based on homomorphic encryption and unsupervised feature selection according to an embodiment of this application. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0033] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0034] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0035] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0036] Furthermore, the terms "installation," "setup," "equipped with," "connection," "linking," and "socketing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral structure; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0038] like Figure 1As shown, the hardware structure of the privacy information protection method based on homomorphic encryption and unsupervised feature selection in this embodiment includes: user 100, local computer 200, and cloud server 300. Traffic data is generated when user 100 accesses cloud server 300 through local computer 200. The rapid development of cloud services provides users with more diversified data processing methods. Due to the large variety of users, the datasets they provide will cover various types, such as email information, background user database information, personal health status, and other information involving personal privacy or corporate trade secrets. When users outsource or perform feature selection processing on the above information (which will be synchronized with cloud server 300), there is a risk of leakage of sensitive information contained therein.

[0039] like Figure 2 As shown, the method includes the following steps S201 to S202:

[0040] Step S201: When a user outsources the processing of privacy information, the privacy information is encrypted using a preset homomorphic encryption algorithm to encrypt the feature selection operation before being uploaded to the server for preset data processing operations.

[0041] In step S202, and / or, when the user performs feature selection processing on the privacy information, the privacy information is protected while performing the feature selection processing based on the preset homomorphic encryption algorithm and using an unsupervised feature selection model.

[0042] The entire feature selection process described above is carried out under the condition of encrypted feature matrix elements. Therefore, the cloud server will not obtain any feature information when implementing unsupervised feature selection. Although it can know which dimensions of features are selected as key content through the non-zero elements of k, it will never obtain the specific content of the features, thus achieving privacy protection in unsupervised feature selection.

[0043] As can be seen from the above description, this application achieves the following technical effects:

[0044] When users outsource the processing of their privacy information, the privacy information is encrypted using a preset homomorphic encryption algorithm before being uploaded to the server for preset data processing. Alternatively, when users perform feature selection processing on their privacy information, the privacy information is protected simultaneously based on the preset homomorphic encryption algorithm and an unsupervised feature selection model. By using the preset homomorphic encryption algorithm and the unsupervised feature selection model, the goals of homomorphic encryption of sensitive information and unsupervised feature selection of feature data are achieved, thereby realizing the technical effect of protecting user privacy data and solving the technical problem of high-dimensional feature sets used for modeling and the ease with which information leakage can occur during processing.

[0045] In step S201 above, when a user needs to outsource the processing of privacy information, the feature selection operation is encrypted using the preset homomorphic encryption algorithm before being uploaded to the server for preset data processing. In other words, the data is encrypted using the preset homomorphic encryption algorithm before being uploaded to the server.

[0046] As an optional implementation, homomorphic encryption algorithms include a key generation process, an encryption process, and a decryption process.

[0047] As an alternative implementation, the noise introduced by the homomorphic encryption algorithm needs to be removed.

[0048] In another scenario of step S202 above, namely when the user performs feature selection processing on privacy information, the privacy information is protected while performing the feature selection processing based on the preset homomorphic encryption algorithm and the unsupervised feature selection model.

[0049] As an alternative implementation, considering that the data collected in the actual network environment lacks sufficient category labels, the importance of features cannot be determined by the correlation between features and labels, thus making accurate feature selection impossible. Simultaneously, considering the privacy of feature information, the unsupervised feature selection method based on homomorphic encryption in this application embodiment is adopted, selecting important features based on the self-representation of encrypted features.

[0050] As a preferred embodiment, the unsupervised feature selection model further includes: self-representational selection of target feature data based on homomorphic encryption features. That is, the target feature data selected by the unsupervised feature selection model is selected based on the importance of the encryption feature itself.

[0051] In a preferred embodiment, when the user performs feature selection processing on privacy information, the privacy information is protected simultaneously based on the preset homomorphic encryption algorithm and an unsupervised feature selection model. This includes: obtaining a feature matrix X based on the unsupervised feature selection model; and determining the importance of different features using the feature matrix X obtained based on the preset homomorphic encryption algorithm and the projection matrix A, wherein the importance of a... i Let a represent the i-th row of matrix A. i =(a i1 ,a i2 ,…,a iM ), i∈[1,D].

[0052] In practice, the column vectors of matrix A can reflect the importance of different features. When only k columns of matrix A are not zero, only k features of the corresponding feature matrix X are selected, and the rest are not selected.

[0053] In a preferred embodiment, the projection matrix A includes: based on the column vectors of matrix A reflecting the importance of different features, when only k columns of matrix A are not zero, then k features of the corresponding feature matrix X are selected, and the rest are not selected. The reconstruction loss term is then... If a i If the vector is 0, then the corresponding i-th dimension feature x i Its contribution is 0.

[0054] In a preferred embodiment, after the feature selection process, the method further includes:

[0055] If the error between the reconstruction matrix AX and the matrix X is minimized, and only k columns of the projection matrix A are non-zero, then the constrained optimization problem of the reconstruction loss term can be expressed in the following form:

[0056]

[0057] Introducing a square matrix r of dimension m, where the elements on the diagonal are either 0 or 1, the number of 1s on the diagonal is k, and the row / column indices and all other elements in the non-zero columns of matrix A are 0, while introducing a regularization term to balance the algorithm's complexity and the fit of parameter optimization, the above constrained optimization problem is transformed into the following form:

[0058]

[0059] When the i-th row element of X is selected, r ii =1, otherwise r ii =0,λ A Used to balance algorithm complexity with the fit of parameter optimization.

[0060] In practice, when the i-th row element of X is selected, r ii =1, otherwise r ii =0. λ A To balance algorithm complexity with the fit of parameter optimization, avoiding overfitting and excessive algorithm complexity, the optimization problem is transformed into a Lagrangian function using the alternating direction multiplier method. An iterative optimization method is then used to solve for the variable, yielding the value of k, matrix r, and matrix A. Based on the k non-zero columns of matrix A, elements in matrix X are selected, choosing the k rows corresponding to the k non-zero columns of A. This feature selection process is independent of the column labels of the features, achieving unsupervised feature selection.

[0061] Preferably, to verify the reliability of the results, the user performs unsupervised feature selection locally using a feature matrix that has not undergone homomorphic encryption. Then, the user calculates the k' non-zero columns of the resulting projection matrix A' and compares whether the k' non-zero columns are equal to k, and whether the positions of the k' non-zero elements in A' are completely consistent with the positions of the k non-zero elements in A. If they are consistent, the selected k-dimensional features are further input into multiple detection models to verify the effectiveness of the feature selection.

[0062] As a preferred embodiment, when a user outsources the processing of privacy information, the privacy information is encrypted using a preset homomorphic encryption algorithm for the feature selection operation before being uploaded to the server for preset data processing operations. This also includes:

[0063] The actual data of privacy information is mapped to plaintext space M through a preset preprocessing algorithm;

[0064] We need to convert floating-point data into binary data within plaintext space. Let D′ be the matrix after transforming the data feature matrix D, meaning that all elements of the transformed feature matrix D′ belong to the plaintext space M. Then, each element of matrix D′ can be represented as...

[0065] d j =(d j1 ,d j2 ,L,d jN ),d ij ∈{0,1}.

[0066] In practical implementation, let D be the data feature matrix, d j Let d be the j-th row of matrix D, where each row represents a sample, and the row number represents the number of samples. j =(d j1 ,d j2 ,L,d jN ), j∈[1,L], each column represents a 1-dimensional feature, the column number i∈[1,N] represents the dimension of the feature, and each dimensional feature is represented by d i To represent this, in most real-world scenarios, data cannot be directly used as plaintext for encryption algorithms. A preprocessing algorithm is needed to map the actual data to the plaintext space M. Currently, common feature transformation formats include floating-point data and binary data. Therefore, it is first necessary to convert the floating-point data into binary data within the plaintext space. Let D′ be the matrix after transforming the data feature matrix D, meaning that all elements of the transformed feature matrix D′ belong to the plaintext space M. Then, each element of matrix D′ can be represented in the following form d j =(d j1 ,d j2 ,L,d jN ),dij ∈{0,1}.

[0067] As a preferred embodiment, the user outsources the processing of privacy information. After encrypting the feature selection operation using a preset homomorphic encryption algorithm, the privacy information is uploaded to the server for preset data processing operations. This also includes: when encrypting the feature selection operation using the preset homomorphic encryption algorithm, the same plaintext can be encrypted multiple times to obtain multiple different ciphertext results.

[0068] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0069] According to embodiments of this application, a privacy information protection device based on homomorphic encryption and unsupervised feature selection for implementing the above-described method is also provided, such as... Figure 3 As shown, the device includes:

[0070] The homomorphic encryption module 301 is used to encrypt the feature selection operation of the privacy information using a preset homomorphic encryption algorithm when the user outsources the processing of privacy information, and then upload it to the server for preset data processing operations.

[0071] The unsupervised feature selection module 302 is used to protect the privacy information while performing feature selection processing based on the preset homomorphic encryption algorithm and the unsupervised feature selection model when the user performs feature selection processing on the privacy information.

[0072] In the homomorphic encryption module 301 of this application embodiment, when a user needs to outsource the processing of privacy information, the feature selection operation is encrypted using the preset homomorphic encryption algorithm before being uploaded to the server for preset data processing operations. In other words, the data is encrypted using the preset homomorphic encryption algorithm before being uploaded to the server.

[0073] As an optional implementation, homomorphic encryption algorithms include a key generation process, an encryption process, and a decryption process.

[0074] As an alternative implementation, the noise introduced by the homomorphic encryption algorithm needs to be removed.

[0075] In another scenario, namely when the user performs feature selection processing on privacy information, the unsupervised feature selection module 302 of this application embodiment protects the privacy information while performing the feature selection processing based on the preset homomorphic encryption algorithm and the unsupervised feature selection model.

[0076] As an alternative implementation, considering that the data collected in the actual network environment lacks sufficient category labels, the importance of features cannot be determined by the correlation between features and labels, thus making accurate feature selection impossible. Simultaneously, considering the privacy of feature information, the unsupervised feature selection method based on homomorphic encryption in this application embodiment is adopted, selecting important features based on the self-representation of encrypted features.

[0077] For homomorphic encryption module 301, the following process is executed:

[0078] Key generation: Let λ be the security parameter of the encryption system. Input λ into the key generation system, and output three parameters ρ, η, and γ, where... Let p be an odd integer of η bits, then p←(2Z+1)I(2 η-1 ,2 η Let p be the private key sk, and calculate t0 = q0·p as the public key, where q0 ← (2Z+1)I(1,2). γ p).

[0079] Encryption: Input plaintext m∈{0,1} and public key t0, respectively from [1,2... γ p) and (-2 ρ ,2 ρ Select random numbers q′ and r, then calculate and output the ciphertext according to the following formula.

[0080] C=E(m,pk)=(m+2·r+q′·p)modt0

[0081] Decryption: Input ciphertext C and private key sk, then output plaintext m according to the following formula.

[0082] m = D(C, sk) = (C mod p) mod 2

[0083] Because homomorphic encryption introduces random noise, encrypting the same plaintext multiple times can yield multiple different ciphertext results.

[0084] The unsupervised feature selection module 302 is used to execute the following process:

[0085] Let D be the data feature matrix, d j Let d be the j-th row of matrix D, where each row represents a sample, and the row number represents the number of samples. j =(d j1 ,d j2 ,L,d jN ), j∈[1,L], each column represents a 1-dimensional feature, the column number i∈[1,N] represents the dimension of the feature, and each dimensional feature is represented by d iTo represent this, in most real-world scenarios, data cannot be directly used as plaintext for encryption algorithms. A preprocessing algorithm is needed to map the actual data to the plaintext space M. Currently, common feature transformation formats include floating-point data and binary data. Therefore, it is first necessary to convert the floating-point data into binary data within the plaintext space. Let D′ be the matrix after transforming the data feature matrix D, meaning that all elements of the transformed feature matrix D′ belong to the plaintext space M. Then, each element of matrix D′ can be represented in the following form d j =(d j1 ,d j2 ,L,d jN ),d ij ∈{0,1}.

[0086] Let X be the feature matrix after homomorphic encryption of D′ (the elements in D′ are encrypted using the homomorphic encryption algorithm mentioned above, i.e., each element in X is the homomorphic ciphertext form of the corresponding element in the feature matrix), A be the projection matrix, and a i Let a represent the i-th row of matrix A. i =(a i1 ,a i2 ,…,a iM ), i∈[1,D]. The column vectors of matrix A can reflect the importance of different features. When only k columns of matrix A are not zero, only k features of the corresponding feature matrix X are selected, and the rest are not selected. Then the reconstruction loss term can be represented as If a i If the vector is 0, then the corresponding i-th dimension feature x i The contribution is 0. To simultaneously minimize the error between the reconstruction matrix AX and the feature matrix X, and to ensure that only k columns of the projection matrix A are non-zero, the constrained optimization problem of the reconstruction loss term can be expressed as follows:

[0087]

[0088] Consider the above ||A|| 2，0 Since the global optimum cannot be directly found for the constrained problem, we introduce a square matrix r of dimension m, where the elements on the diagonal of r are either 0 or 1, the number of 1s on the diagonal is k, and the row / column indices and all other elements in the non-zero columns of matrix A are 0. We also introduce a regularization term to balance the algorithm's complexity with the fit of the parameter optimization. The above optimization problem can then be transformed into the following form:

[0089]

[0090] When the i-th row element of X is selected, r ii =1, otherwise r ii =0. λ ATo balance algorithm complexity with the fit of parameter optimization, avoiding overfitting and excessive algorithm complexity, the optimization problem is transformed into a Lagrangian function using the alternating direction multiplier method. An iterative optimization method is then used to solve for the variable, yielding the value of k, matrix r, and matrix A. Based on the k non-zero columns of matrix A, elements in matrix X are selected, choosing the k rows corresponding to the k non-zero columns of A. This feature selection process is independent of the column labels of the features, achieving unsupervised feature selection.

[0091] To verify the reliability of the results, the user performs unsupervised feature selection locally using a feature matrix that has not undergone homomorphic encryption. Then, the user calculates the k' non-zero columns of the resulting projection matrix A' and compares them to see if the k' non-zero columns are equal to k, and whether the positions of the k' non-zero elements in A' are exactly the same as the positions of the k non-zero elements in A. If they are consistent, the selected k-dimensional features are further input into multiple detection models to verify the effectiveness of the feature selection.

[0092] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0093] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A privacy information protection method based on homomorphic encryption and unsupervised feature selection, characterized in that, include: When a user outsources the processing of privacy information, the feature selection operation is encrypted using a preset homomorphic encryption algorithm before being uploaded to the server for preset data processing operations. When the user performs feature selection processing on their privacy information, the privacy information is protected while performing the feature selection processing based on the preset homomorphic encryption algorithm and an unsupervised feature selection model. When the user performs feature selection processing on privacy information, the privacy information is protected simultaneously based on the preset homomorphic encryption algorithm and an unsupervised feature selection model, including: Based on the unsupervised feature selection model, the feature matrix is ​​obtained. ; The feature matrix obtained based on the preset homomorphic encryption algorithm and projection matrix To determine the importance of different features, wherein the Representation matrix The OK, , , The projection matrix ,include: Based on the matrix The column vectors reflect the importance of different features, when the matrix only When a column is not zero, then the corresponding characteristic matrix have If one feature is selected and the rest are not selected, then the reconstruction loss term is: ;if If the vector is 0, then the corresponding first vector is... 3D features Its contribution is 0; After the feature selection process, the process further includes: Simultaneously satisfy the reconstruction matrix With the matrix The error between them is minimized, and the projection matrix only If the column is not zero, the constrained optimization problem of reconstructing the loss term can be expressed in the following form: Introducing dimensions as square array ,in, If the elements on the diagonal are either 0 or 1, then the number of 1s on the diagonal is . One, row index / column index and matrix If all other elements in the non-zero column are 0, and a regularization term is introduced to balance the algorithm's complexity with the fit of parameter optimization, then the above constrained optimization problem is transformed into the following form: when The When a row element is selected, ,otherwise , Used to balance algorithm complexity with the fit of parameter optimization.

2. The method according to claim 1, characterized in that, The unsupervised feature selection model further includes: self-representational selection of target feature data based on homomorphic encryption features.

3. The method according to claim 1, characterized in that, When a user outsources the processing of privacy information, the privacy information is encrypted using a preset homomorphic encryption algorithm after the feature selection operation is performed, and then uploaded to the server for preset data processing operations. This also includes: The actual data of privacy information is mapped to plaintext space through a preset preprocessing algorithm. ; It is necessary to convert floating-point data into binary data within plaintext space, so that... For the data feature matrix The transformed matrix, i.e., the characteristic matrix after transformation. All elements belong to plaintext space. Then the matrix Each element is represented as 。 4. The method according to claim 1, characterized in that, The process of outsourcing the processing of user privacy information, encrypting the feature selection operation using a preset homomorphic encryption algorithm, and then uploading the information to the server for preset data processing operations, further includes: When using a preset homomorphic encryption algorithm to encrypt the feature selection operation, the same plaintext is encrypted multiple times to obtain multiple different ciphertext results.

5. A privacy information protection device based on homomorphic encryption and unsupervised feature selection, characterized in that, include: The homomorphic encryption module is used to encrypt the feature selection operation of the privacy information using a preset homomorphic encryption algorithm when the user outsources the processing of privacy information, and then upload it to the server for preset data processing operations. An unsupervised feature selection module is used to protect the privacy information while performing feature selection processing based on the preset homomorphic encryption algorithm and an unsupervised feature selection model when the user performs feature selection processing on the privacy information. This includes: Based on the unsupervised feature selection model, the feature matrix is ​​obtained. ; The feature matrix obtained based on the preset homomorphic encryption algorithm and projection matrix To determine the importance of different features, wherein the Representation matrix The OK, , , The projection matrix ,include: Based on the matrix The column vectors reflect the importance of different features, when the matrix only When a column is not zero, then the corresponding characteristic matrix have If one feature is selected and the rest are not selected, then the reconstruction loss term is: ;if If the vector is 0, then the corresponding first vector is... 3D features The contribution is 0. After the feature selection process, the following is also included: Simultaneously satisfy the reconstruction matrix With the matrix The error between them is minimized, and the projection matrix only If the column is not zero, the constrained optimization problem of reconstructing the loss term can be expressed in the following form: Introducing dimensions as square array ,in, If the elements on the diagonal are either 0 or 1, then the number of 1s on the diagonal is . One, row index / column index and matrix If all other elements in the non-zero column are 0, and a regularization term is introduced to balance the algorithm's complexity with the fit of parameter optimization, then the above constrained optimization problem is transformed into the following form: when The When a row element is selected, ,otherwise , Used to balance algorithm complexity with the fit of parameter optimization.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to perform the method described in any one of claims 1 to 5 when executed.

7. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method as described in any one of claims 1 to 5.

Images