Data desensitization method and system

Abstract

The invention relates to the technical field of data security, and discloses a data desensitization method and system. The method comprises the following steps: analyzing a data access request of a user, and determining a target data structure; based on the target data structure, a target classification and grading label is determined, and the classification and grading label is generated by analyzing the initial service information and the system data source by using a large model in advance and is matched with the corresponding data structure in advance based on a preset matching rule; a target desensitization strategy is determined based on the target classification and grading label, and the desensitization strategy is determined by analyzing the classification and grading label of the corresponding data in the system data source through a large model in advance; obtaining to-be-accessed data based on the data access request; and performing desensitization processing on the to-be-accessed data by using the target desensitization strategy to obtain desensitized data, and sending the desensitized data to the user. According to the technical scheme, the data classification and grading efficiency and the desensitization strategy making accuracy can be improved.

Description

Technical Field

[0001] This invention relates to the field of data security technology, and in particular to a data de-identification method and system. Background Technology

[0002] Data security governance is currently facing severe challenges: Classification and grading are inefficient and costly: Traditional methods rely heavily on human experience to define classification rules and grading standards. When faced with massive, heterogeneous, and dynamically changing data, they are inefficient, prone to errors, and difficult to cover all data scenarios.

[0003] Rigid rules and poor adaptability: Predefined static rules are difficult to adapt to business changes, the emergence of new data types, and complex data semantic relationships, resulting in inaccurate or incomplete classification and grading results.

[0004] Desensitization strategies are complex to formulate and have weak correlation: Desensitization rules are usually formulated independently of the classification and grading results, require a lot of manual configuration, make it difficult to ensure that the strategy is accurately matched with the data sensitivity, and the strategy is updated late.

[0005] Therefore, to address the above shortcomings, a data anonymization method and system are needed. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to improve the efficiency of data classification and grading and the accuracy of de-identification strategy formulation in the process of data security governance. In view of the shortcomings of the prior art, this invention provides a data de-identification method and system.

[0007] To address the aforementioned technical problems, this invention provides a data desensitization method, executed by a data desensitization system, comprising: Parse the user's data access request to determine the target data structure that the user wants to access; Based on the target data structure, target classification and grading labels corresponding to the target data structure are determined. The target classification and grading label is one of the classification and grading labels. The classification and grading labels are generated in advance by analyzing the initial business information and system data sources using a large model, and are matched with the corresponding data structure in advance based on preset matching rules. Based on the target classification and grading labels, a target de-identification strategy associated with the target classification and grading labels is determined, wherein the target de-identification strategy is one of the de-identification strategies, and the de-identification strategy is determined in advance by analyzing the classification and grading labels of the corresponding data in the system data source using the large model; Based on the data access request, retrieve the data to be accessed corresponding to the target data structure from the system data source; The data to be accessed is de-identified using the target de-identification strategy to obtain de-identified data, which is then sent to the user.

[0008] The present invention also provides a data anonymization system for performing a data anonymization method as described above, comprising: The target data structure determination module is used to parse the user's data access request and determine the target data structure to be accessed by the user. The target classification and grading label determination module is used to determine the target classification and grading label corresponding to the target data structure based on the target data structure. The target classification and grading label is one of the classification and grading labels. The classification and grading label is generated in advance by analyzing the initial business information and system data source using a large model, and is matched with the corresponding data structure in advance based on preset matching rules. The target desensitization strategy determination module is used to determine the target desensitization strategy associated with the target classification and grading label based on the target classification and grading label, wherein the target desensitization strategy is one of the desensitization strategies, and the desensitization strategy is determined in advance by analyzing the classification and grading labels of the corresponding data in the system data source using the large model; The data to be accessed module is used to obtain the data to be accessed corresponding to the target data structure from the data source based on the data access request; The data desensitization module is used to desensitize the data to be accessed using the target desensitization strategy, obtain desensitized data, and send it to the user.

[0009] The data desensitization method and system of the present invention have the following beneficial effects: By leveraging the semantic understanding and generation capabilities of large-scale models, data classification and grading labels are automatically and intelligently generated, with the entire process possessing dynamic learning and adaptive capabilities. Based on the automatically generated classification and grading labels, the large-scale model intelligently generates matching desensitization strategies, ensuring a high degree of matching between the desensitization strategies and data sensitivity. Sensitive data is desensitized in real time and accurately during data access. The innovative application of large-scale models to the generation of classification and grading labels and desensitization strategies significantly improves the automation and intelligence level of the desensitization process, greatly reduces the workload of manually defining rules and configuring strategies, and lowers operating costs and the risk of human error. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of a data desensitization method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a data desensitization system provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] Figure 1 This is a flowchart illustrating a data anonymization method provided in an embodiment of the present invention. The method is executed by a data anonymization system, such as... Figure 1 As shown, the method includes: S110. Parse the user's data access request and determine the target data structure to be accessed by the user.

[0013] Preferably, in this embodiment, a proxy service layer is deployed in the data anonymization system as an intermediate proxy layer for users accessing the system's data source. This proxy service layer intercepts user data access requests, parses them, identifies the target data structure the user intends to access, finds its corresponding target classification and grading tags and the target anonymization strategy associated with those tags, and anonymizes the data to be accessed by the user using the target anonymization strategy before sending it to the user. By implementing the data anonymization process through a proxy layer, the business coupling is low, the intrusion into the business system is minimal, and it is easy to integrate and deploy.

[0014] The data access request can be an SQL (Structured Query Language) statement. This request can include the data structure corresponding to the data the user intends to access. If the user has specific anonymization requirements for the data, the request can also include a description of those requirements. In this embodiment, the data structure can be a table and / or fields. For example, tables can include user tables, product tables, order tables, and category tables, and fields can include ID card information, salary data, and date data.

[0015] S120. Based on the target data structure, determine the target classification and grading label corresponding to the target data structure, wherein the target classification and grading label is one of the classification and grading labels, and the classification and grading label is generated in advance by analyzing the initial business information and system data source using a large model, and is matched with the corresponding data structure in advance based on preset matching rules.

[0016] In this embodiment, the initial business information can be industry and / or business type information pre-entered by the system creator. The system data source is all data sources existing in the system when the user accesses it; these are pre-created by the system creator according to business needs, and can be created all at once or added multiple times. The preset matching rules can be manually matched by the system creator or matched using a large model, etc.

[0017] S130. Based on the target classification and grading labels, determine the target desensitization strategy associated with the target classification and grading labels, wherein the target desensitization strategy is one of the desensitization strategies, and the desensitization strategy is determined in advance by analyzing the classification and grading labels of the corresponding data in the system data source using the large model.

[0018] In this embodiment, the large model analyzes and understands the sensitivity and business scenarios of the corresponding data based on the classification and grading labels, and determines the appropriate de-identification strategy. If the system creator provides specific de-identification requirements, the large model will generate corresponding de-identification strategies based on the above content and the specific de-identification requirements. It is understandable that different classification and grading labels correspond to different de-identification strategies.

[0019] S140. Based on the data access request, obtain the data to be accessed corresponding to the target data structure from the system data source.

[0020] For example, if the target data structure is an ID card field, then the corresponding data to be accessed is the information of each ID card corresponding to the ID card field.

[0021] S150. The data to be accessed is de-identified using the target de-identification strategy to obtain de-identified data and send it to the user.

[0022] This invention proposes a data anonymization method, comprising: parsing a user's data access request to determine the target data structure to be accessed by the user; determining a target classification and grading label corresponding to the target data structure based on the target data structure, wherein the target classification and grading label is one of the classification and grading labels, which is generated in advance by analyzing initial business information and system data sources using a large model, and is matched with the corresponding data structure in advance based on preset matching rules; determining a target anonymization strategy associated with the target classification and grading label based on the target classification and grading label, wherein the target anonymization strategy is one of the anonymization strategies, which is determined in advance by analyzing the classification and grading labels of the corresponding data in the system data source using the large model; obtaining the data to be accessed corresponding to the target data structure from the system data source based on the data access request; and performing anonymization processing on the data to be accessed using the target anonymization strategy to obtain anonymized data and sending it to the user. By leveraging the semantic understanding and generative capabilities of large-scale models, data classification and grading labels are automatically and intelligently generated, with the entire process possessing dynamic learning and adaptive capabilities. Based on the automatically generated classification and grading labels, the large-scale model intelligently generates matching desensitization strategies, ensuring a high degree of matching between the desensitization strategies and data sensitivity. During data access, the desensitization strategies are used to perform real-time and precise desensitization processing on sensitive data, improving the accuracy of data desensitization. This innovative application of large-scale models to the generation of classification and grading labels and desensitization strategies significantly improves the automation and intelligence level of the desensitization process, greatly reduces the workload of manually defining rules and configuring strategies, and lowers operating costs and the risk of human error.

[0023] Based on the above embodiments, further, before determining the target classification and grading label corresponding to the target data structure based on the target data structure, the method further includes: The initial business information and system data sources are analyzed in advance using a large model to generate the classification and grading labels.

[0024] Preferably, the data desensitization system in this embodiment is configured with a large model-driven dynamic classification and grading engine, which is used to generate classification and grading labels using a large model.

[0025] Preferably, the step of pre-analyzing the initial business information and system data sources using a large model to generate the classification and grading labels includes: Based on the initial business information, the large model is invoked to generate initial classification labels related to the initial business information, and the initial classification labels are updated in the classification label library. The initial business information is pre-input by the system creator. Continuously monitor whether there are any new data sources. If so, sample the data from the new data sources to obtain sampled data. The new data sources are created in advance by the system creator as needed, and the system data sources include each new data source. The large model is used to analyze the sampled data to obtain sampled classification labels, and the sampled classification labels are updated to the classification label library. Based on preset security level classification rules or by using a large model to analyze data sensitivity, the classification tags in the classification tag library are assigned corresponding security levels to obtain the classification and grading tags.

[0026] In this embodiment, the system, based on user-inputted information such as the company's industry and business type, calls the large model interface to generate a series of initial category tags that may be related to the company's industry and business type, and updates them to the category tag library, which stores the category tags. Each time the system creator creates a new data source, the system automatically or on demand samples the data from the new data source and inputs the sampled data into the large model. The large model analyzes the sampled data content, identifies potential new category tags not included in the category tag library, and dynamically updates the category tag library. For example, category tags may include ID card tags, salary tags, and date tags. After obtaining the category tag library, the generated category tags are assigned corresponding security levels, which may include public, internal, secret, and top secret levels. For example, if the category tag is an ID card tag and the corresponding security level is secret, then the corresponding category tag is an ID card secret tag.

[0027] The classification and grading labeling system in this embodiment can be dynamically expanded and adjusted according to new data sources, possessing strong adaptability and evolutionary capabilities.

[0028] Based on the above embodiments, further comprising, after obtaining the classification and grading label, the method includes: Based on a preset regular expression, each data structure in the system data source is matched and labeled with the classification and grading tags, wherein the data structure is a table or field; Alternatively, based on the system creator's description of each data structure, a large model is used to generate matching suggestions between each data structure and the classification and grading labels. After receiving confirmation information from the system creator regarding the matching suggestions, each data structure is matched and labeled with the classification and grading labels based on the matching suggestions. Alternatively, a large model can be used to analyze each data structure and the classification and grading labels, automatically matching and labeling each data structure with the classification and grading labels.

[0029] In this embodiment, after obtaining the classification and grading labels, the labels can be matched and marked with the corresponding data structures based on preset matching rules. The preset matching rules are the three matching rules described above. Using these three matching rules to match the data structures with the classification and grading labels improves the accuracy of the matching. For example, if the data structure is an ID card field, and the classification and grading label matched using the preset matching rules is an ID card secret label, then the ID card secret label is marked on the ID card field.

[0030] Based on the above embodiments, further, before determining the target desensitization strategy associated with the target classification and grading label based on the target classification and grading label, the method further includes: The system uses the large model to analyze the classification and grading labels of the corresponding data in the system's data source in advance to determine the desensitization strategy.

[0031] Preferably, the data desensitization system in this embodiment is equipped with an intelligent desensitization strategy generator, which is used to generate desensitization strategies using a large model.

[0032] Preferably, the step of pre-analyzing the classification and grading labels of the corresponding data in the system data source using the large model to determine the desensitization strategy includes: The classification and grading labels and data types of the corresponding data in the system's data source are input into the large model as input information. Based on the input information, the large model is used to perform business scenario and sensitivity analysis on the corresponding data. Based on the preset desensitization algorithm library, a desensitization strategy corresponding to the corresponding data is generated. The desensitization strategy is associated with the classification and grading labels of the corresponding data.

[0033] In this embodiment, the trigger condition for generating the de-identification strategy can be after the system data has completed the hierarchical classification label matching and marking. For example, the data type can include strings, numbers, and dates. The de-identification algorithm library is configured with various de-identification algorithms, which can include, for example, replacement, masking, hashing, encryption, and generalization algorithms. It is understood that the de-identification strategy can be a single de-identification method or a combination of multiple de-identification methods; no special limitation is made here. For example, if the classification label of the corresponding data is an ID card secret label, and its corresponding data type is a string, the ID card secret label and the string type are input into the large model. The large model performs business scenario and sensitivity analysis on the ID card data based on the ID card secret label and the string type to determine the de-identification strategy for the ID card data.

[0034] The above desensitization strategy is automatically generated based on intelligent classification and grading results, ensuring a high degree of matching between the desensitization strategy and data sensitivity.

[0035] Based on the above embodiments, the input information further includes desensitization requirement description information of the corresponding data, and the generated desensitization strategy is associated with the classification and grading labels of the corresponding data and the desensitization requirement description information.

[0036] In this embodiment, the trigger condition for generating the de-identification strategy can be that the system creator inputs the de-identification requirement description information for the corresponding data. For example, if the classification and grading label for the corresponding data is an ID card secret label, its corresponding data type is a string, and its de-identification requirement description information is hiding the last 6 characters of the ID card, then the ID card secret label, the string type, and the hidden last 6 characters of the ID card are input into the large model. The large model performs business scenario and sensitivity analysis on the ID card data based on the ID card secret label, the string type, and the hidden last 6 characters of the ID card to determine the ID card data de-identification strategy.

[0037] Based on the above embodiments, it further includes: The historical classification and grading operation records are input into the large model, and the classification and grading labels are evaluated using the large model to generate classification and grading label optimization suggestions. Based on the proposed optimization suggestions for the classification and grading labels, the classification and grading labels are optimized. Input the desensitization strategy sample and the corresponding data samples before and after desensitization into the large model, use the large model to evaluate the desensitization strategy sample, and generate desensitization strategy optimization suggestions; Based on the proposed desensitization strategy optimization suggestions, the desensitization strategy samples were optimized.

[0038] In this embodiment, the system records in detail the classification and grading process, the generation of de-identification strategies, user access, and the results of de-identification execution. A large model is used to analyze historical records, automatically performing post-event quality checks (e.g., checking the accuracy of classification and grading and evaluating the effectiveness of de-identification), identifying problems, generating optimization suggestions, and feeding them back to the front-end process (e.g., adjusting classification and grading labels, optimizing matching rules, and updating de-identification strategies), forming a closed loop of continuous improvement.

[0039] The system records key operation logs in detail, including: the initial generation of classification labels during the classification and grading process, the dynamic update record of classification labels, the classification and grading results, and the matching mark operation (which may include operation mode, operator, and result, etc.); the triggering conditions, input information, and generated desensitization strategy content during the desensitization strategy generation process; the data access requests, user information, access time, and target data structure during the user access process; and the desensitization strategy, data samples before and after desensitization, and execution result status during the desensitization execution process.

[0040] For example, historical classification and grading operation records may include sampling data and data structure matching results. The large model is used to evaluate the classification and grading labels, including the rationality of the classification label definition, the accuracy of data structure matching, and the appropriateness of the grading. Optimization suggestions for classification and grading labels may include merging classification labels, adjusting grading, and modifying matching rules.

[0041] This embodiment inputs a sample of the de-identification strategy and data samples before and after de-identification into a large model to evaluate the effectiveness of the de-identification strategy, balancing data availability and security, and identifying potential information leakage risks or over-identification issues. For example, de-identification strategy optimization suggestions may include adjusting the de-identification algorithm strength and modifying the de-identification strategy itself.

[0042] Provide quality inspection reports, suggestions for optimizing classification and grading labels, and suggestions for optimizing desensitization strategies to the administrator or directly feed them back to the system's front-end workflow. Administrators can then adjust the classification and grading label library, optimize matching rules, modify grading strategies, or update desensitization strategies accordingly. The system can also automatically apply these suggestions based on preset rules, achieving closed-loop continuous optimization.

[0043] In this embodiment, the system records the entire operation process and uses a large model for post-operation quality inspection, realizing a fully intelligent closed loop from data identification (classification and grading), strategy formulation (de-identification rule generation) to strategy execution (real-time de-identification) and effect evaluation (quality inspection and optimization), continuously optimizing the classification and grading model and de-identification strategy.

[0044] Figure 2 This is a schematic diagram of a data anonymization system provided in an embodiment of the present invention. This system is used to execute a data anonymization method as described in any of the above embodiments, such as... Figure 2 As shown, the system includes: The target data structure determination module 210 is used to parse the user's data access request and determine the target data structure to be accessed by the user. The target classification and grading label determination module 220 is used to determine the target classification and grading label corresponding to the target data structure based on the target data structure. The target classification and grading label is one of the classification and grading labels. The classification and grading label is generated in advance by analyzing the initial business information and system data source using a large model, and is matched with the corresponding data structure in advance based on preset matching rules. The target desensitization strategy determination module 230 is used to determine the target desensitization strategy associated with the target classification and grading label based on the target classification and grading label, wherein the target desensitization strategy is one of the desensitization strategies, and the desensitization strategy is determined in advance by analyzing the classification and grading labels of the corresponding data in the system data source using the large model; The data to be accessed acquisition module 240 is used to acquire the data to be accessed corresponding to the target data structure from the data source based on the data access request; The data desensitization module 250 is used to desensitize the data to be accessed using the target desensitization strategy, obtain desensitized data, and send it to the user.

[0045] This invention proposes a data anonymization system for executing a data anonymization method as described in any of the above embodiments, comprising: a target data structure determination module, used to parse a user's data access request and determine the target data structure to be accessed by the user; a target classification and grading label determination module, used to determine a target classification and grading label corresponding to the target data structure based on the target data structure, wherein the target classification and grading label is one of the classification and grading labels, which is generated in advance by analyzing initial business information and system data sources using a large model, and is matched with the corresponding data structure in advance based on preset matching rules; a target anonymization strategy determination module, used to determine a target anonymization strategy associated with the target classification and grading label based on the target classification and grading label, wherein the target anonymization strategy is one of the anonymization strategies, which is determined in advance by analyzing and determining the classification and grading labels of the corresponding data in the system data source using the large model; a data to be accessed acquisition module, used to acquire data to be accessed corresponding to the target data structure from the data source based on the data access request; and a data anonymization module, used to perform anonymization processing on the data to be accessed using the target anonymization strategy to obtain anonymized data and send it to the user. By leveraging the semantic understanding and generative capabilities of large-scale models, data classification and grading labels are automatically and intelligently generated, with the entire process possessing dynamic learning and adaptive capabilities. Based on the automatically generated classification and grading labels, the large-scale model intelligently generates matching desensitization strategies, ensuring a high degree of matching between the desensitization strategies and data sensitivity. During data access, the desensitization strategies are used to perform real-time and precise desensitization processing on sensitive data, improving the accuracy of data desensitization. This innovative application of large-scale models to the generation of classification and grading labels and desensitization strategies significantly improves the automation and intelligence level of the desensitization process, greatly reduces the workload of manually defining rules and configuring strategies, and lowers operating costs and the risk of human error.

[0046] Based on the above technical solutions, the system may further include a classification and grading label determination module, which can be used to: analyze the initial business information and system data source in advance using a large model before determining the target classification and grading label corresponding to the target data structure based on the target data structure, and generate the classification and grading label.

[0047] Based on the above technical solutions, the classification and grading label determination module can further be used to: generate initial classification labels related to the initial business information by calling the large model, and update the initial classification labels to the classification label library, wherein the initial business information is pre-input by the system creator; continuously monitor whether there are new data sources, and if so, sample the data from the new data sources to obtain sampled data, wherein the new data sources are pre-created by the system creator as needed, and the system data sources include each new data source; analyze the sampled data using the large model to obtain sampled classification labels, and update the sampled classification labels to the classification label library; assign corresponding security levels to the classification labels in the classification label library based on preset security level classification rules or by analyzing data sensitivity using the large model, thereby obtaining the classification and grading labels.

[0048] Based on the above technical solutions, the classification and grading label determination module can further be used to: after obtaining the classification and grading labels, match and label each data structure in the system data source with the classification and grading labels based on a preset regular expression, wherein the data structure is a table or field; or, based on the system creator's description information of each data structure, use a large model to generate matching suggestions between each data structure and the classification and grading labels, and after receiving confirmation information from the system creator regarding the matching suggestions, match and label each data structure with the classification and grading labels based on the matching suggestions; or, use a large model to analyze each data structure and the classification and grading labels, and automatically match and label each data structure with the classification and grading labels.

[0049] Based on the above technical solutions, the system may further include a desensitization strategy generation module, which can be used to: analyze the classification and grading labels of the corresponding data in the system data source in advance using the large model before determining the target desensitization strategy associated with the target classification and grading labels based on the target classification and grading labels, and determine the desensitization strategy.

[0050] Based on the above technical solutions, the desensitization strategy generation module can further be used to: input the classification and grading labels and data types of the corresponding data in the system data source into the large model as input information; based on the input information, use the large model to perform business scenario and sensitivity analysis on the corresponding data, and generate a desensitization strategy corresponding to the corresponding data based on a preset desensitization algorithm library, wherein the desensitization strategy is associated with the classification and grading labels of the corresponding data.

[0051] Based on the above technical solutions, the input information further includes desensitization requirement description information of the corresponding data, and the generated desensitization strategy is associated with the classification and grading labels of the corresponding data and the desensitization requirement description information.

[0052] Based on the above technical solutions, the system may further include an audit and closed-loop optimization module, specifically used for: inputting historical classification and grading operation records into the large model; using the large model to evaluate the classification and grading labels and generate classification and grading label optimization suggestions; optimizing the classification and grading labels according to the classification and grading label optimization suggestions; inputting de-identification strategy samples and corresponding pre- and post-de-identification data samples into the large model; using the large model to evaluate the de-identification strategy samples and generate de-identification strategy optimization suggestions; and optimizing the de-identification strategy samples according to the de-identification strategy optimization suggestions.

[0053] The data desensitization system provided in this embodiment of the invention can execute a data desensitization method provided in this embodiment of the invention, and its execution steps and beneficial effects will not be described in detail here.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data anonymization method, characterized in that, Performed by the data anonymization system, including: Parse the user's data access request to determine the target data structure that the user wants to access; Based on the target data structure, target classification and grading labels corresponding to the target data structure are determined. The target classification and grading label is one of the classification and grading labels. The classification and grading labels are generated in advance by analyzing the initial business information and system data sources using a large model, and are matched with the corresponding data structure in advance based on preset matching rules. Based on the target classification and grading labels, a target de-identification strategy associated with the target classification and grading labels is determined, wherein the target de-identification strategy is one of the de-identification strategies, and the de-identification strategy is determined in advance by analyzing the classification and grading labels of the corresponding data in the system data source using the large model; Based on the data access request, retrieve the data to be accessed corresponding to the target data structure from the system data source; The data to be accessed is de-identified using the target de-identification strategy to obtain de-identified data, which is then sent to the user.

2. The method according to claim 1, characterized in that, Before determining the target classification and grading label corresponding to the target data structure based on the target data structure, the method further includes: The initial business information and system data sources are analyzed in advance using a large model to generate the classification and grading labels.

3. The method according to claim 2, characterized in that, The process of pre-analyzing initial business information and system data sources using a large model to generate the classification and grading labels includes: Based on the initial business information, the large model is invoked to generate initial classification labels related to the initial business information, and the initial classification labels are updated in the classification label library. The initial business information is pre-input by the system creator. Continuously monitor whether there are any new data sources. If so, sample the data from the new data sources to obtain sampled data. The new data sources are created in advance by the system creator as needed, and the system data sources include each new data source. The large model is used to analyze the sampled data to obtain sampled classification labels, and the sampled classification labels are updated to the classification label library. Based on preset security level classification rules or by using a large model to analyze data sensitivity, the classification tags in the classification tag library are assigned corresponding security levels to obtain the classification and grading tags.

4. The method according to claim 3, characterized in that, After obtaining the classification and grading labels, the process also includes: Based on a preset regular expression, each data structure in the system data source is matched and labeled with the classification and grading tags, wherein the data structure is a table or field; Alternatively, based on the system creator's description of each data structure, a large model is used to generate matching suggestions between each data structure and the classification and grading labels. After receiving confirmation information from the system creator regarding the matching suggestions, each data structure is matched and labeled with the classification and grading labels based on the matching suggestions. Alternatively, a large model can be used to analyze each data structure and the classification and grading labels, automatically matching and labeling each data structure with the classification and grading labels.

5. The method according to any one of claims 1-4, characterized in that, Before determining the target desensitization strategy associated with the target classification and grading label based on the target classification and grading label, the method further includes: The system uses the large model to analyze the classification and grading labels of the corresponding data in the system's data source in advance to determine the desensitization strategy.

6. The method according to claim 5, characterized in that, The step of pre-analyzing the classification and grading labels of the corresponding data in the system's data source using the large model to determine the de-identification strategy includes: The classification and grading labels and data types of the corresponding data in the system's data source are input into the large model as input information. Based on the input information, the large model is used to perform business scenario and sensitivity analysis on the corresponding data. Based on the preset desensitization algorithm library, a desensitization strategy corresponding to the corresponding data is generated. The desensitization strategy is associated with the classification and grading labels of the corresponding data.

7. The method according to claim 5, characterized in that, The input information also includes descriptions of the data anonymization requirements, and the generated anonymization strategies are associated with the data classification and grading labels and the anonymization requirements descriptions.

8. The method according to any one of claims 1-4, characterized in that, Also includes: The historical classification and grading operation records are input into the large model, and the classification and grading labels are evaluated using the large model to generate classification and grading label optimization suggestions. Based on the proposed optimization suggestions for the classification and grading labels, the classification and grading labels are optimized. Input the desensitization strategy sample and the corresponding data samples before and after desensitization into the large model, use the large model to evaluate the desensitization strategy sample, and generate desensitization strategy optimization suggestions; Based on the proposed desensitization strategy optimization suggestions, the desensitization strategy samples were optimized.

9. A data anonymization system, characterized in that, A data anonymization method for performing any one of claims 1-8 includes: The target data structure determination module is used to parse the user's data access request and determine the target data structure to be accessed by the user. The target classification and grading label determination module is used to determine the target classification and grading label corresponding to the target data structure based on the target data structure. The target classification and grading label is one of the classification and grading labels. The classification and grading label is generated in advance by analyzing the initial business information and system data source using a large model, and is matched with the corresponding data structure in advance based on preset matching rules. The target desensitization strategy determination module is used to determine the target desensitization strategy associated with the target classification and grading label based on the target classification and grading label, wherein the target desensitization strategy is one of the desensitization strategies, and the desensitization strategy is determined in advance by analyzing the classification and grading labels of the corresponding data in the system data source using the large model; The data to be accessed module is used to obtain the data to be accessed corresponding to the target data structure from the data source based on the data access request; The data desensitization module is used to desensitize the data to be accessed using the target desensitization strategy, obtain desensitized data, and send it to the user.

10. The system according to claim 9, characterized in that, The system also includes a classification and grading label determination module, which is used to analyze the initial business information and system data source in advance using a large model to generate the classification and grading labels before determining the target classification and grading labels corresponding to the target data structure based on the target data structure.

Images