Dynamic desensitization method for data sharing scene

Abstract

The invention provides a data sharing scene-oriented dynamic desensitization method, which comprises the following steps of: receiving an original data set and set target data availability, and converting various data in the original data set into structured data; calculating a standard deviation based on the structured data of the kth type of data; determining an initial privacy parameter of the kth type of data based on the standard deviation of the kth type of data and a preset privacy parameter coefficient, and desensitizing the kth type of data based on the initial privacy parameter of the kth type of data; then, determining the information entropy and the theoretical maximum entropy of the kth type of data, and further calculating the entropy retention rate of the kth type of data; based on the difference value between the entropy retention rate of the kth type of data and the availability of the target data, dynamically adjusting the privacy parameter coefficient of the kth type of data by adopting a PID control algorithm until the difference value between the entropy retention rate of the kth type of data and the availability of the target data is smaller than a first preset threshold value, and determining the final privacy parameter coefficient of the kth type of data, and carrying out desensitization treatment.

Description

Technical Field

[0001] This application relates to the field of data security technology, specifically to a dynamic de-identification method, system, electronic device, and storage medium for data sharing scenarios. Background Technology

[0002] Currently, the mainstream data anonymization methods that focus on ensuring data availability and improving data privacy are mainly based on natural language processing (NLP) methods, such as those based on BERT and LSTM networks. NLP-based anonymization methods can accurately identify sensitive information in data and anonymize it accordingly, maximizing data availability. However, these methods do not consider the correlation between non-sensitive and sensitive information, making them vulnerable to inference attacks that could lead to the recovery of private information.

[0003] To prevent inference attacks, de-identification methods based on Generative Adversarial Networks (GANs) have emerged in recent years. These methods primarily involve adversarial training between the generator and discriminator to generate non-real data while ensuring the distribution of the generated data closely approximates the original data. While GAN-based de-identification methods can retain some important features of the original data after de-identification, it is difficult to accurately measure whether the de-identified data meets usability requirements.

[0004] To address the difficulty in measuring the usability of GAN-based data masking methods, usability assessment-based data masking methods have gained increasing attention. Researchers use predefined data quality rules for automated checks, evaluating data usability based on rules such as data integrity, accuracy, and consistency. However, rule-based masking methods require rules to be defined according to specific scenarios, making them highly limited by the usage scenario and data type, and lacking universality.

[0005] Therefore, in view of the above problems, this application proposes a dynamic de-identification method for data sharing scenarios. Summary of the Invention

[0006] This application provides a dynamic desensitization method for data sharing scenarios. Through information theory quantitative evaluation and closed-loop dynamic adjustment, it achieves a precise and adaptive optimal balance between the strength of data privacy protection and usability.

[0007] To achieve the above objectives, this application adopts the following technical solution:

[0008] Firstly, this application provides a dynamic de-identification method for data sharing scenarios, the method comprising:

[0009] S101. Receive the original dataset and the target data availability set by the user, and convert each type of data in the original dataset into structured data respectively;

[0010] S102. Based on the structured data of the k-th class, calculate the standard deviation of the k-th class of data;

[0011] S103. Based on the standard deviation of the k-th data and the preset privacy parameter coefficients, determine the initial privacy parameters of the k-th data, and then perform desensitization processing on the k-th data based on the initial privacy parameters of the k-th data;

[0012] S104. Based on the multiple states of the de-identified k-th class of data, calculate the true probability distribution of each state, and then determine the information entropy of the k-th class of data based on the true probability distribution of the multiple states in the k-th class of data.

[0013] S105. Assuming that multiple states are uniformly distributed in the k-th class of data, calculate the ideal probability distribution for each state, and then determine the theoretical maximum entropy of the k-th class of data based on the ideal probability distribution of multiple states in the k-th class of data.

[0014] S106. Based on the information entropy and theoretical maximum entropy of the k-th data, calculate the entropy retention rate of the k-th data, where the entropy retention rate represents data availability;

[0015] S107. Based on the difference between the entropy retention rate of the k-th type of data and the availability of the target data, a PID control algorithm is used to dynamically adjust the privacy parameter coefficient of the k-th type of data;

[0016] S108. Repeat steps S103 to S107 until the difference between the entropy retention rate of the k-th type of data and the availability of the target data is less than the first preset threshold, determine the final privacy parameter coefficient of the k-th type of data, and then perform desensitization processing on the k-th type of data;

[0017] Where k is a positive integer greater than 1.

[0018] Secondly, this application provides a dynamic de-identification system for data sharing scenarios, the system comprising:

[0019] The data receiving and processing module is used to receive the raw dataset and the target data availability set by the user, and to convert various types of data in the raw dataset into structured data respectively;

[0020] The standard deviation calculation module is used to calculate the standard deviation of the k-th class of data based on the structured data of the k-th class.

[0021] The initial desensitization module is used to determine the initial privacy parameters of the k-th type of data based on the standard deviation of the k-th type of data and the preset privacy parameter coefficients, and then perform desensitization processing on the k-th type of data based on the initial privacy parameters of the k-th type of data.

[0022] The information entropy calculation module is used to calculate the true probability distribution of each state based on the multiple states of the de-identified k-th class of data, and then determine the information entropy of the k-th class of data based on the true probability distribution of the multiple states in the k-th class of data.

[0023] The maximum entropy calculation module is used to calculate the ideal probability distribution of each state, assuming that multiple states are uniformly distributed in the k-th class of data, and then determine the theoretical maximum entropy of the k-th class of data based on the ideal probability distribution of multiple states in the k-th class of data.

[0024] The entropy retention rate calculation module is used to calculate the entropy retention rate of the k-th type of data based on the information entropy and the theoretical maximum entropy, where the entropy retention rate represents the data availability.

[0025] The PID control adjustment module is used to dynamically adjust the privacy parameter coefficient of the k-th type of data based on the difference between the entropy retention rate of the k-th type of data and the availability of the target data using a PID control algorithm.

[0026] The iterative convergence judgment module is used to repeatedly execute steps S103 to S107 until the difference between the entropy retention rate of the k-th type of data and the availability of the target data is less than the first preset threshold, determine the final privacy parameter coefficient of the k-th type of data, and then perform desensitization processing on the k-th type of data.

[0027] Where k is a positive integer greater than 1.

[0028] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the dynamic desensitization method of the first aspect for data sharing scenarios.

[0029] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium including storage of a computer program or instructions, which, when executed, cause the dynamic desensitization method of the first aspect for data sharing scenarios to be performed.

[0030] In this embodiment, a quantitative evaluation system based on information entropy and conditional entropy transforms the fuzzy concepts of "availability" and "privacy risk" into calculable objective indicators (i.e., entropy retention rate and privacy contribution). A PID control algorithm is then introduced, using the entropy retention rate as a feedback signal to dynamically adjust global privacy parameters, enabling data availability to automatically converge and stabilize at the user-defined target value. Finally, global parameters are differentiated based on the privacy contribution of each field, applying stronger protection to high-risk fields while maintaining overall availability. This method automatically adapts to different data characteristics, avoiding the drawbacks of traditional methods that rely on trial and error, ultimately outputting de-identified data that meets availability requirements while possessing superior privacy protection.

[0031] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A flowchart illustrating a dynamic desensitization method for data sharing scenarios provided in an embodiment of this application;

[0034] Figure 2 This is a flowchart illustrating the PID control algorithm in the dynamic desensitization method for data sharing scenarios provided in the embodiments of this application.

[0035] Figure 3 This is a schematic diagram illustrating the effect of the dynamic desensitization method for data sharing scenarios provided in the embodiments of this application. Detailed Implementation

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. At the same time, in the description of the embodiments of this application, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0037] Figure 1 This is a flowchart illustrating a dynamic desensitization method for data sharing scenarios provided in an embodiment of this application.

[0038] The process of this dynamic de-identification method for data sharing scenarios is as follows:

[0039] Step S101. Receive the original dataset and the target data availability set by the user, and convert each type of data in the original dataset into structured data.

[0040] It should be noted that the original dataset includes various types of data, including numerical data, categorical data, time-based data, text data, image data, audio data, etc., without any specific restrictions.

[0041] In this embodiment of the application, various types of data in the original dataset are converted into structured data, including:

[0042] For numerical data, retain the original numerical format;

[0043] For categorical data, label encoding is performed to convert it into discrete numerical values;

[0044] For time-based data, it is broken down into multiple independent numerical features;

[0045] For text-based data, word segmentation is performed and the data is converted into a numerical distribution based on word frequency statistics.

[0046] For image data, the distribution of pixel values ​​is statistically analyzed after grayscale processing.

[0047] For audio data, the distribution of statistical spectral features is obtained after frequency domain transformation and feature extraction.

[0048] For example, categorical data: "male" and "female" are encoded as 0 and 1 respectively; time-based data: 2024-01-15 14:30:00 is broken down into features such as year, month, day, hour, minute, and second.

[0049] In addition, each type of data needs to be processed and desensitized. The following explanation will take the k-th type of data as an example, where k is a positive integer greater than 1.

[0050] Step S102. Calculate the standard deviation of the k-th class of data based on the structured data of the k-th class.

[0051] Step S103. Based on the standard deviation of the k-th data and the preset privacy parameter coefficients, determine the initial privacy parameters of the k-th data, and then perform de-identification processing on the k-th data based on the initial privacy parameters of the k-th data.

[0052] That is, the formula for calculating the privacy parameter ε is: ε=α×σ; where α is the preset privacy parameter coefficient and σ is the standard deviation of the k-th class of data.

[0053] In this embodiment of the application, the preset rules for the preset privacy parameter coefficients are as follows:

[0054] When the availability of the target data is less than 0.7, the preset privacy parameter coefficient α is set to 1.0;

[0055] When the availability of the target data is greater than or equal to 0.7, the preset privacy parameter coefficient α is set to 0.5.

[0056] For example, if the user sets a target data availability U target =0.8, then the preset privacy parameter coefficient α =0.5.

[0057] In other words, when users expect low availability, in order to configure a lower initial availability that is closer to the target, the preset privacy parameter coefficient is set to 1;

[0058] When users expect high availability, the default privacy parameter coefficient is set to 0.5 in order to configure a higher initial availability closer to the target.

[0059] It should also be noted that the target data availability U target The value range is from 0 to 1.

[0060] Step S104. Based on the multiple states of the k-th class of data after desensitization, calculate the true probability distribution P(G) for each state. h Then, based on the true probability distribution of multiple states in the k-th class of data, the information entropy of the k-th class of data is determined.

[0061] Suppose that there are n states in the k-th class of data, with probabilities of occurrence p1, p2, ..., pn. n Next, the information entropy of the data can be calculated using the obtained probability distribution, as shown in the following formula:

[0062]

[0063] It should be noted that if the original data values ​​are relatively evenly distributed before anonymization, the entropy value will be high; conversely, if after generalization, most of the data is concentrated in a few intervals, the probability distribution will become unbalanced, leading to a significant decrease in entropy. Therefore, the magnitude of the entropy decrease can be used as a measure of the loss of data complexity.

[0064] For example, suppose the attribute "income" in the k-th data category has multiple different values ​​after subdivision, and its distribution is relatively dispersed, resulting in a high level of information entropy. After generalization, this income data is divided into three categories: "low income," "middle income," and "high income." If most of the data is concentrated in the "middle income" category, while the proportions of "low income" and "high income" are relatively low, the overall probability distribution will become extremely unbalanced, leading to a significant decrease in the calculated information entropy. By comparing the changes in entropy values ​​before and after anonymization, the usability of the data can be assessed more intuitively.

[0065] Step S105. Assuming that the multiple states in the k-th class of data are uniformly distributed, calculate the ideal probability distribution of each state, and then determine the theoretical maximum entropy of the k-th class of data based on the ideal probability distribution of the multiple states in the k-th class of data.

[0066] That is, for n states, the entropy reaches its maximum when the data is uniformly distributed: H max =log2n.

[0067] Step S106. Based on the information entropy and theoretical maximum entropy of the k-th data, calculate the entropy retention rate of the k-th data, where the entropy retention rate represents data availability.

[0068] That is, the entropy retention rate (ERR) is: This value can also be understood as the current data availability U.

[0069] Step S107. Based on the difference between the entropy retention rate of the k-th type of data and the availability of the target data, a PID control algorithm is used to dynamically adjust the privacy parameter coefficient of the k-th type of data.

[0070] Step S108. Repeat steps S103 to S107 until the difference between the entropy retention rate of the k-th type of data and the availability of the target data is less than the first preset threshold, determine the final privacy parameter coefficient of the k-th type of data, and then perform desensitization processing on the k-th type of data.

[0071] In this embodiment of the application, the first preset threshold is one-thousandth of the target data availability.

[0072] It should be noted that the PID control algorithm (Proportional-Integral-Derivative) is a classic closed-loop control strategy widely used in industrial automation, robot control, autonomous driving, and other fields. The PID controller calculates the error (the difference between the setpoint and the actual value) and combines the effects of the proportional, integral, and derivative parameters to achieve precise adjustment of the system.

[0073] Proportional (P) control: Adjusts the control input proportionally to the magnitude of the error, resulting in a rapid response, but may have steady-state error. Integral (I) control: Accumulates and corrects the error, eliminating steady-state error, but may cause overshoot and oscillation. Derivative (D) control: Predicts the error based on its rate of change, improving system response speed and reducing overshoot, but is sensitive to noise. By adjusting the PID parameters (i.e., the weighting coefficients of P, I, and D), the dynamic response of the system can be optimized, achieving a balance between accuracy, stability, and speed.

[0074] In this application, a PID control algorithm is used to dynamically adjust the preset privacy parameter coefficient α. The core formula of PID control is as follows:

[0075]

[0076] Among them, e(t)=U target -U,K p For proportional gain, fast response error, K i For integral gain, to eliminate steady-state error, K d The differential gain is used to predict the trend of error change.

[0077] By iterating the above formula repeatedly, the dynamic control of α can be achieved until the condition |e(t)| < 0.001U is met. target This means the data availability error is less than 0.1%. The overall control framework for the PID controller is shown below. Figure 2 .

[0078] In this embodiment of the application, step S8 determines the final privacy coefficient of the k-th type of data, and then performs de-identification processing on the k-th type of data, including:

[0079] Step S109. The k-th data includes multiple privacy fields. Calculate the true probability distribution of the i-th privacy field, and then determine the information entropy of the i-th privacy field based on the true probability distribution of the i-th privacy field.

[0080] That is,

[0081] Step S110. The k-th data includes multiple non-privacy fields. Calculate the conditional entropy of the i-th privacy field given that the j-th non-privacy field is known.

[0082] That is, for the privacy field S i Non-privacy field X j The conditional entropy is:

[0083]

[0084] Step S111. Based on the information entropy of the i-th privacy field and the conditional entropy of the i-th privacy field given the j-th non-privacy field, calculate the entropy decay ratio of the j-th non-privacy field to the i-th privacy field.

[0085] That is, the entropy decay ratio of the j-th non-privacy field to the i-th privacy field is...

[0086] Step S112. Repeat steps S109 to S111 until the entropy decay ratio of the j-th non-privacy field to all privacy fields is determined, and then the privacy contribution of the j-th non-privacy field is determined.

[0087] That is, the privacy contribution of the j-th non-privacy field is Where m is the number of privacy fields.

[0088] Step S113. Repeat steps S109 to S112 until the privacy contribution of all non-privacy fields is determined, and then determine the average privacy contribution of all non-privacy fields.

[0089] That is, Where n is the number of non-privacy fields.

[0090] Step S114. Preset an adjustment factor. For the j-th non-privacy field, calculate the tuning coefficient of the j-th non-privacy field based on the privacy contribution of the j-th non-privacy field and the average privacy contribution of all non-privacy fields.

[0091] That is, Here, γ is the adjustment factor, which controls the adjustment range, and is usually 0 < γ < 1.

[0092] The reason for the above steps is that if the privacy contribution of a certain field is higher than the global average Y... - If Yj is higher than the mean, it indicates that the field contributes significantly to privacy information and has a higher risk of leakage, thus requiring a higher privacy parameter. Conversely, if Yj is lower than the mean, it indicates that the field contributes less to privacy information and its privacy parameter can be reduced.

[0093] for Fields, expressions If the value is positive, multiplying it by a positive coefficient γ will make the term within the square brackets greater than 1, thereby increasing the privacy parameter of the field.

[0094] for If the field is negative, the privacy parameter will be reduced accordingly.

[0095] This method not only allows for precise control of privacy parameters in each field, but also ensures a relative balance of overall privacy parameter resources.

[0096] Step S115. Calculate the standard deviation of the data for the j-th non-privacy field, and multiply the final privacy parameter coefficient of the k-th data class by the standard deviation of the data for the j-th non-privacy field to obtain the basic privacy parameter of the j-th non-privacy field.

[0097] That is, the basic privacy parameter of the j-th non-privacy field is ε. j =α*σ j Where α is the final privacy parameter coefficient of the k-th class of data, σ j Let be the standard deviation of the j-th non-privacy field.

[0098] Step S116. Based on the tuning coefficient of the j-th non-privacy field and the basic privacy parameter of the j-th non-privacy field, determine the final privacy parameter of the j-th non-privacy field, and perform de-identification processing on the j-th non-privacy field based on the final privacy parameter of the j-th non-privacy field.

[0099] That is, the final privacy parameter of the j-th non-privacy field after adjustment is ε′. j =ε j *N j .

[0100] Where i and j are positive integers greater than 1.

[0101] In this embodiment of the application, after step S116, the method further includes:

[0102] Step S117. After desensitizing all non-privacy fields, evaluate the data availability of the k-th data to obtain the data availability evaluation result of the k-th data.

[0103] Step S118. Repeat steps S101 to S117 until the difference between the availability assessment result of the k-th type of data and the availability of the target data is less than the second preset threshold.

[0104] In this embodiment, the second preset threshold is one-thousandth of the target data availability.

[0105] In this embodiment of the application, the mean value of the tuning coefficient of the non-privacy parameter in the k-th type of data is 1.

[0106] That is, This can be understood as the overall average level of privacy parameters remaining unchanged before and after the adjustment, and the usability remaining basically unchanged.

[0107] In summary, this embodiment of the application, based on a quantitative evaluation system of information entropy and conditional entropy, transforms the fuzzy concepts of "availability" and "privacy risk" into calculable objective indicators (i.e., entropy retention rate and privacy contribution). Furthermore, a PID control algorithm is introduced, using the entropy retention rate as a feedback signal to dynamically adjust global privacy parameters, enabling data availability to automatically converge and stabilize at the user-defined target value. Finally, global parameters are differentiated based on the privacy contribution of each field, applying stronger protection to high-risk fields while maintaining overall availability. This method can automatically adapt to different data characteristics, avoiding the drawbacks of traditional methods that rely on trial and error, and ultimately outputs de-identified data that meets availability requirements while possessing superior privacy protection strength.

[0108] The above combination Figures 1-2 This paper describes a dynamic de-identification method for data sharing scenarios provided in the embodiments of this application. The following describes the application of the dynamic de-identification method for data sharing scenarios provided in the embodiments of this application.

[0109] The usability assurance test experiment applied the anonymized data to train three models of different complexities. For the simpler logistic regression model, 10,000 samples were generated using the `make_classification` dataset built into scikit-learn, with each sample containing 8 valid features and 2 redundant features, and the validation set to test set ratio was 7:3. For the simpler convolutional neural network and the more complex neural network using a VGG16 pre-trained model, the Kaggle cat and dog classification dataset was used, containing 1,500 cat and 1,500 dog images, with a training set to test set ratio of 2:1.

[0110] To visually represent the experimental results, and since the values ​​of α, U, and e are similar, the results for the three values ​​are presented in a table, and the number of iterations is used as the x-axis to plot the graph. Here, U... target Experimental images of noise desensitization under three settings (0.5, 0.8, and 0.9), along with the average time, average number of iterations, and maximum usability error for 20 experiments, are shown in the following figures. Figure 3 As shown.

[0111] The experimental results show that the average time of the 20 experiments was in the millisecond range, and the average and maximum number of iterations were both less than 100. The maximum error was 0.0805%, and the average error was 0.0562%, both less than 0.1%. This proves that the desensitization method can effectively ensure the availability of the desensitized data and meet the needs of subsequent use.

[0112] The above combination Figures 1-2 This application describes a dynamic desensitization method and application for data sharing scenarios provided in the embodiments of this application. The following describes the dynamic desensitization system for data sharing scenarios provided in the embodiments of this application.

[0113] The system includes: a data receiving and processing module, a standard deviation calculation module, an initial desensitization module, an information entropy calculation module, a maximum entropy calculation module, an entropy retention rate calculation module, a PID control adjustment module, and an iterative convergence judgment module, as detailed below.

[0114] The data receiving and processing module is used to receive the raw dataset and the target data availability set by the user, and to convert various types of data in the raw dataset into structured data respectively;

[0115] The standard deviation calculation module is used to calculate the standard deviation of the k-th class of data based on the structured data of the k-th class.

[0116] The initial desensitization module is used to determine the initial privacy parameters of the k-th type of data based on the standard deviation of the k-th type of data and the preset privacy parameter coefficients, and then perform desensitization processing on the k-th type of data based on the initial privacy parameters of the k-th type of data.

[0117] The information entropy calculation module is used to calculate the true probability distribution of each state based on the multiple states of the de-identified k-th class of data, and then determine the information entropy of the k-th class of data based on the true probability distribution of the multiple states in the k-th class of data.

[0118] The maximum entropy calculation module is used to calculate the ideal probability distribution of each state, assuming that multiple states are uniformly distributed in the k-th class of data, and then determine the theoretical maximum entropy of the k-th class of data based on the ideal probability distribution of multiple states in the k-th class of data.

[0119] The entropy retention rate calculation module is used to calculate the entropy retention rate of the k-th type of data based on the information entropy and the theoretical maximum entropy, where the entropy retention rate represents the data availability.

[0120] The PID control adjustment module is used to dynamically adjust the privacy parameter coefficient of the k-th type of data based on the difference between the entropy retention rate of the k-th type of data and the availability of the target data using a PID control algorithm.

[0121] The iterative convergence judgment module is used to repeatedly execute steps S103 to S107 until the difference between the entropy retention rate of the k-th type of data and the availability of the target data is less than the first preset threshold, determine the final privacy parameter coefficient of the k-th type of data, and then perform desensitization processing on the k-th type of data.

[0122] Where k is a positive integer greater than 1.

[0123] Furthermore, the specific implementation of the above system is basically similar to the method implementation, so the description is relatively simple. For relevant details, please refer to the description of the method implementation. Moreover, it should be noted that in the various modules of the system of this application, the components are logically divided according to the functions they are to perform. However, this application is not limited to this and can re-divide or combine the components as needed.

[0124] In another aspect, the present invention provides an electronic device for implementing the above-described dynamic desensitization method for data sharing scenarios. This electronic device is not limited to a terminal device or server within a system. The electronic device includes, but is not limited to, a memory and a processor. The memory stores a computer program, and the processor is configured to execute the steps of any of the above method embodiments via the computer program.

[0125] In another aspect, the present invention provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional embodiments of the dynamic de-identification method for data sharing scenarios described above. The computer program is configured to execute the steps in any of the above method embodiments at runtime.

[0126] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and still achieve the desired result. Furthermore, the specific order or sequential order shown in the drawings is not necessarily required to achieve the desired result; in some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0127] The above description is merely an 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 principle of this application should be included within the scope of the claims of this application.

Claims

1. A dynamic desensitization method for data sharing scenarios, characterized in that, include: S101. Receive the original dataset and the target data availability set by the user, and convert each type of data in the original dataset into structured data respectively; S102. Based on the structured data of the k-th class, calculate the standard deviation of the k-th class of data; S103. Based on the standard deviation of the k-th data and the preset privacy parameter coefficients, determine the initial privacy parameters of the k-th data, and then perform desensitization processing on the k-th data based on the initial privacy parameters of the k-th data; S104. Based on the multiple states of the de-identified k-th class of data, calculate the true probability distribution of each state, and then determine the information entropy of the k-th class of data based on the true probability distribution of the multiple states in the k-th class of data. S105. Assuming that multiple states are uniformly distributed in the k-th class of data, calculate the ideal probability distribution for each state, and then determine the theoretical maximum entropy of the k-th class of data based on the ideal probability distribution of multiple states in the k-th class of data. S106. Based on the information entropy and theoretical maximum entropy of the k-th data, calculate the entropy retention rate of the k-th data, wherein the entropy retention rate represents data availability; S107. Based on the difference between the entropy retention rate of the k-th type of data and the availability of the target data, a PID control algorithm is used to dynamically adjust the privacy parameter coefficient of the k-th type of data; S108. Repeat steps S103 to S107 until the difference between the entropy retention rate of the k-th type of data and the availability of the target data is less than the first preset threshold, determine the final privacy parameter coefficient of the k-th type of data, and then perform desensitization processing on the k-th type of data; Wherein, k is a positive integer greater than 1.

2. The dynamic desensitization method for data sharing scenarios according to claim 1, characterized in that, The step S108, which involves determining the final privacy coefficient of the k-th data category and then performing de-identification processing on the k-th data category, includes: S109. The k-th type of data includes multiple privacy fields. Calculate the true probability distribution of the ith privacy field, and then determine the information entropy of the ith privacy field based on the true probability distribution of the ith privacy field. S110. The k-th type of data includes multiple non-privacy fields. Calculate the conditional entropy of the i-th privacy field when the j-th non-privacy field is known. S111. Based on the information entropy of the i-th privacy field and the conditional entropy of the i-th privacy field when the j-th non-privacy field is known, calculate the entropy decay ratio of the j-th non-privacy field to the i-th privacy field; S112. Repeat steps S109 to S111 until the entropy decay ratio of the j-th non-privacy field to all privacy fields is determined, and then the privacy contribution of the j-th non-privacy field is determined. S113. Repeat steps S109 to S112 until the privacy contribution of all non-privacy fields is determined, and then determine the average privacy contribution of all non-privacy fields. S114. Preset an adjustment factor. For the j-th non-privacy field, calculate the tuning coefficient of the j-th non-privacy field based on the privacy contribution of the j-th non-privacy field and the average privacy contribution of all non-privacy fields. S115. Calculate the data standard deviation of the j-th non-privacy field, and multiply the final privacy parameter coefficient of the k-th data with the data standard deviation of the j-th non-privacy field to obtain the basic privacy parameter of the j-th non-privacy field; S116. Based on the tuning coefficient of the j-th non-privacy field and the basic privacy parameter of the j-th non-privacy field, determine the final privacy parameter of the j-th non-privacy field, and perform desensitization processing on the j-th non-privacy field based on the final privacy parameter of the j-th non-privacy field. Where i and j are positive integers greater than 1.

3. The dynamic desensitization method for data sharing scenarios according to claim 2, characterized in that, After step S116, the method further includes: S117. After desensitizing all non-privacy fields, the data usability of the k-th data is evaluated to obtain the usability evaluation result of the k-th data. S118. Repeat steps S101 to S117 until the difference between the availability assessment result of the k-th type of data and the availability of the target data is less than the second preset threshold.

4. The dynamic desensitization method for data sharing scenarios according to claim 1, characterized in that, The step of converting various types of data in the original dataset into structured data includes: For numerical data, retain the original numerical format; For categorical data, label encoding is performed to convert it into discrete numerical values; For time-based data, it is broken down into multiple independent numerical features; For text-based data, word segmentation is performed and the data is converted into a numerical distribution based on word frequency statistics. For image data, the distribution of pixel values ​​is statistically analyzed after grayscale processing. For audio data, the distribution of statistical spectral features is obtained after frequency domain transformation and feature extraction.

5. The dynamic desensitization method for data sharing scenarios according to claim 1, characterized in that, The preset rules for the preset privacy parameter coefficients are as follows: When the availability of the target data is less than 0.7, the preset privacy parameter coefficient is set to 1.0; When the availability of the target data is greater than or equal to 0.7, the preset privacy parameter coefficient is set to 0.

5.

6. The dynamic desensitization method for data sharing scenarios according to claim 3, characterized in that, Both the first preset threshold and the second preset threshold are one-thousandth of the target data availability.

7. The dynamic desensitization method for data sharing scenarios according to claim 1, characterized in that, The mean value of the tuning coefficients of the non-privacy parameters in the k-th type of data is 1.

8. A dynamic desensitization system for data sharing scenarios, characterized in that, The system includes: The data receiving and processing module is used to receive the raw dataset and the target data availability set by the user, and to convert various types of data in the raw dataset into structured data respectively; The standard deviation calculation module is used to calculate the standard deviation of the k-th class of data based on the structured data of the k-th class. The initial desensitization module is used to determine the initial privacy parameters of the k-th type of data based on the standard deviation of the k-th type of data and the preset privacy parameter coefficients, and then perform desensitization processing on the k-th type of data based on the initial privacy parameters of the k-th type of data. The information entropy calculation module is used to calculate the true probability distribution of each state based on the multiple states of the de-identified k-th class of data, and then determine the information entropy of the k-th class of data based on the true probability distribution of the multiple states in the k-th class of data. The maximum entropy calculation module is used to calculate the ideal probability distribution of each state, assuming that multiple states are uniformly distributed in the k-th class of data, and then determine the theoretical maximum entropy of the k-th class of data based on the ideal probability distribution of multiple states in the k-th class of data. The entropy retention rate calculation module is used to calculate the entropy retention rate of the k-th type of data based on the information entropy and the theoretical maximum entropy, wherein the entropy retention rate represents the data availability; The PID control adjustment module is used to dynamically adjust the privacy parameter coefficient of the k-th type of data based on the difference between the entropy retention rate of the k-th type of data and the availability of the target data using a PID control algorithm. The iterative convergence judgment module is used to repeatedly execute steps S103 to S107 until the difference between the entropy retention rate of the k-th type of data and the availability of the target data is less than the first preset threshold, determine the final privacy parameter coefficient of the k-th type of data, and then perform desensitization processing on the k-th type of data. Wherein, k is a positive integer greater than 1.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the dynamic desensitization method for data sharing scenarios as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a computer program or instructions that, when executed, cause the method as described in any one of claims 1-7 to be performed.

Images