A multi-dimensional data level association discrimination method based on re-identification attack

Abstract

The present invention proposes a multi-dimensional data level association discrimination method based on re-identification attack, comprising the following steps: S1, processing the sampled data into a structured format to obtain a sample data table; S2, selecting a corresponding identification method according to the attributes of the data in the sample data table to identify the data type and the initial sensitivity level; S3, calculating the re-identification risk assessment value of the data in the sample data table through encoding and data generation technology; S4, adjusting the initial sensitivity level of the data using the re-identification risk assessment value; the present invention can effectively assess the re-identification risk of data, thereby scientifically and rationally managing the data in a graded manner according to the value of different data.

Description

Technical Field

[0001] The present invention relates to the field of data security technology, and in particular to a multi-dimensional data level association discrimination method based on re-identification attack. Background Art

[0002] In the modern information society, data has become a core resource to promote scientific and technological progress and business development. With the widespread application of big data technology, enterprises and institutions can extract valuable insights from large amounts of data to make more informed decisions. However, in the process of data processing and analysis, the value of different data varies greatly. Therefore, how to effectively determine the value of data and classify and grade the data accordingly has become a key issue in data management.

[0003] Reference Chinese patent name: A method for constructing a data classification and grading domain knowledge base based on information extraction, its application number: CN202210896400.XA, discloses a method for constructing a data classification and grading domain knowledge base based on information extraction, and relates to the field of natural language processing technology. The present invention includes a document acquisition step, a document preprocessing step, a text data extraction step, a table information extraction step, a data classification and grading tuple extraction step, and a data classification and grading domain knowledge base construction step. The invention proposes an automatic policy and regulation parsing framework, and constructs the classification and grading information into a domain knowledge base, giving full play to the guiding role of policies and regulations on data classification and grading, which can effectively bridge the gap between national-level data protection concerns and specific countermeasures of organizations, and the framework can carry out more research in the future.

[0004] The value of data is usually closely related to its uniqueness, scarcity and sensitivity. High-value data usually contains more detailed and useful information and has a higher potential for use in business and research. Especially in scenarios that require decision support, market analysis or R&D innovation, high-value data often plays a vital role. However, due to its sensitivity and high utilization value, these data are often more likely to become the target of malicious attacks, especially re-identification attacks.

[0005] A re-identification attack is an attack that re-identifies the identity of an individual in a data set by matching de-identified data with other data sources. Although de-identified data has removed obvious personal identifiers, attackers can use some key features (such as date of birth, gender, zip code, etc.) to match, thereby revealing the real identity information hidden behind the data. Since high-value data is often richer and more complex in feature dimensions, it is easier to be re-identified and faces a higher risk of leakage. In order to accurately determine the value of data in data management and classify data according to the risk of re-identification, existing technical means are insufficient. On the one hand, traditional de-identification technology mainly focuses on how to reduce the identification risk of data rather than evaluating the value of the data itself. On the other hand, the risk of re-identification attacks is usually difficult to quantify, especially in complex scenarios involving multi-dimensional data associations, which requires a technical means that can comprehensively evaluate data risks and values. Summary of the invention

[0006] The present invention solves the problem that the prior art is unable to classify data according to the re-identification risk, and proposes a multi-dimensional data level association discrimination method based on re-identification attacks, which can effectively evaluate the re-identification risk of data, thereby scientifically and reasonably managing the classification of different data according to their value.

[0007] In order to achieve the above objectives, the following technical solutions are proposed: A multi-dimensional data level association discrimination method based on re-identification attack includes the following steps: S1, processing the sampled data into a structured format to obtain a sample data table; S2, selecting a corresponding identification method according to the attributes of the data in the sample data table to identify the data type and initial sensitivity level; S3, calculate the re-identifiable risk assessment value of the data in the sample data table through coding and data generation techniques; S4, using the re-identifiable risk assessment value to adjust the initial sensitivity level of the data.

[0008] The present invention evaluates the difficulty of data re-identification by simulating re-identification attacks, thereby indirectly evaluating the value of data and classifying the data. Specifically, the present method uses re-identification attack technology to conduct risk assessment on different columns and their combinations in the data, and determines their recognition probability in multi-source data matching. According to the recognition probability, the value of the data is classified, and the sensitivity level of the data is adjusted accordingly.

[0009] This method can not only help data managers effectively identify high-value data, but also provide a scientific basis for data classification and protection by quantifying re-identification risks. In practical applications, data managers can use this method to optimize data storage, sharing and utilization strategies, thereby maximizing the commercial and research value of data while ensuring data security.

[0010] In summary, the introduction of the multidimensional data association level determination model provides an innovative tool for modern data management, which can effectively solve the key problems in data value judgment and risk assessment. At a time when data is becoming an increasingly important asset, this technology has broad application prospects and important practical significance.

[0011] The present invention simulates attacks on different data columns and their combinations in structured data tables, calculates the probability of re-identification, and re-evaluates the security level of the data based on the re-identification risk value. The model uses multi-source discrimination technology to determine the column combination with high re-identification probability through extensive data expansion of the target data set, and adjusts the data level accordingly to improve the ability to accurately manage and control data assets.

[0012] Preferably, the S1 specifically includes the following steps: importing sampled data and processing it into a structured format, and replacing blank data with an empty string.

[0013] The S1 of the present invention specifically includes the following steps: importing data and preprocessing the data into a structured format, specifically including: reading a data table, reading the data in each row and column of the table in a string format, replacing blank data with an empty string, and generating a sample data table T. The sampling amount of the data must meet certain number requirements, with a minimum of no less than 1,000 items, and more than 5,000 items are recommended.

[0014] Preferably, the attributes of the data in S2 include data length and character composition type, the data length includes fixed length and variable length, and the character composition type includes numeric type and extended character type as well as letter type and Chinese character type.

[0015] The data attributes of the present invention include data length and character composition type, wherein the data length includes fixed length and variable length, and the character composition type includes numerical type, letter type, Chinese character type and extended character type.

[0016] Preferably, the recognition method includes regular expression recognition and keyword matching recognition as well as named entity recognition and generation rule recognition, and data with different attributes adopt one or more of the recognition methods.

[0017] Preferably, for numerical data, the data type is inferred based on the numerical range and aggregation characteristics.

[0018] Preferably, the S2 specifically comprises the following steps: S21, identifying each column of the sample data table according to the basic attributes of the data to obtain the attribute recognition result of the data; S22, selecting different recognition methods to judge the data type according to the attribute recognition result; S23, the initial level of the data is obtained by matching the data type with the data classification and grading standards of each industry. If the data type cannot be identified, the data is considered to be non-sensitive data and is defined as the lowest sensitivity level.

[0019] Preferably, the S3 specifically includes the following steps: using an encoder to process each column in the sample data table, mapping different string values ​​in the column to different numbers; using a GaussCopulas model to fit and generate data, and calculating re-identifiable risk assessment values ​​of the original data and the generated data.

[0020] The S3 of the present invention specifically includes the following steps: encoding the data using an encoder, fitting and generating the data through a GaussCopulas model, calculating the re-identification probability of the generated data and the original data, and evaluating the re-identification risk of the data based on the calculation result of the re-identification risk.

[0021] Preferably, the re-identifiable risk assessment value calculation process is as follows: S31, use LabelEncoder encoder to process each column in the sample data table; S32, passing the encoded sample data table as Tnum to the GaussCopulas model for fitting; S33, using the fitted GaussCopulas model to generate data, rounding the generated data and trimming the values ​​that exceed the numerical range of the corresponding column in Tnum to obtain Tsysthsize; S34, sampling data Tsample from Tsysthsize; S35, assuming that there are N rows of data in Tsample, and the nth row of data in Tsample obtains p rows of the same data in Tsysthsize, the calculation process of the re-identifiable risk valuation R is as follows:

[0022] Preferably, the S4 specifically comprises the following steps: Set columns whose data types cannot be identified to the lowest sensitivity level; Adjust the sensitivity level of the columns whose re-identifiable risk estimates are greater than 50% in the initial sensitivity level by upgrading them in stages based on the size of the re-identifiable risk estimates.

[0023] The beneficial effects of the present invention are as follows: the present invention simulates attacks on different data columns and their combinations in a structured data table, calculates the probability of re-identification, and re-evaluates the security level of the data based on the re-identification risk value. The model uses multi-source discrimination technology to determine the column combination with high re-identification probability through extensive data expansion of the target data set, and adjusts the data level accordingly to improve the ability to accurately manage and control data assets. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the method of the present invention.

[0025] Figure 2 It is a specific flow chart of S4 of the present invention. DETAILED DESCRIPTION

[0026] Embodiment 1: This embodiment proposes a multi-dimensional data level association determination method based on re-identification attack, referring to Figure 1 , specifically including the following steps: S1, data preprocessing: import data and preprocess it into a structured format. The sampling amount of data must meet certain number requirements to obtain a sample data table T; The S1 of the present invention specifically includes the following steps: importing data and preprocessing the data into a structured format, specifically including: reading a data table, reading the data in each row and column of the table in a string format, replacing blank data with an empty string, and generating a sample data table T. The sampling amount of the data must meet certain number requirements, with a minimum of no less than 1,000 items, and more than 5,000 items are recommended.

[0027] S2, data type identification: identify the data category of the sample data table T according to the attributes of the data, and use a specific identification method to determine the data type and initial sensitivity level; the attributes of the data of the present invention include data length and character composition type, wherein the data length includes fixed length and variable length, and the character composition type includes numeric type, letter type, Chinese character type and extended character type.

[0028] S3, re-identification risk assessment: Calculate the re-identification probability of the data in the sample data table T through encoding and data generation technology, that is, the re-identification risk assessment value, and evaluate the re-identification risk of the data; S4, data classification adjustment: adjust the sensitivity level of the data according to the re-identification risk assessment result. S4 of the present invention specifically includes the following steps: Figure 2, based on the results of step two and step three, if the re-identifiable risk valuation of T is greater than 50%, then one level is added to the original level, specifically: for columns that can identify the data type, they are graded and upgraded based on the re-identifiable risk value. For columns that cannot identify the data type, they are set to the lowest sensitivity level. The graded upgrade can be upgraded in intervals according to the size of the re-identifiable risk valuation, such as 50%-60% upgrade 1 level, 60%-70% upgrade 2 levels, and so on. The present invention also takes into account whether the data column with the lowest sensitivity level in step two has a risk of re-identification, and sets the data column with a re-identifiable risk value greater than 0 in the lowest sensitivity level as the re-identification sensitivity level.

[0029] The present invention evaluates the difficulty of data re-identification by simulating re-identification attacks, thereby indirectly evaluating the value of data and classifying the data. Specifically, the present method uses re-identification attack technology to conduct risk assessment on different columns and their combinations in the data, and determines their recognition probability in multi-source data matching. According to the recognition probability, the value of the data is classified, and the sensitivity level of the data is adjusted accordingly.

[0030] This method can not only help data managers effectively identify high-value data, but also provide a scientific basis for data classification and protection by quantifying re-identification risks. In practical applications, data managers can use this method to optimize data storage, sharing and utilization strategies, thereby maximizing the commercial and research value of data while ensuring data security.

[0031] In summary, the introduction of the multidimensional data association level determination model provides an innovative tool for modern data management, which can effectively solve the key problems in data value judgment and risk assessment. At a time when data is becoming an increasingly important asset, this technology has broad application prospects and important practical significance.

[0032] The present invention simulates attacks on different data columns and their combinations in structured data tables, calculates the probability of re-identification, and re-evaluates the security level of the data based on the re-identification risk value. The model uses multi-source discrimination technology to determine the column combination with high re-identification probability through extensive data expansion of the target data set, and adjusts the data level accordingly to improve the ability to accurately manage and control data assets.

[0033] Embodiment 2: This embodiment optimizes S2 on the basis of embodiment 1 and proposes a multi-dimensional data level association discrimination method based on re-identification attack. Figure 1 , specifically including the following steps: S1, data preprocessing: import data and preprocess it into a structured format. The sampling amount of data must meet certain number requirements to obtain a sample data table T; The S1 of the present invention specifically includes the following steps: importing data and preprocessing the data into a structured format, specifically including: reading a data table, reading the data in each row and column of the table in a string format, replacing blank data with an empty string, and generating a sample data table T. The sampling amount of the data must meet certain number requirements, with a minimum of no less than 1,000 items, and more than 5,000 items are recommended.

[0034] S2, data type identification: identify the data category of the sample data table T according to the attributes of the data, and use a specific identification method to determine the data type and initial sensitivity level; the attributes of the data of the present invention include data length and character composition type, wherein the data length includes fixed length and variable length, and the character composition type includes numeric type, letter type, Chinese character type and extended character type.

[0035] The S2 of the present invention specifically includes the following steps: according to the data classification and grading standards of various industries, according to the attributes of the data, the category and level of the data are identified, specifically including: S21, for a column of the sample data table T, identify the sample to be identified according to the basic attributes of the data. The attributes of the data are shown in Table 1 below: Table 1 Data attribute table S22, according to the data attribute recognition result, adopt different data recognition schemes, and the specific recognition method is shown in the following Table 2: Table 2 Comparison table of data attributes and identification methods The specific identification method implementation method specifically includes: Inferring data types based on numerical ranges and aggregate features: First, calculate the distribution and range of values ​​in the data set, such as minimum, maximum, mean, standard deviation, etc. Then infer data types based on these statistical features. Regular expressions: Write regular expressions to match specific data formats, such as dates (\d{4}-\d{2}-\d{2}), phone numbers, email addresses, etc. By applying regular expressions to scan data, the matching pattern may represent the type of data.

[0036] Generation rule identification: For data with known generation rules, the data is identified by judging whether it meets the generation rule identification type, such as ID card number and license plate number.

[0037] Keyword matching: Searching a dataset for specific keywords or phrases, which are often associated with a particular type of data.

[0038] Named Entity Recognition: Identify entity types in data, such as place names, organization names, and personal names, through natural language processing models (such as BERT or spaCy). The model identifies and classifies entities based on context and word features to infer data types.

[0039] S23, after identifying the data type, determine the initial level of the data by querying national, industry and enterprise-related standards; if the data type cannot be identified, it is defined as non-sensitive data, and no level is the lowest sensitivity level.

[0040] S3, re-identification risk assessment: Calculate the re-identification probability of the data in the sample data table T through encoding and data generation technology, that is, the re-identification risk assessment value, and evaluate the re-identification risk of the data; S4, data classification adjustment: adjust the sensitivity level of the data according to the re-identification risk assessment result. S4 of the present invention specifically includes the following steps: Figure 2 , based on the results of step two and step three, if the re-identifiable risk valuation of T is greater than 50%, then one level is added to the original level, specifically: for columns that can identify the data type, they are graded and upgraded based on the re-identifiable risk value. For columns that cannot identify the data type, they are set to the lowest sensitivity level. The graded upgrade can be upgraded in intervals according to the size of the re-identifiable risk valuation, such as 50%-60% upgrade 1 level, 60%-70% upgrade 2 levels, and so on. The present invention also takes into account whether the data column with the lowest sensitivity level in step two has a risk of re-identification, and sets the data column with a re-identifiable risk value greater than 0 in the lowest sensitivity level as the re-identification sensitivity level.

[0041] The present invention evaluates the difficulty of data re-identification by simulating re-identification attacks, thereby indirectly evaluating the value of data and classifying the data. Specifically, the present method uses re-identification attack technology to conduct risk assessment on different columns and their combinations in the data, and determines their recognition probability in multi-source data matching. According to the recognition probability, the value of the data is classified, and the sensitivity level of the data is adjusted accordingly.

[0042] This method can not only help data managers effectively identify high-value data, but also provide a scientific basis for data classification and protection by quantifying re-identification risks. In practical applications, data managers can use this method to optimize data storage, sharing and utilization strategies, thereby maximizing the commercial and research value of data while ensuring data security.

[0043] In summary, the introduction of the multidimensional data association level determination model provides an innovative tool for modern data management, which can effectively solve the key problems in data value judgment and risk assessment. At a time when data is becoming an increasingly important asset, this technology has broad application prospects and important practical significance.

[0044] The present invention simulates attacks on different data columns and their combinations in structured data tables, calculates the probability of re-identification, and re-evaluates the security level of the data based on the re-identification risk value. The model uses multi-source discrimination technology to determine the column combination with high re-identification probability through extensive data expansion of the target data set, and adjusts the data level accordingly to improve the ability to accurately manage and control data assets.

[0045] Embodiment 3: This embodiment optimizes S3 on the basis of embodiment 2, and proposes a multi-dimensional data level association discrimination method based on re-identification attack. Figure 1 , specifically including the following steps: S1, data preprocessing: import data and preprocess it into a structured format. The sampling amount of data must meet certain number requirements to obtain a sample data table T; The S1 of the present invention specifically includes the following steps: importing data and preprocessing the data into a structured format, specifically including: reading a data table, reading the data in each row and column of the table in a string format, replacing blank data with an empty string, and generating a sample data table T. The sampling amount of the data must meet certain number requirements, with a minimum of no less than 1,000 items, and more than 5,000 items are recommended.

[0046] S2, data type identification: identify the data category of the sample data table T according to the attributes of the data, and use a specific identification method to determine the data type and initial sensitivity level; the attributes of the data of the present invention include data length and character composition type, wherein the data length includes fixed length and variable length, and the character composition type includes numeric type, letter type, Chinese character type and extended character type.

[0047] The S2 of the present invention specifically includes the following steps: according to the data classification and grading standards of various industries, according to the attributes of the data, the category and level of the data are identified, specifically including: S21, for a column of the sample data table T, identify the sample to be identified according to the basic attributes of the data, the attributes of the data are shown in Table 1 of Example 2: S22, according to the data attribute recognition result, adopt different data recognition schemes, and the specific recognition method is shown in Table 2 of Example 2: The specific identification method implementation method specifically includes: Inferring data types based on numerical ranges and aggregate features: First, calculate the distribution and range of values ​​in the data set, such as minimum, maximum, mean, standard deviation, etc. Then infer data types based on these statistical features. Regular expressions: Write regular expressions to match specific data formats, such as dates (\d{4}-\d{2}-\d{2}), phone numbers, email addresses, etc. By applying regular expressions to scan data, the matching pattern may represent the type of data.

[0048] Generation rule identification: For data with known generation rules, the data is identified by judging whether it meets the generation rule identification type, such as ID card number and license plate number.

[0049] Keyword matching: Searching a dataset for specific keywords or phrases, which are often associated with a particular type of data.

[0050] Named Entity Recognition: Identify entity types in data, such as place names, organization names, and personal names, through natural language processing models (such as BERT or spaCy). The model identifies and classifies entities based on context and word features to infer data types.

[0051] S23, after identifying the data type, determine the initial level of the data by querying national, industry and enterprise-related standards; if the data type cannot be identified, it is defined as non-sensitive data, and no level is the lowest sensitivity level.

[0052] S3, re-identification risk assessment: Calculate the re-identification probability of the data in the sample data table T through encoding and data generation technology, that is, the re-identification risk assessment value, and evaluate the re-identification risk of the data; The S3 of the present invention specifically includes the following steps: encoding the data using an encoder, fitting and generating the data through a GaussCopulas model, calculating the re-identification probability of the generated data and the original data, and evaluating the re-identification risk of the data based on the calculation result of the re-identification risk.

[0053] The specific calculation process of the re-identifiable risk assessment value of the present invention is as follows: S31, use the LabelEncoder encoder to process each column in the data table T, map different string values ​​in the column to different numbers, each column should have an independent mapping relationship, and different columns cannot be mixed.

[0054] S32, passing the encoded data table as Tnum to the GaussCopulas model for fitting.

[0055] S33, use the fitted GaussCopulas model to generate data. When the amount of data in data table T is no more than 5,000, the amount of data generated is generally ten times that of data table T. It can be adjusted according to the specific situation, but the amount of data generated cannot be less than that of data table T. Then the generated data is rounded to the nearest integer, and the values ​​that exceed the numerical range of the corresponding column in Tnum are trimmed. The processed generated data is used as Tsysthsize.

[0056] S34, samples data from Tsysthsize. The amount of sampled data is usually consistent with the data table T, and can also be adjusted according to specific circumstances. The sampled data is set to Tsample.

[0057] S35, suppose Tsample has N rows of data in total, and the nth row of data in Tsample can obtain p rows of the same data in Tsysthsize, denoted as pn, then the re-identifiable risk valuation of T is R.

[0058]

[0059] S4, data classification adjustment: adjust the sensitivity level of the data according to the re-identification risk assessment result. S4 of the present invention specifically includes the following steps: Figure 2 , based on the results of step two and step three, if the re-identifiable risk valuation of T is greater than 50%, then one level is added to the original level, specifically: for columns that can identify the data type, they are graded and upgraded based on the re-identifiable risk value. For columns that cannot identify the data type, they are set to the lowest sensitivity level. The graded upgrade can be upgraded in intervals according to the size of the re-identifiable risk valuation, such as 50%-60% upgrade 1 level, 60%-70% upgrade 2 levels, and so on. The present invention also takes into account whether the data column with the lowest sensitivity level in step two has a risk of re-identification, and sets the data column with a re-identifiable risk value greater than 0 in the lowest sensitivity level as the re-identification sensitivity level.

[0060] The present invention evaluates the difficulty of data re-identification by simulating re-identification attacks, thereby indirectly evaluating the value of data and classifying the data. Specifically, the present method uses re-identification attack technology to conduct risk assessment on different columns and their combinations in the data, and determines their recognition probability in multi-source data matching. According to the recognition probability, the value of the data is classified, and the sensitivity level of the data is adjusted accordingly.

[0061] This method can not only help data managers effectively identify high-value data, but also provide a scientific basis for data classification and protection by quantifying re-identification risks. In practical applications, data managers can use this method to optimize data storage, sharing and utilization strategies, thereby maximizing the commercial and research value of data while ensuring data security.

[0062] In summary, the introduction of the multidimensional data association level determination model provides an innovative tool for modern data management, which can effectively solve the key problems in data value judgment and risk assessment. At a time when data is becoming an increasingly important asset, this technology has broad application prospects and important practical significance.

[0063] The present invention simulates attacks on different data columns and their combinations in structured data tables, calculates the probability of re-identification, and re-evaluates the security level of the data based on the re-identification risk value. The model uses multi-source discrimination technology to determine the column combination with high re-identification probability through extensive data expansion of the target data set, and adjusts the data level accordingly to improve the ability to accurately manage and control data assets.

Claims

1. A multi-dimensional data level association discrimination method based on re-identification attack, characterized in that: The following steps are involved: S1, processing the sampled data into a structured format to obtain a sample data table; S2, selecting a corresponding identification method according to the attributes of the data in the sample data table to identify the data type and initial sensitivity level; S3, calculate the re-identifiable risk assessment value of the data in the sample data table through coding and data generation techniques; S4, adjust the initial sensitivity level of the data using the re-identifiable risk assessment value.

2. According to the method for determining multidimensional data level association based on re-identification attack according to claim 1, it is characterized in that: The S1 specifically includes the following steps: importing sampled data and processing it into a structured format, and replacing blank data with an empty string.

3. A multi-dimensional data level association discrimination method based on re-identification attack according to claim 1 or 2, characterized in that: The sampling amount of the data is at least 1000.

4. The multi-dimensional data level association determination method based on re-identification attack according to claim 1 is characterized in that: The attributes of the data in S2 include data length and character composition type. The data length includes fixed length and variable length. The character composition type includes numeric type, extended character type, letter type and Chinese character type.

5. The multi-dimensional data level association determination method based on re-identification attack according to claim 4 is characterized in that: The recognition method includes regular expression recognition and keyword matching recognition as well as named entity recognition and generation rule recognition.

6. The multi-dimensional data level association determination method based on re-identification attack according to claim 5 is characterized in that: For numeric data, the data type is inferred based on the numeric range and aggregation characteristics.

7. A multi-dimensional data level association determination method based on re-identification attack according to any one of claims 4 to 6, characterized in that: The S2 specifically includes the following steps: S21, identifying each column of the sample data table according to the basic attributes of the data to obtain the attribute recognition result of the data; S22, selecting different recognition methods to judge the data type according to the attribute recognition result; S23, the initial level of the data is obtained by matching the data type with the data classification and grading standards of each industry. If the data type cannot be identified, the data is considered to be non-sensitive data and is defined as the lowest sensitivity level.

8. The multi-dimensional data level association determination method based on re-identification attack according to claim 1 is characterized in that: The S3 specifically includes the following steps: using an encoder to process each column in the sample data table, mapping different string values ​​in the column to different numbers; using a GaussCopulas model to fit and generate data, and calculating the re-identifiable risk assessment value of the original data and the generated data.

9. The multi-dimensional data level association determination method based on re-identification attack according to claim 8 is characterized in that: The re-identifiable risk assessment value calculation process is as follows: S31, use LabelEncoder encoder to process each column in the sample data table; S32, passing the encoded sample data table as Tnum to the GaussCopulas model for fitting; S33, using the fitted GaussCopulas model to generate data, rounding the generated data and trimming the values ​​that exceed the numerical range of the corresponding column in Tnum to obtain Tsysthsize; S34, sampling data Tsample from Tsysthsize; S35, assuming that there are N rows of data in Tsample, and the nth row of data in Tsample obtains p rows of the same data in Tsysthsize, the calculation process of the re-identifiable risk valuation R is as follows:

10. A multi-dimensional data level association determination method based on re-identification attack according to claim 8 or 9, characterized in that: The S4 specifically comprises the following steps: Set columns whose data types cannot be identified to the lowest sensitivity level; Adjust the sensitivity level of the columns whose re-identifiable risk estimates are greater than 50% in the initial sensitivity level by upgrading them in stages based on the size of the re-identifiable risk estimates.

Images