An active defense method and system based on differential privacy protection

Abstract

This invention discloses a proactive defense method and system based on differential privacy protection. The method includes: obtaining a sensitive data matrix; the sensitive data matrix includes sample vectors, which are vectorized representations of sensitive data within a business system; projecting the sensitive data matrix onto an eigenvector matrix to obtain a transformation matrix; the eigenvector matrix includes eigenvectors of the sensitive data matrix; adding noise that satisfies differential privacy to the transformation matrix to obtain a noise matrix; projecting the noise matrix onto the eigenvector matrix to obtain a hidden matrix of the sensitive data matrix; and publishing the hidden matrix for access by visitors. This application adds effective noise to sensitive data within the protected business system while ensuring data availability, preventing attackers from successfully accessing sensitive data in the business system. This constructs a self-controllable proactive defense system, ensuring data security.

Description

Technical Field

[0001] The present invention relates to the field of network security protection, and in particular to an active defense method and system based on differential privacy protection. Background Art

[0002] We are currently entering the intelligent age from the information age. The scale of data is rapidly expanding, the value of data applications is being recognized, and data flows have brought huge economic benefits and wealth. While data elements are being marketized and capitalized, data security and privacy security issues are becoming more prominent. With the widespread application of big data, cloud computing, and mobile Internet, governments and enterprises are paying more and more attention to data, and huge benefits have been generated through data mining and data analysis. However, in the process of data mining, analysis, and use, a large amount of privacy information carried by the data is leaked, and privacy protection faces new challenges. Therefore, in order to meet the security needs of sensitive information and data in different systems, it is necessary to design a method to scientifically and actively protect sensitive information and data.

[0003] In view of this situation, the existing technology generally implements the protection of sensitive information and data through the following methods:

[0004] Method 1: K-anonymity model, a data anonymization method. By obfuscating the attributes of the data, given an original data set, an anonymized data set is generated to ensure that the individual identities in the data set cannot be recovered under the experimental availability of the data. That is, any quasi-identifier information in the public data, the same combination appears at least K times.

[0005] Method 2: Diversity: There must be at least one well-replaceable sensitive attribute value in an equivalence class. For a data table, each equivalence class must satisfy L-diversity. That is, if an attacker knows the details of an individual's equivalence class, and there are more than L sensitive attribute values ​​in the equivalence class, the attacker cannot learn the individual's more confidential information, and the individual's privacy is protected.

[0006] The first method mentioned above has the following disadvantages:

[0007] When the order of the public data records is the same as the original records, the attacker can guess the anonymized records.

[0008] The above second method has the following disadvantages:

[0009] In a real data set, attribute values ​​are likely to be skewed or semantically similar. L-diversity ensures diversity and cannot recognize semantic similarity. Attackers can obtain relevant information that is useful to them through semantics with a high degree of similarity.

[0010] In summary, there are contradictions in existing data protection: if the scale of noise added is too small to ensure data availability, there is a risk of leaking user privacy; on the contrary, adding excessive noise will cause serious distortion of anonymous data and lose the original data value. Summary of the invention

[0011] The purpose of the present invention is to provide an active defense method and system based on differential privacy protection, which can achieve stronger data availability under the same protection.

[0012] To achieve the above object, the present invention provides the following solutions:

[0013] An active defense method based on differential privacy protection, including:

[0014] Obtain a sensitive data matrix; the sensitive data matrix includes a sample vector, and the sample vector includes a vectorized representation of sensitive data in the business system; the sensitive data is data that needs to be protected;

[0015] According to the eigenvector matrix, the sensitive data matrix is ​​projected to obtain a conversion matrix; the eigenvector matrix includes the eigenvectors of the sensitive data matrix;

[0016] Adding noise satisfying differential privacy to the transformation matrix to obtain a noise matrix;

[0017] Projecting the noise matrix according to the eigenvector matrix to obtain a hidden matrix of the sensitive data matrix;

[0018] The hidden matrix is ​​published for access by visitors.

[0019] Optionally, before projecting the sensitive data matrix according to the eigenvector matrix, the method further includes:

[0020] Obtaining the eigenvalues ​​and eigenvectors of the sensitive data matrix;

[0021] Select target eigenvectors to form the eigenvector matrix; the target eigenvectors include eigenvectors corresponding to eigenvalues ​​with the largest values.

[0022] Optionally, before adding noise satisfying differential privacy to the transformation matrix to obtain the noise matrix, the method further includes:

[0023] The noise p satisfying differential privacy is determined according to the number of rows m of the eigenvector matrix, the number of columns k of the eigenvector matrix, and the differential privacy budget function ε.

[0024] Optionally, adding noise satisfying differential privacy to the transformation matrix to obtain a noise matrix specifically includes:

[0025] According to the projected transformation matrix F n×k And the noise p that satisfies differential privacy, determine the noise matrix F noise , F noise =F n×k +p.

[0026] Optionally, projecting the sensitive data matrix according to the eigenvector matrix specifically includes:

[0027] The sensitive data matrix is ​​projected using a first projection formula, where the first projection formula is:

[0028] F n×k =X×V k

[0029] Among them, F n×k represents the conversion matrix, X represents the sensitive data matrix, V k Denotes the eigenvector matrix.

[0030] Optionally, projecting the noise matrix according to the eigenvector matrix specifically includes: projecting the noise matrix using a second projection formula, where the second projection formula is:

[0031] F X‘ =F noise ×V k

[0032] Among them, F X‘ represents the hidden matrix of the sensitive data matrix, V k represents the eigenvector matrix, F noise Denote the noise matrix

[0033] After the above steps, the sensitive data in the protected business system has less noise added and higher availability, thus achieving the goal of preventing attackers from obtaining sensitive data information.

[0034] The present invention also provides an active defense system based on differential privacy protection, comprising:

[0035] A sensitive data matrix acquisition module is used to acquire a sensitive data matrix; the sensitive data matrix includes a sample vector, and the sample vector includes a vectorized representation of sensitive data in the business system; the sensitive data is data that needs to be protected;

[0036] A projection module, used to project the sensitive data matrix according to a characteristic vector matrix to obtain a conversion matrix; the characteristic vector matrix includes the characteristic vectors of the sensitive data matrix;

[0037] A noise matrix determination module, used for adding noise satisfying differential privacy to the transformation matrix to obtain a noise matrix;

[0038] The projection module is further used to project the noise matrix according to the eigenvector matrix to obtain a hidden matrix of the sensitive data matrix;

[0039] The publishing module is used to publish the hidden matrix for visitors to access.

[0040] The projection sample matrix after the above modules uses less noise, ensuring the availability of data for normal visitors while preventing attackers from obtaining sensitive data.

[0041] Optionally, also include:

[0042] A feature obtaining module, used to obtain the eigenvalues ​​and eigenvectors of the sensitive data matrix;

[0043] The eigenvector matrix determination module is used to select target eigenvectors to form the eigenvector matrix; the target eigenvectors include the eigenvectors corresponding to the eigenvalues ​​with the largest numerical values.

[0044] Optionally, also include:

[0045] A projection function determination module is used to determine the noise p that satisfies differential privacy according to the number of rows m of the eigenvector matrix, the number of columns k of the eigenvector matrix, and the differential privacy budget function ε.

[0046] Optionally, the noise matrix determination module specifically includes:

[0047] A noise matrix determination unit is used to determine the noise matrix according to the conversion matrix F n×k And the noise p that satisfies differential privacy, determine the noise matrix F noise , F noise =F n×k +p.

[0048] According to the specific embodiment provided by the present invention, the following technical effects are disclosed: the present application first projects the sensitive data matrix according to the eigenvector of the sensitive data matrix to obtain a transformation matrix, wherein the sensitive data matrix is ​​a vectorized representation of the sensitive data in the protected business system. Then, noise that satisfies differential privacy is added to the transformation matrix to obtain a noise matrix. According to the above-mentioned eigenvector matrix, the noise matrix is ​​projected to obtain a hidden matrix of the sensitive data matrix. Finally, the hidden matrix is ​​published for access by visitors. That is to say, when the current attacker accesses the data, the present application uses this method to add effective noise to the sensitive data in the protected business system, while ensuring data availability to ensure that the attacker cannot successfully access the sensitive data of the business system, thereby building an autonomous and controllable active defense system to ensure data security. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0050] Figure 1 Schematic diagram of the process of an active defense method based on differential privacy protection in an embodiment of the present invention;

[0051] Figure 2 4 is a framework diagram of an active defense system in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, words such as "first" and "second" are used to distinguish the same or similar items with basically the same functions and effects. For example, the first wireless communication module and the second wireless communication module are only used to distinguish different wireless communication modules, and their order is not limited. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order.

[0053] It should be noted that, in this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The purpose of the present invention is to provide an active defense method and system based on differential privacy protection to ensure data security.

[0056] See also Figure 1 , the present application provides an active defense method based on differential privacy protection, the method comprising the following steps:

[0057] Step 101: Obtain a sensitive data matrix; the sensitive data matrix includes sample vectors, and the sample vectors include vectorized representations of sensitive data in a business system; the sensitive data is data that needs to be protected.

[0058] Step 102: Project the sensitive data matrix according to the eigenvector matrix to obtain a conversion matrix; the eigenvector matrix includes the eigenvectors of the sensitive data matrix.

[0059] Step 103: Add noise that satisfies differential privacy to the transformation matrix to obtain a noise matrix.

[0060] Step 104: Project the noise matrix according to the eigenvector matrix to obtain a hidden matrix of the sensitive data matrix.

[0061] Step 105: The hidden matrix is ​​published for visitors to access, ensuring the availability of data for normal visitors to use while preventing attackers from obtaining sensitive data.

[0062] In some embodiments, before step 102, the method further includes:

[0063] Obtaining the eigenvalues ​​and eigenvectors of the sample matrix;

[0064] The target eigenvector is selected to form the above eigenvector matrix, and the target eigenvector includes the eigenvector corresponding to the eigenvalue with the largest value. Specifically, the target eigenvector may include the eigenvectors corresponding to the first several eigenvalues ​​sorted from large to small.

[0065] In some embodiments, before step 103, the method further includes:

[0066] The noise p satisfying differential privacy is determined according to the number of rows m of the eigenvector matrix, the number of columns k of the eigenvector matrix, and the differential privacy budget function ε.

[0067] The following is a detailed introduction to the above methods:

[0068] 1) Obtain sensitive data in the business system and create a sample set of all sensitive data.

[0069] 2) Standardize the sample set X and calculate the eigenvector D and matrix V k .

[0070] Represent each sample in the sample set with a vector, and then combine all samples to form an original sample matrix X with m rows and n columns. m×n (m samples, each sample has n features).

[0071] To avoid the samples being affected, the sample set X needs to be standardized, denoted as X = {x i ,i=1,2,...n}.

[0072] After standardization, find the standardized sample matrix X m×n The characteristic value D, D = diag{λ 1 ,λ 2 ,...,λ m},λ 1 ≥λ 2 ≥...≥λ m ≥0;λ 1 ,λ 2 ,...,λ m is the sample matrix X m×n diagonal elements.

[0073] For the matrix V k The following methods can be used to determine:

[0074] Arrange the above eigenvalues ​​in ascending order, take the eigenvectors corresponding to the k eigenvalues ​​with the largest values, and these k eigenvectors form the matrix V k (m rows and k columns).

[0075] 3) Using differential privacy protection processing, the sample set X is projected and transformed to obtain the projected sample set X′.

[0076] The above differential privacy protection measures include adding a noise algorithm that obeys the Laplace distribution. The noise algorithm that obeys the Laplace distribution uses the principal component analysis algorithm. The core idea is data dimensionality reduction: projecting the sample set from the high-dimensional space to the low-dimensional space to obtain a projected sample set. The projected sample set has low noise and high availability, and can prevent attackers who access the terminal from obtaining sensitive data of the protected business system to the greatest extent.

[0077] The specific implementation is as follows:

[0078] According to the above eigenvector matrix V k , project the above standardized sample matrix to obtain the projected matrix (i.e. the above transformation matrix) F n×k =X×V k .

[0079] Determine the noise p that satisfies differential privacy: ε represents the differential privacy budget function. The transformation function sensitivity of the projection transformation is The privacy budget is ε / 2,

[0080] Determine the noise matrix F noise : F noise =F n×k +p.

[0081] Determine the hidden matrix F X‘ : F X‘ =F noise ×V k

[0082] 4) The access request of the access terminal passes through the security detection module to access the sample set X'; the attacker cannot obtain sensitive data of the protected business system.

[0083] It should be noted that the access terminal can be any one or more combinations of laptops, desktops, servers, mobile devices, embedded devices, SCADA systems, and IoT devices. The security module has functions such as big data threat intelligence collision, self-learning whitelist, dynamic whitelist, blacklist, firewall, and malicious code trapping. If the security detection module determines that the visitor is malicious, when the visitor accesses sensitive data, the system will publish a hidden matrix for the visitor to access to ensure the security of sensitive data.

[0084] Corresponding to the above-mentioned active defense method based on differential privacy protection, the present application also provides an active defense system based on differential privacy protection, which includes: a differential privacy module, see Figure 2 The differential privacy module is set between the security detection module and the business system. When the security detection module detects that the visitor is malicious and the visitor requests to access sensitive data, the system publishes a hidden matrix for the visitor to access to ensure the security of sensitive data. When the security detection module detects that the visitor is not malicious, the visitor directly accesses the business system according to the visitor's access request.

[0085] It should be noted that see Figure 2 The above security detection module is composed of multiple security modules. When any security module detects that the visitor is malicious, the visitor is identified as malicious.

[0086] The above differential privacy module includes:

[0087] The matrix acquisition module is used to acquire a sensitive data matrix; the sensitive data matrix includes a sample vector, and the sample vector includes a vectorized representation of sensitive data in the business system; the sensitive data is data that needs to be protected. The specific method is to represent each sensitive data in the protected business system with a vector, and all data are combined to form a sample matrix;

[0088] A projection module, used to project the sensitive data matrix according to a characteristic vector matrix to obtain a conversion matrix; the characteristic vector matrix includes the characteristic vectors of the sensitive data matrix;

[0089] A noise matrix determination module, used for adding noise satisfying differential privacy to the transformation matrix to obtain a noise matrix;

[0090] The projection module is further used to project the noise matrix according to the eigenvector matrix to obtain a hidden matrix of the sensitive data matrix;

[0091] The publishing module is used to publish the hidden matrix for visitors to access. The projection sample matrix after the above module uses less noise, ensuring the availability of data for normal visitors to use, while preventing attackers from obtaining sensitive data.

[0092] In some embodiments, the differential privacy module further includes:

[0093] A projection function determination module, used to determine the noise p that satisfies differential privacy according to the number of rows m of the eigenvector matrix, the number of columns k of the eigenvector matrix, and the differential privacy budget function ε;

[0094] ε represents the differential privacy budget function. The transformation function sensitivity of the projection transformation is The privacy budget is ε / 2,

[0095] In some embodiments, the noise matrix determination module specifically includes:

[0096] A noise matrix determination unit is used to determine the noise matrix according to the conversion matrix F n×k And the noise p that satisfies differential privacy, determine the noise matrix F noise , F noise =F n×k +p.

[0097] This application uses this method to add effective noise to sensitive data in the protected business system when the current attacker accesses the data, while ensuring data availability to ensure that the attacker cannot successfully access the sensitive data in the business system, building an autonomous and controllable active defense system to ensure data security.

[0098] The existing differential privacy model projects the data set based on the attribute correlation of the original data set using the dimensionality reduction idea and adds random noise to the projection. The noise sensitivity is 1, the noise data is Lap(2m / ε), and the total noise is This application uses the maximum information coefficient that is closer to the actual value to analyze the original data set, and the dimensionality reduction is more ideal. The original data is projected and transformed into a lower-dimensional space. Under the same degree of privacy protection, less noise is required, and the sensitivity is The noise data is The total noise is Less total noise than previous methods.

[0099] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0100] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. An active defense method based on differential privacy protection, characterized in that: include: Obtaining a sensitive data matrix; the sensitive data matrix includes a sample vector, and the sample vector includes a vectorized representation of sensitive data in the business system; The sensitive data is data that needs to be protected; According to the eigenvector matrix, the sensitive data matrix is ​​projected to obtain a conversion matrix; The eigenvector matrix includes eigenvectors of the sensitive data matrix; The noise p satisfying differential privacy is determined according to the number of rows m of the eigenvector matrix, the number of columns k of the eigenvector matrix, and the differential privacy budget function ε. Adding noise satisfying differential privacy to the transformation matrix to obtain a noise matrix; Projecting the noise matrix according to the eigenvector matrix to obtain a hidden matrix of the sensitive data matrix; The hidden matrix is ​​published for access by visitors.

2. The active defense method based on differential privacy protection according to claim 1, characterized in that: Before projecting the sensitive data matrix according to the eigenvector matrix, the method further includes: Obtaining the eigenvalues ​​and eigenvectors of the sensitive data matrix; Select target eigenvectors to form the eigenvector matrix; the target eigenvectors include eigenvectors corresponding to eigenvalues ​​with the largest values.

3. The active defense method based on differential privacy protection according to claim 1, characterized in that: The step of adding noise satisfying differential privacy to the transformation matrix to obtain a noise matrix specifically includes: According to the projected transformation matrix F n×k And the noise p that satisfies differential privacy, determine the noise matrix F noise , F noise =F n×k +p.

4. The active defense method based on differential privacy protection according to claim 1, characterized in that: The projecting the sensitive data matrix according to the eigenvector matrix specifically includes: The sensitive data matrix is ​​projected using a first projection formula, where the first projection formula is: F n×k =X×V k Among them, F n×k represents the conversion matrix, X represents the sensitive data matrix, V k Denotes the eigenvector matrix.

5. The active defense method based on differential privacy protection according to claim 1, characterized in that: The projecting the noise matrix according to the eigenvector matrix specifically includes: The noise matrix is ​​projected using a second projection formula, where the second projection formula is: F X‘ =F noise ×V k Among them, F X‘ represents the hidden matrix of the sensitive data matrix, V k represents the eigenvector matrix, F noise Denotes the noise matrix.

6. An active defense system based on differential privacy protection, characterized in that: include: A sensitive data matrix acquisition module is used to acquire a sensitive data matrix; The sensitive data matrix includes a sample vector, and the sample vector includes a vectorized representation of sensitive data in the business system; the sensitive data is data that needs to be protected; A projection module, used to project the sensitive data matrix according to the eigenvector matrix to obtain a conversion matrix; The eigenvector matrix includes eigenvectors of the sensitive data matrix; A projection function determination module is used to determine the noise p that satisfies differential privacy according to the number of rows m of the eigenvector matrix, the number of columns k of the eigenvector matrix, and the differential privacy budget function ε. A noise matrix determination module, used for adding noise satisfying differential privacy to the transformation matrix to obtain a noise matrix; The projection module is further used to project the noise matrix according to the eigenvector matrix to obtain a hidden matrix of the sensitive data matrix; The publishing module is used to publish the hidden matrix for visitors to access.

7. The active defense system based on differential privacy protection according to claim 6, further comprising: A feature obtaining module, used to obtain the eigenvalues ​​and eigenvectors of the sensitive data matrix; The eigenvector matrix determination module is used to select target eigenvectors to form the eigenvector matrix; the target eigenvectors include the eigenvectors corresponding to the eigenvalues ​​with the largest numerical values.

8. The active defense system based on differential privacy protection according to claim 6, characterized in that: The noise matrix determination module specifically includes: A noise matrix determination unit is used to determine the noise matrix according to the conversion matrix F n×k And the noise p that satisfies differential privacy, determine the noise matrix F noise , F noise =F n×k +p.

Images