Privacy field processing method and device, equipment, medium and program product

Abstract

The embodiment of the invention provides a privacy field processing method and device, equipment, a medium and a program product, and relates to the field of artificial intelligence. The method comprises the steps of obtaining metadata and access log information of fields in a database, and extracting multi-dimensional features of the fields through the metadata and the access log information; inputting the multi-dimensional features into a semantic recognition model for feature fusion analysis, and outputting a privacy type classification result of the field; and generating a privacy protection strategy of the field according to a privacy type classification result. Based on the method, full-process automation of intelligent identification, dynamic grading and strategy configuration of the privacy fields can be realized, and the privacy field processing efficiency and the privacy protection adaptation degree are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a method, apparatus, device, medium, and program product for processing privacy fields. Background Technology

[0002] Taking banking as an example in the financial sector, with the acceleration of digital transformation in banking operations, data assets are a crucial banking resource. Bank databases store a large amount of sensitive data, including user personal information, transaction records, and account behavior. To protect data security, privacy protection measures are necessary for this sensitive data.

[0003] Privacy fields can be understood as data that requires privacy protection. Existing methods for handling privacy fields primarily rely on manual rules such as field name matching and regular expression recognition. These existing methods are ill-suited to the demands of efficient privacy field protection.

[0004] For example, when dealing with large volumes of data for privacy field identification and protection, manual rules struggle to accurately identify privacy fields and provide differentiated privacy protection for fields with different privacy levels. This inefficiency becomes even more pronounced when privacy fields in the database undergo frequent changes. Therefore, existing methods suffer from limitations in efficiently providing highly adaptable privacy protection for privacy fields. Summary of the Invention

[0005] This application provides a method, apparatus, device, medium, and program product for processing privacy fields, which aims to achieve highly efficient and adaptive privacy protection when processing privacy fields.

[0006] In a first aspect, embodiments of this application provide a method for processing privacy fields, the method comprising:

[0007] Obtain metadata and access log information of fields in the database, and extract multidimensional features of fields using the metadata and access log information;

[0008] Multidimensional features are input into a semantic recognition model for feature fusion analysis, and the privacy type classification results of the output fields are output.

[0009] Generate privacy protection strategies for fields based on privacy type classification results.

[0010] Secondly, embodiments of this application provide a privacy field processing apparatus, the apparatus comprising:

[0011] The acquisition module is used to acquire metadata and access log information of fields in the database, and extract multidimensional features of the fields through the metadata and access log information;

[0012] The output module is used to input multidimensional features into the semantic recognition model for feature fusion analysis and output the privacy type classification results of the fields.

[0013] The generation module is used to generate privacy protection strategies for fields based on the privacy type classification results.

[0014] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0016] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0017] The privacy field processing method, apparatus, device, medium, and program products provided in this application embodiment address the limitation of existing static rule bases that rely solely on field name matching by using a semantic recognition model to perform feature fusion analysis on the multi-dimensional features of fields. Feature fusion analysis weights and fuses the field's metadata with access log information, optimizing the accuracy of privacy field identification. Based on this method, the ability to identify implicit privacy fields can be significantly improved. Furthermore, the privacy protection strategy generated from the classification results reduces manual intervention and can specifically generate privacy protection strategies adapted to privacy fields, adapting to business scenarios where privacy fields frequently change. Therefore, this method can automatically generate privacy protection strategies for privacy fields, achieving full automation of intelligent identification, dynamic classification, and strategy configuration of privacy fields, improving the efficiency of privacy field processing and the adaptability of privacy protection. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] Figure 1 A flowchart illustrating the privacy field processing method provided in this application embodiment;

[0020] Figure 2 This is a schematic diagram of the structure of the privacy field processing system provided in the embodiments of this application;

[0021] Figure 3 This is a schematic diagram of the privacy field processing device provided in the embodiments of this application;

[0022] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0023] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0025] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0026] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0027] The following is an explanation of some terms and concepts used in the embodiments of this application:

[0028] Privacy Field: This refers to information fields in a database that can directly or indirectly identify an individual, such as ID card number, mobile phone number, bank card number, email address, etc.

[0029] Sensitivity Leveling: Based on the degree of risk posed by information leakage, privacy fields are divided into different security levels to guide privacy protection strategies of varying strengths.

[0030] Semantic Recognition Model: An artificial intelligence (AI) model based on natural language processing technology, used to understand field names and contextual semantics and perform classification.

[0031] Data Masking Policy: A security strategy that preserves the format of sensitive data or partially hides it, used in development testing or open interface scenarios.

[0032] Access Log Feature: By analyzing behavioral characteristics such as database query frequency, access roles, and call parameters, this feature assists artificial intelligence models in identifying potential privacy fields.

[0033] In this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0034] In the embodiments of this application, the use of terms such as "first" and "second" is to distinguish between identical or similar items that have essentially the same function and effect. For example, "first electronic device" and "second electronic device" are merely used to distinguish different electronic devices and do not limit their order of execution. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply that they are different.

[0035] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0036] For example, with the comprehensive digital transformation of banking operations, core banking systems, data platforms, and open banking platforms have accumulated massive amounts of sensitive customer data, including personal identification information (such as ID card numbers and mobile phone numbers), account information (such as bank card numbers and transaction details), and behavioral data (such as login records and consumption preferences). While this data supports business innovation (such as intelligent risk control and precision marketing), it also faces serious privacy risks.

[0037] For example, open banking interfaces may lead to sensitive data being accessed by third parties without authorization due to incorrect field identification, and shared datasets in data platforms may cause compliance issues due to insufficient classification. Traditional privacy protection solutions rely on manual rule bases (such as field name matching, regular expressions, and data type validation) for privacy field identification and classification. However, bank databases have complex field naming rules (such as field names with implicit privacy, such as contact1 and info_001), variable data structures (such as nested tables and unstructured fields), and need to be dynamically adapted to different business scenarios (such as partial desensitization in development and testing environments and encryption in production environments).

[0038] Furthermore, banks must meet stringent requirements for full lifecycle management of data compliance, including automatic identification of privacy fields, sensitivity classification, configuration of de-identification strategies, and operational audit traceability. Existing methods suffer from low efficiency, poor accuracy, and high compliance risks when dealing with these scenarios. There is an urgent need for an automated solution that combines semantic analysis and a dynamic strategy engine to achieve accurate identification of privacy fields, fine-grained classification, and adaptive strategy configuration, thereby improving business agility while ensuring data security.

[0039] For some existing technologies, the identification and classification of privacy fields mainly rely on static rule bases (such as field name keyword matching, regular expressions, and data type validation). For example, privacy fields are identified by predefined field name lists (such as phone, id_card) or regular expressions (such as “^[1-9]\d{5}\d{4}$” matching ID card numbers); sensitivity is determined by field type (such as “VARCHAR(18)”) or data distribution characteristics (such as uniqueness, character set). The typical process of this type of solution includes: manual maintenance of the rule base, field metadata collection, rule matching, sensitivity label allocation, and strategy configuration. However, existing methods have many limitations based on the above processing procedures.

[0040] For example, manual maintenance is costly. Since the rule base requires manual setting of sensitivity for each field, and rules need continuous updates when business systems frequently change, it is difficult to adapt to dynamic environments. Another example is lower recognition accuracy. Static rules cannot recognize semantically implicit privacy fields (such as mobile_no, contact1) and have insufficient ability to recognize unstructured data, such as JavaScript Object Notation (JSON) fields, resulting in lower recognition accuracy.

[0041] Existing methods suffer from coarse-grained grading. Relying solely on field types or simple rules, they cannot differentiate between fields of varying risk levels, such as ID numbers (highly sensitive) and customer tags (lowly sensitive). Furthermore, existing methods lack intelligent auditing. The inability to record the logical path between field identification and policy decisions makes compliance reviews difficult. Moreover, the strategies employed by existing methods are easily circumvented. Because they only anonymize fields, attackers can bypass protection through function concatenation or multi-field combination queries.

[0042] In summary, existing technologies rely on static rule bases for privacy field identification. However, bank databases have complex field naming conventions and variable data structures, such as nested tables and JSON fields, which may lead to insufficient identification accuracy. Furthermore, tiered rules based on field type or simple rules cannot differentiate between different risk levels, resulting in a mismatch between privacy protection strategies and risks. In addition, policy configuration relies on manual maintenance, making it difficult to adapt to frequent system changes, and it lacks the ability to protect against attackers bypassing de-identification through function concatenation.

[0043] Starting from the actual needs of banking operations, the inventors discovered that existing static rule bases cannot handle the technical challenges of complex field naming and dynamic business scenarios. To address this, they proposed using semantic recognition models (such as a converter-based bidirectional encoder representation model) to analyze field names and contextual annotations, combining this with sample data features (such as character sets and uniqueness) to improve recognition accuracy. Simultaneously, by integrating access log behaviors (such as query frequency and calling roles) through a feature fusion module, they further optimized the identification and classification of privacy fields.

[0044] Building upon this foundation, an adaptive policy engine can automatically generate de-identification and encryption policies based on the hierarchical results, and transactional repair controls ensure policy execution consistency. Finally, an audit module records the semantic recognition model's reasoning path, achieving interpretability and compliance.

[0045] In view of this, embodiments of this application provide a method for processing privacy fields. This method combines a semantic recognition model with an adaptive policy engine to achieve automated identification, dynamic classification, and policy configuration of privacy fields in a database. Based on a semantic recognition model, this approach analyzes the semantic features, contextual information, and access behavior data of fields to overcome the limitations of existing static rule bases and solve the problems of low accuracy in privacy field identification, coarse-grained classification, and reliance on manual policy configuration. Simultaneously, by interfacing with the database system through a dynamic policy engine, real-time deployment and transactional remediation control of privacy protection policies are achieved, ensuring a balance between data security and business efficiency.

[0046] The method provided in this application can be applied to scenarios such as core banking systems, data platforms, and open banking platforms. In core banking systems, sensitive fields such as customer identity information and transaction records need to be automatically identified and classified; in data platforms, fine-grained privacy protection is required for shared datasets; and in open banking interfaces, the privacy policies of different third-party callers need to be dynamically adapted. Furthermore, core banking systems can be integrated with database policy centers to support real-time policy deployment and audit traceability.

[0047] It should be noted that the privacy field processing methods, devices, equipment, media, and program products provided in this application can be used in the field of artificial intelligence technology, or in any field other than artificial intelligence, such as information security, financial technology, or other related fields. The application fields of the privacy field processing methods, devices, equipment, media, and program products in this application are not limited.

[0048] The technical solutions of this application will be described in detail below with reference to specific embodiments. The specific embodiments described below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0049] Figure 1 This is a flowchart illustrating the privacy field processing method provided in an embodiment of this application. The executing entity of this method can be an electronic device with corresponding data storage and computing capabilities, such as a computer, server, or server cluster. Figure 1 As shown, the method includes:

[0050] S101: Obtain the metadata and access log information of the fields in the database, and extract the multidimensional features of the fields through the metadata and access log information.

[0051] For example, a database field can be understood as a data unit stored in the database to represent a specific business attribute (such as mobile_no, transaction_detail). Metadata can be understood as data that describes data, including field names, data types, comments, table names, etc. (such as the metadata corresponding to the field mobile_no is of type VARCHAR and the comment is "user contact information").

[0052] Access log information can be understood as data that records database access behavior, including query frequency, calling role, interface type, etc., such as the mobile_no field being frequently accessed by the risk control system. Multidimensional features can be understood as feature vectors that include metadata, access log information, and statistical features of sample data (such as the multidimensional features of the mobile_no field including field name, VARCHAR type, and frequently queried role).

[0053] S102, input the multi-dimensional features into the semantic recognition model for feature fusion analysis, and output the privacy type classification results of the fields.

[0054] For example, a semantic recognition model can be understood as a machine learning model based on natural language processing, such as a bidirectional encoder representation model based on a transducer. Semantic recognition models can be used to analyze the semantics and contextual information of fields. Feature fusion analysis can be understood as weighting and combining multi-dimensional features and inputting them into a classification network to generate privacy-type classification results (e.g., the field mobile_no is classified as contact information).

[0055] S103, Generate privacy protection strategies for fields based on privacy type classification results.

[0056] For example, the privacy type classification result can be understood as the privacy type label of the field output by the semantic recognition model (such as ID number, contact information). The privacy protection strategy can be understood as the strategy to protect the privacy field, such as de-identification, encryption or access control strategy (such as ID number field matching encryption strategy) generated based on the privacy type classification result.

[0057] For example, to implement the above steps, after obtaining the metadata and access log information of the database fields, multi-dimensional features can be extracted from the database field metadata and access log information, and used as input to the semantic recognition model. The semantic recognition model combines field semantics (such as the field name mobile_no and the annotation user contact information) and sample data statistical features (such as character set and uniqueness), and performs weighted analysis through a feature fusion layer to output privacy type classification results (such as contact information and ID number).

[0058] Furthermore, the classification results can be input into the policy generation module, which then matches them with predefined de-identification, encryption, or access control policy templates to generate privacy protection policies (such as matching the ID number field with a symmetric encryption policy). The overall process formed by the above steps, through dynamic analysis of multi-dimensional feature fusion and semantic recognition models, achieves automatic identification and classification of privacy fields, and connects with the database policy center to complete policy deployment.

[0059] As can be seen, existing methods rely on static rule bases for privacy field identification and classification, resulting in high manual maintenance costs, low identification accuracy, and coarse classification granularity. In contrast, the method in this application employs a semantic recognition model, combining field names, annotations, statistical features of sample data, and access log behavior features. Through a feature fusion layer, it performs multi-dimensional feature analysis, outputting privacy type labels and sensitivity classification results.

[0060] The privacy field processing method provided in this application addresses the limitation of existing static rule bases that rely solely on field name matching by using a semantic recognition model to perform feature fusion analysis on the multidimensional features of fields. For example, the field `mobile_no` is identified as contact information through a semantic model, rather than relying on the keyword "mobile" in a predefined rule base. The feature fusion layer weightedly fuses statistical features of sample data (such as character set distribution and uniqueness) with access log behavior (such as frequently queried roles) to further optimize recognition accuracy. Based on this method, the ability to identify implicit privacy fields (such as `contact1` and `info_001`) can be significantly improved. Simultaneously, the method generates privacy protection strategies (such as ID number matching encryption strategies and customer tag matching partial desensitization strategies) that match the risk level based on the classification results, reducing manual intervention and adapting to business scenarios with frequent system changes. Ultimately, this method can automatically generate desensitization or encryption strategies that match the risk level, achieving full automation of the intelligent identification, dynamic classification, and strategy configuration of privacy fields.

[0061] In one possible implementation, multidimensional features of a field are extracted through metadata and access log information, including: extracting multidimensional features of a field through field name, data type, table name and / or field comments in the field's metadata, query frequency, calling role and / or interface type in the field's access log information, and sample data statistical features of the field, wherein the sample data statistical features of the field include at least one of character set distribution, uniqueness and data length.

[0062] For example, a field name can be understood as a unique name that identifies the field (e.g., mobile_no). A data type can be understood as the data storage format of the field (e.g., VARCHAR, INT). A table name can be understood as the name of the database table to which the field belongs (e.g., user_profile). A field comment can be understood as descriptive text about the field (e.g., text explaining that the field contains user contact information).

[0063] Query frequency can be understood as the frequency with which a field is accessed (e.g., mobile_no is accessed 100 times / day). Calling role can be understood as the system or user role that initiates the access request for the field (e.g., a bank's risk control system, data analysis system, etc.). Interface type can be understood as the type of interface that initiates the access request for the field (e.g., open banking interface, internal application programming interface (API), etc.).

[0064] In fields such as machine learning and data analysis, statistical characteristics of sample data represent indicators that quantitatively describe the distribution, central tendency, dispersion, morphological characteristics, and relationships between variables of sample data. In this embodiment, the statistical characteristics of the field's sample data include at least one of character set distribution, uniqueness, and data length.

[0065] Character set distribution can be understood as the distribution of character types in the field data (e.g., the character set of the mobile_no field is a set of numbers). Uniqueness can be understood as the probability of duplicate data in the field data (e.g., the uniqueness of the ID card number field is 100%). Data length can be understood as the range of character lengths in the field data (e.g., the length of the mobile_no field is 11).

[0066] In the process of extracting multidimensional features, the first step is to obtain metadata (such as field names, data types, table names, and comments) from the database fields to describe their basic attributes. The second step is to analyze access log information (such as query frequency, calling roles, and interface types) to characterize the field's usage scenarios. Finally, the character set distribution, uniqueness, and data length of the sample data are statistically analyzed to determine the field's data characteristics. Through the extraction of these three types of features, a multidimensional feature vector for the field is formed, which serves as input to the semantic recognition model, providing a data foundation for subsequent privacy type classification.

[0067] In this embodiment, by refining the extraction process of multi-dimensional features, the semantic recognition model is ensured to comprehensively capture the privacy risk attributes of fields. For example, the field `mobile_no` is significantly improved in its ability to identify hidden privacy fields through the fusion analysis of metadata (field name, VARCHAR type), access logs (high-frequency queries initiated by the risk control system), and statistical features of sample data (character set is a numeric set). This technique provides richer input information to the semantic recognition model through the structured extraction of multi-dimensional features, thereby enhancing the accuracy and robustness of privacy field identification.

[0068] Building upon the methods provided in the above embodiments, a context-aware multimodal feature fusion algorithm can also be introduced. For example, field context awareness can be introduced into the semantic recognition module. By fusing multi-dimensional semantic information such as field names, field comments, table names, and / or associated fields, the ability to identify hidden privacy fields can be enhanced. For instance, if the field contact1 is located in the table user_profile and is associated with the phone field, it can be identified as contact information.

[0069] Specifically, the semantic recognition module can employ a multimodal transformer (Transformer) module to achieve context awareness. This module encodes field names, field comments, table names, and related field text information into vectors, and then concatenates these vectors with statistical features of the sample data (such as character sets and uniqueness). The Transformer module's attention mechanism dynamically weights the importance of different contextual information. For example, the field "mobile" might be identified as a transaction device in the table "transaction_log," while in the table "user_info," it might be identified as contact information.

[0070] Multimodal feature fusion significantly improves the accuracy of identifying semantically hidden privacy fields, especially in scenarios with ambiguous or vague field names (such as contact1 and info_001), reducing false positives and false negatives. Simultaneously, a context-aware mechanism enhances the model's understanding of the actual purpose of the fields, avoiding classification errors caused by duplicate field names.

[0071] For example, the lack of interpretable documentation of the semantic recognition model's reasoning process may prevent the privacy field identification and policy configuration processes from meeting regulatory compliance requirements. Therefore, the method provided in this application can introduce a full-process log tracking mechanism in the auditing and interpretability module to record the semantic recognition path and feature fusion weight allocation of the semantic recognition model, providing a decision-making basis for sensitivity classification (such as matching ID card numbers to Level 1 sensitive rules).

[0072] In one possible implementation, after generating the privacy protection strategy for the field based on the privacy type classification result, the method further includes: recording the inference path of the semantic recognition model during the feature fusion analysis process, and the feature fusion weights assigned to each feature during feature fusion; and generating an interpretability log of the privacy type classification result of the field based on the inference path and the feature fusion weights.

[0073] For example, the inference path can be understood as the decision-making process of a semantic recognition model to classify a field into privacy types, such as the field `mobile_no` being parsed by the semantic recognition model as contact information. Feature fusion weights can be understood as the contribution of multi-dimensional features in the classification process, such as a field where statistical features from sample data account for 60% and access log features account for 40%. Interpretable logs can be understood as structured data recording the basis for classification decisions, such as logs in JSON format.

[0074] After generating the privacy protection strategy, the inference path and feature fusion weight allocation of the semantic recognition model can be recorded, and an interpretable log can be generated. This interpretable log stores the basis for classification decisions in a structured format, providing data support for subsequent regulatory audits. Through full-process log tracking, a structured audit report (such as in JSON format) can be generated, detailing the semantic reasoning process of field classification, the allocation logic of feature fusion weights, and the rule matching basis for sensitivity grading.

[0075] For example, the field `transaction_detail` is classified as a secondary sensitive field because access logs show it is frequently used in the risk control system, and is matched with a partial anonymization strategy. This makes all identification and strategy decision-making processes traceable and auditable, meeting the banking industry's compliance requirements for data processing transparency, while also supporting regulatory agencies' compliance reviews of the entire privacy protection process.

[0076] In this embodiment of the application, by recording the reasoning path of the semantic recognition model in the feature fusion analysis process, as well as the feature fusion weights assigned to each feature during feature fusion, and generating an interpretability log of the privacy type classification results of the field, the entire process of privacy field identification and policy configuration is made traceable.

[0077] For example, existing de-identification strategies are easily bypassed by attackers through function concatenation or multi-field combination queries, leading to the risk of sensitive data leakage. To address this, the method provided in this application can introduce an attack pattern recognition module based on Graph Neural Network (GNN) during the strategy execution phase. This module models the query statements and parameter call relationships in the access logs as a graph structure and identifies high-risk attack patterns (such as the combination of "de-identified fields + concatenated functions + fixed parameters") through graph embedding technology.

[0078] In one possible implementation, after generating the privacy protection strategy for the field based on the privacy type classification result, the method further includes: obtaining access logs of access to the database; analyzing the access behavior to the database based on the access logs and identifying high-risk attack patterns; and adjusting the privacy protection strategy for the field based on the identification result of the high-risk attack patterns.

[0079] For example, a query statement can be understood as a statement that requests access to the database, such as a Structured Query Language (SQL) statement, like the SQL statement to query a mobile phone number: SELECT * FROM user_profile WHERE masked_mobile = 1234***5678.

[0080] Parameter call relationships can be understood as the logical association between query parameters (such as concatenating the masked_mobile field with @example.com). High-risk attack patterns can be understood as behavioral patterns that may bypass de-identification strategies, such as abnormal behaviors like function concatenation and multi-field combined queries.

[0081] After generating the privacy protection policy, during the policy execution phase, for example, by introducing an attack pattern recognition module, the module can analyze query statements, parameter call relationships, and construct complete phone numbers in the access logs to identify high-risk attack patterns, detect abnormal behavior, and determine whether there are potential attack behaviors that bypass the de-identification policy.

[0082] The attack pattern recognition module is a functional module for anomaly detection based on graph neural networks. Based on graph neural networks, it analyzes the graph structure of nodes (fields, functions) and edges (call relationships) to capture complex attack patterns. Furthermore, it can mark identified high-risk attack patterns as high-risk behaviors and, in conjunction with a rule engine, automatically trigger encryption or blocking policies for high-risk queries.

[0083] For example, when the system detects that a user has constructed a complete mobile number using the CONCAT(masked_mobile, "@example.com") command, the attack pattern identification module automatically triggers an encryption policy and blocks the query. Based on the identification results, privacy protection policies can be dynamically adjusted, such as upgrading some de-identification policies to hash encryption policies. Therefore, by analyzing access behavior, the system's defense capabilities against bypass attacks are enhanced.

[0084] In this embodiment, by analyzing access behavior through access logs, such as analyzing query statements and parameter call relationships, high-risk attack patterns (such as function concatenation) can be identified. Furthermore, privacy protection strategies can be dynamically adjusted (such as upgrading to hash encryption). Based on this, the detection coverage of attacks that bypass de-identification strategies can be improved, preventing attackers from obtaining sensitive data through function concatenation or multi-field combination queries, thus enhancing protection. In addition, graph embedding technology can dynamically adapt to new attack patterns, enhancing the system's anti-attack capabilities.

[0085] For example, to address the issue that semantic recognition models may struggle to adapt to frequent changes in field naming rules and the introduction of new field types in banking systems, the method provided in this application can incorporate incremental learning and model drift compensation mechanisms. Specifically, an incremental learning framework is introduced into the semantic recognition model. By continuously receiving feedback from new field naming rules, sample data, and access logs, an online fine-tuning algorithm is used to dynamically update model parameters, preventing model performance from declining due to changes in the business system. Furthermore, a model drift detection module can be used to determine whether incremental learning needs to be triggered.

[0086] In one possible implementation, before inputting multidimensional features into the semantic recognition model for feature fusion analysis, the method further includes: receiving new field naming rules and sample data feedback, wherein the sample data feedback is the user's feedback result on the privacy type classification of the field; and dynamically updating the model parameters of the semantic recognition model through an incremental learning algorithm based on the new field naming rules and sample data feedback.

[0087] For example, incremental learning algorithms can be understood as online incremental learning mechanisms that support online updates of model parameters, such as online incremental learning based on the Follow The Regularized Leader (FTRL) optimizer. After the semantic recognition model is deployed, it continuously receives new field naming rules and sample data feedback (such as field metadata and user annotation feedback), and updates the classification boundary by fine-tuning the model parameters.

[0088] New field naming rules can be understood as the naming conventions for fields added to the business system (such as customer risk tags). For example, when a bank adds a "customer risk tag" field, the semantic recognition model can automatically learn its semantic features and adjust the classification boundaries. Sample data feedback can be understood as the field classification results annotated by users, such as customer_risk_score being classified as a customer risk tag. Model parameters may include, for example, the weight matrix of each network layer in the semantic recognition model, and the weight values ​​of neurons in each network layer.

[0089] Before performing feature fusion analysis on the semantic recognition model, the incremental learning algorithm receives new field naming rules and sample data feedback to dynamically update the model parameters. This step ensures that the semantic recognition model can adapt to changes in the business system and improves its long-term adaptability.

[0090] Incremental learning algorithms enable semantic recognition models to quickly adapt to new field naming rules and changes in data distribution by fine-tuning model parameters. For example, when a new field, customer_risk_score, is introduced, the model learns its semantic features through online fine-tuning and adjusts the classification boundaries to improve recognition accuracy.

[0091] To detect whether incremental learning has been triggered, a model drift detection module can be introduced. This module, for example, is based on relative entropy (Kullback-Leibler Divergence, KL divergence). By statistically analyzing the differences in the distribution of old and new data (such as KL divergence), the model drift detection module can dynamically determine whether incremental learning needs to be triggered, thus preventing the performance of the semantic recognition model from degrading as the business system changes.

[0092] For example, by calculating the KL divergence between the distribution of the new field customer_risk_score and the historical data, and determining that there is a significant difference, incremental learning of the semantic recognition model can be automatically triggered to fine-tune the model parameters.

[0093] In this embodiment, an incremental learning algorithm is used to dynamically update the model parameters of the semantic recognition model, enabling it to quickly adapt to new field naming rules and changes in data distribution, continuously adapt to changes in business systems, avoid model performance degradation, and improve the long-term stability of privacy field recognition capabilities. Simultaneously, a model drift compensation mechanism can dynamically identify changes in data distribution, improving the model's long-term stability and adaptability, ensuring that privacy field recognition capabilities continuously match business needs.

[0094] For example, if the policy generation module cannot dynamically adjust the de-identification strength according to the access scenario, it may lead to a mismatch between the privacy protection policy and business needs. To this end, the method provided in this application embodiment can introduce a scenario classifier into the policy generation module, and dynamically match the de-identification policy template (such as partial masking in the development environment and hash encryption for third-party interface calls) by analyzing the calling role, Internet Protocol address (IP), interface type, and other features of the access request.

[0095] In one possible implementation, generating a privacy protection strategy for a field based on the privacy type classification result includes: matching anonymization strategies, encryption strategies, and / or access control strategies according to the field sensitivity level corresponding to the privacy type classification result; and generating a privacy protection strategy for the field based on the successfully matched anonymization strategies, encryption strategies, and / or access control strategies.

[0096] For example, the field sensitivity level can be understood as the sensitivity level corresponding to the privacy type classification result. For instance, the preset field sensitivity levels include three levels: Level 1, Level 2, and Level 3, with sensitivity decreasing sequentially. Higher levels indicate greater sensitivity, and the consequences of field leakage are more severe. The field sensitivity level corresponding to the current privacy type classification result can be determined based on the mapping relationship between the privacy type classification result and the preset field sensitivity levels.

[0097] Desensitization strategies can be understood as strategies for preserving the format of sensitive data or partially hiding it (such as replacing the middle part of a mobile phone number with asterisks, e.g., 1234***5678). Desensitization strategies include partial masking, hash mapping, and other strategies.

[0098] Encryption strategies can be understood as reversible or irreversible encryption of sensitive data. Examples of encryption strategies include those that encrypt data using algorithms such as the Advanced Encryption Standard (AES) or the RSA (Rivest-Shamir-Adleman) encryption algorithm.

[0099] Access control policies can be understood as restricting the access permissions of specific roles or interfaces, such as prohibiting third parties from accessing the ID number field.

[0100] In the step of generating a privacy protection policy, different policy templates can be matched based on the sensitivity level of the fields corresponding to the privacy type classification results (such as one of level 1, 2, or 3). For example, level 1 sensitive fields (such as ID card numbers) are matched with encryption policies, level 2 sensitive fields (such as mobile phone numbers) are matched with de-identification policies, and level 3 sensitive fields (such as customer tags) are matched with access control policies. This step achieves precise adaptation of protection strength and risk level through flexible matching of policy templates.

[0101] In this embodiment, the precise matching of protection strength and field sensitivity level is achieved through flexible matching of privacy protection strategies. For example, matching encryption strategies with level-one sensitive fields can avoid the risk of data leakage; matching de-identification strategies with level-two sensitive fields can preserve data availability. Based on this, the scenario adaptability of privacy protection strategies can be improved, avoiding data value loss due to excessive de-identification or security risks caused by insufficient protection, while meeting the compliance requirements of different business scenarios.

[0102] In one possible implementation, after matching the desensitization strategy according to the field sensitivity level corresponding to the privacy type classification result, the method further includes: adjusting the desensitization strategy to enhance or weaken the desensitization intensity based on the field sensitivity level corresponding to the privacy type classification result and the access scenario of accessing the database.

[0103] For example, an access scenario can be understood as the context of database access, such as a development environment, a production environment, or a third-party API call.

[0104] Different access scenarios will have different requirements for the strength of data masking. In the process of matching data masking strategies, the strength of the data masking strategy is dynamically adjusted according to the sensitivity level of the field (such as level 1, level 2, and level 3) and the database access scenario (such as development environment, production environment, or third-party interface call).

[0105] A scenario classifier can be understood as a tool for classifying and identifying the current scenario. By analyzing the contextual features of access requests (such as whether the caller is a risk control system or a third-party interface), the scenario classifier can automatically select a de-identification strategy that matches the business requirements based on a preset algorithm.

[0106] A scenario classifier can be introduced into the policy generation module. By analyzing features such as the calling role, IP address, and interface type of the access request, the scenario classifier can determine the current access scenario. Then, combined with a scenario rule base, the de-identification strength of the de-identification policy can be dynamically adjusted. The scenario rule base can include multiple rules corresponding to different access scenarios to adjust the de-identification strength. For example, a rule could be set to "fully encrypt sensitive fields when calling third-party interfaces" to enhance de-identification strength.

[0107] For example, when the scenario classifier detects that the caller is an internal data analysis system, it adjusts the generated de-identification strategy based on the sensitivity level of the fields corresponding to the privacy type classification result and the internal data analysis system (access scenario) to weaken the de-identification strength. For instance, it might use a partial de-identification strategy for transaction detail fields to retain data availability. Conversely, when the scenario classifier detects that the caller is a third-party interface (access scenario), it adjusts the generated de-identification strategy to strengthen its de-identification strength. For example, it might use a full de-identification strategy for transaction detail fields to ensure data security.

[0108] For example, a partial masking desensitization strategy can be used in the development environment (access scenario), while a hash encryption desensitization strength can be used in the production environment. For instance, a first-level sensitive field (such as a mobile phone number) can be partially masked (e.g., 1234***5678) in the development environment, while hash encryption, such as SHA256 (mobile), can be used when calling a third-party interface (access scenario).

[0109] In this embodiment, by dynamically adapting the de-identification strength based on sensitivity levels and access scenarios, the business compatibility of the de-identification strategy can be improved. This ensures that the de-identification strength accurately matches the business needs of the access scenario, avoiding data availability degradation due to excessive de-identification or security risks caused by insufficient protection. Simultaneously, the dynamic strategy mechanism can adapt to the compliance requirements of different business scenarios, enhancing system flexibility and security. For example, partial masking can be used in the development environment to preserve data availability, while hash encryption can be used when calling third-party interfaces to ensure data irreversibility. Based on this, the scenario adaptability of the privacy protection strategy is improved, avoiding data value loss due to excessive de-identification.

[0110] For example, based on the above embodiments, a mechanism for fine-grained sensitivity grading and risk quantification assessment can be introduced. A risk quantification assessment model is introduced into the sensitivity grading module, comprehensively calculating a risk score based on the consequences of field leakage (such as identity theft risk, financial loss), data usage (such as open interfaces, internal analysis), and access roles (such as risk control systems, third-party applications), thus achieving more fine-grained grading. For example, a higher risk score can be assigned to the bank card number field in third-party interface call scenarios.

[0111] Specifically, a hybrid risk assessment model based on rules and machine learning is constructed. The rule layer defines basic risk factors (e.g., ID number leakage risk = level 5), and the machine learning layer trains a risk prediction model (e.g., a regression model) using historical data on data leaks. A risk score is generated by combining field attributes, scenario characteristics, and access behavior. For example, the risk score for the field "Transaction Details" is 3.2 when accessed in the risk control system, but 4.8 when accessed via a third-party API.

[0112] Based on this, a dynamic and fine-grained classification of privacy fields is achieved through risk quantification assessment, avoiding the coarse-grained problem of static classification rules. Simultaneously, by combining the predictive capabilities of historical data breaches, the matching degree between classification results and actual risks is improved, guiding more precise configuration of protection strategies.

[0113] In one possible implementation, after generating the privacy protection strategy for the field based on the privacy type classification result, the method further includes: controlling the consistency of the privacy protection strategy during execution through a transactional repair control algorithm, wherein the transactional repair control algorithm is a control algorithm that supports rollback operations.

[0114] For example, a transactional repair control algorithm can be understood as an algorithm that ensures the consistency of the execution of privacy protection strategies based on database transaction repair control. For instance, a transactional repair control algorithm can roll back inconsistent privacy protection strategy execution results to roll back operations that fail to protect against security failures.

[0115] After generating a privacy protection policy, electronic devices can ensure the consistency of the policy's execution through transactional repair control algorithms, such as preset repair code based on database transactions. For example, if the privacy protection policy deployment fails, the electronic device automatically rolls back to the state before the policy change, avoiding database service interruption due to execution anomalies.

[0116] In this embodiment, the consistency of the privacy protection strategy during execution is controlled by a transactional repair control algorithm, which can avoid database service interruption, improve the reliability and stability of privacy protection strategy execution, and ensure the continuity of the privacy protection process.

[0117] The privacy field processing method provided in this application can be understood as a privacy field identification and classification system for bank databases based on a semantic recognition model and an adaptive policy engine. By analyzing field naming, sample data characteristics, access log behavior, and contextual semantics, it automatically completes the detection and risk classification of privacy-sensitive fields and generates corresponding de-identification, encryption, or access control policies.

[0118] The following is combined Figure 2 The privacy field processing method provided in the embodiments of this application will be further described. Figure 2 This is a schematic diagram of the structure of a privacy field processing system provided in the embodiments of this application. This privacy field processing system can be used to execute the privacy field processing methods provided in the above embodiments. The privacy field processing system includes multiple functional units, such as... Figure 2 As shown, it includes a data acquisition module, an AI semantic recognition module, a feature fusion module, a sensitivity classification module, a policy generation module, and an audit and explainability module.

[0119] The system comprises the following modules: a data acquisition module for extracting field metadata and sample features; an AI semantic recognition module for identifying field semantic categories using a converter-based bidirectional encoder representation model; a feature fusion module for integrating field features and log behavior features; a sensitivity grading module for outputting sensitivity levels and matching them to policy templates; a policy generation module for automatically generating and deploying de-identification and encryption policies; and an auditing and explainability module for recording recognition results and AI inference paths.

[0120] Specifically, the data acquisition module can extract field metadata (field name, type, comments, table name context, etc.) from the database system, and combine it with sample data statistical characteristics (length distribution, character set characteristics, data uniqueness) and access log information (query frequency, calling role) and other information.

[0121] The AI ​​semantic recognition module can use a transducer-based bidirectional encoder representation model to extract field names and contextual annotation semantics; it is then input into a classification network via a feature fusion layer; the output is the field privacy type (such as name, ID number, contact information, account information, etc.) and its confidence level. The model can be continuously optimized through incremental learning.

[0122] The sensitivity grading module can classify fields according to the identification results, the bank's internal security regulations, and the origin information of internet data breaches. For example, it can be divided into the following three levels: Level 1 (highly sensitive): ID card number, bank card number, etc.; Level 2 (medium sensitive): mobile phone number, email address; Level 3 (generally sensitive): customer tags, region, etc.

[0123] The strategy generation and execution module can automatically generate de-identification strategies (mask, hash, partial display, etc.) or encryption strategies (AES, RSA, etc.) based on the hierarchical results; dynamic configuration is achieved through integration with the database strategy center. Transactional repair control: All repair operations are executed based on transactions, ensuring execution consistency and rollback capabilities.

[0124] The auditing and explainability module can log and track the recognition process of the AI ​​semantic recognition module; output explainable reports explaining the field classification and strategy recommendation basis; and support regulatory compliance audit traceability. Visual interface or application programming interface: It supports displaying comparison results and remediation task status through a web front-end, while also providing an open RESTful API for easy integration with existing bank systems.

[0125] The privacy field processing system can execute the privacy field processing method provided in any of the above embodiments to achieve the goals of automated identification and classification, improved identification accuracy, intelligent recommendation of privacy protection strategies, compliance and explainability, and continuous learning and optimization.

[0126] For example, by implementing the above methods, intelligent identification and automated classification of privacy fields in the database can be achieved, reducing manual rule maintenance work. Through AI semantic modeling, the identification accuracy is improved compared to traditional regular expression methods. Based on the classification results, de-identification or encryption strategies are automatically configured, reducing the risk of manual configuration. The system provides model inference audit reports, meeting the traceability requirements of banking regulatory authorities. The model can continuously learn from newly accessed systems and log feedback, improving its adaptability.

[0127] Figure 3 This is a schematic diagram of the privacy field processing device provided in the embodiments of this application, as shown below. Figure 3 As shown, this application embodiment provides a privacy field processing device, which includes:

[0128] The acquisition module 301 is used to acquire the metadata and access log information of the fields in the database, and extract the multidimensional features of the fields through the metadata and access log information;

[0129] Output module 302 is used to input multidimensional features into the semantic recognition model for feature fusion analysis and output the privacy type classification result of the field.

[0130] The generation module 303 is used to generate privacy protection strategies for fields based on the privacy type classification results.

[0131] In one possible implementation, the acquisition module 301 is specifically used for:

[0132] Multidimensional features of a field are extracted by analyzing the field name, data type, table name, and / or field comments in the field's metadata, the query frequency, calling role, and / or interface type in the field's access log information, and the statistical features of the field's sample data. The statistical features of the field's sample data include at least one of the following: character set distribution, uniqueness, and data length.

[0133] In one possible implementation, the device further includes a recording module for:

[0134] Record the reasoning path of the semantic recognition model during feature fusion analysis, as well as the feature fusion weights assigned to each feature during feature fusion;

[0135] Based on inference paths and feature fusion weights, an interpretability log is generated for the privacy type classification results of the generated fields.

[0136] In one possible implementation, the apparatus further includes an attack detection module for:

[0137] Retrieve access logs for database access;

[0138] Analyze database access behavior based on access logs and identify high-risk attack patterns;

[0139] Based on the identification results of high-risk attack patterns, adjust the privacy protection strategy for the fields.

[0140] In one possible implementation, the device further includes an incremental learning module for:

[0141] Receive feedback on new field naming rules and sample data. The sample data feedback is the user's feedback result on the privacy type classification of the field.

[0142] Based on the new field naming rules and sample data feedback, the model parameters of the semantic recognition model are dynamically updated through an incremental learning algorithm.

[0143] In one possible implementation, the generation module 303 is specifically used for:

[0144] Based on the sensitivity level of the field corresponding to the privacy type classification result, match the de-identification strategy, encryption strategy and / or access control strategy;

[0145] Generate a privacy protection policy for the field based on the successfully matched de-identification policy, encryption policy, and / or access control policy.

[0146] In one possible implementation, the generation module 303 is further configured to:

[0147] Based on the sensitivity level of the fields corresponding to the privacy type classification results, and the access scenarios for accessing the database, the de-identification strategy is adjusted to enhance or weaken the de-identification strength.

[0148] In one possible implementation, the device further includes a control module for:

[0149] The consistency of privacy protection policies during execution is controlled through a transactional repair control algorithm, which includes a control algorithm that supports rollback operations.

[0150] The privacy field processing apparatus provided in this application embodiment can be used to execute the technical solution of the privacy field processing method in any of the above embodiments of this application. Its implementation principle and technical effect are similar, and will not be described again here.

[0151] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 4 As shown, the electronic device of this embodiment may include: at least one processor 401; and a memory 402 communicatively connected to the at least one processor; wherein the memory 402 stores instructions that can be executed by the at least one processor 401, and the instructions are executed by the at least one processor 401 to cause the electronic device to perform the method as described in any of the above embodiments.

[0152] Optionally, the memory 402 can be either standalone or integrated with the processor 401.

[0153] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the foregoing embodiments, and will not be repeated here.

[0154] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method of any of the foregoing embodiments.

[0155] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method of any of the foregoing embodiments.

[0156] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0157] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0158] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU) or other general-purpose processors. The processor can also be a Digital Signal Processor (DSP) or an Application Specific Integrated Circuit (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0159] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be various media that can store program code, such as USB flash drives, portable hard drives, read-only memory (ROM), disks or optical discs.

[0160] The aforementioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Examples of storage media include Static Random-Access Memory (SRAM) or Electrically Erasable Programmable Read Only Memory (EEPROM).

[0161] Storage media can be, for example, erasable programmable read-only memory (EPROM) or programmable read-only memory (PROM). Storage media can also be read-only memory (ROM), magnetic storage, flash memory, magnetic disks, or optical disks. Storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0162] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside within an application-specific integrated circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components within an electronic device or host device.

[0163] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0164] The sequence numbers of the embodiments in this application are merely for description and do not represent the superiority or inferiority of the embodiments. Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0165] Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0166] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0167] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0168] It should be further noted that although the steps in the flowchart are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0169] Furthermore, at least some steps in the flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0170] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0171] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0172] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for processing privacy fields, characterized in that, The method includes: Obtain metadata and access log information of fields in the database, and extract multidimensional features of the fields using the metadata and access log information; The multidimensional features are input into a semantic recognition model for feature fusion analysis, and the privacy type classification result of the field is output. Generate a privacy protection strategy for the field based on the privacy type classification result.

2. The method according to claim 1, characterized in that, The step of extracting multidimensional features of the field using the metadata and access log information includes: The multidimensional features of the field are extracted by using the field name, data type, table name and / or field comments in the field's metadata, the query frequency, calling role and / or interface type in the field's access log information, and the statistical features of the field's sample data. The statistical features of the field's sample data include at least one of character set distribution, uniqueness and data length.

3. The method according to claim 1, characterized in that, After generating the privacy protection strategy for the field based on the privacy type classification result, the method further includes: Record the reasoning path of the semantic recognition model during the feature fusion analysis process, as well as the feature fusion weights assigned to each feature during feature fusion; Based on the inference path and the feature fusion weights, an interpretability log of the privacy type classification result of the field is generated.

4. The method according to claim 1, characterized in that, After generating the privacy protection strategy for the field based on the privacy type classification result, the method further includes: Obtain access logs for accessing the database; The access behavior to the database is analyzed based on the access logs, and high-risk attack patterns are identified. Based on the identification results of the high-risk attack patterns, the privacy protection strategy for the field is adjusted.

5. The method according to claim 1, characterized in that, Before inputting the multidimensional features into the semantic recognition model for feature fusion analysis, the method further includes: Receive feedback on new field naming rules and sample data, wherein the sample data feedback is the user's feedback result on the privacy type classification of the field; Based on the new field naming rules and the sample data feedback, the model parameters of the semantic recognition model are dynamically updated through an incremental learning algorithm.

6. The method according to any one of claims 1-5, characterized in that, The step of generating the privacy protection strategy for the field based on the privacy type classification result includes: Based on the field sensitivity level corresponding to the privacy type classification result, match the de-identification strategy, encryption strategy and / or access control strategy; Based on the successfully matched de-identification strategy, encryption strategy, and / or access control strategy, generate the privacy protection strategy for the field.

7. The method according to claim 6, characterized in that, After matching the desensitization strategy according to the field sensitivity level corresponding to the privacy type classification result, the method further includes: Based on the sensitivity level of the fields corresponding to the privacy type classification results, and the access scenarios for accessing the database, the de-identification strategy is adjusted to enhance or weaken the de-identification strength of the strategy.

8. The method according to any one of claims 1-5, characterized in that, After generating the privacy protection strategy for the field based on the privacy type classification result, the method further includes: The consistency of the privacy protection strategy during execution is controlled by a transactional repair control algorithm, which includes a control algorithm that supports rollback operations.

9. A privacy field processing device, characterized in that, The device includes: The acquisition module is used to acquire metadata and access log information of fields in the database, and extract multidimensional features of the fields through the metadata and access log information; The output module is used to input the multidimensional features into the semantic recognition model for feature fusion analysis and output the privacy type classification result of the field. The generation module is used to generate a privacy protection strategy for the field based on the privacy type classification result.

10. An electronic device, characterized in that, include: Memory and processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-8.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-8.

Images