Data privacy protection management system based on artificial intelligence

Abstract

The invention relates to the technical field of data privacy protection, in particular to a data privacy protection management system based on artificial intelligence, which is characterized in that a resource library building module is used for extracting privacy protection knowledge from multi-source data and building a structured resource library; and the privacy classification module classifies data privacy information and evaluates a sensitive level through intelligent identification and analysis, and provides a basis for a differential protection strategy. The privacy protection module fuses privacy types and sensitive levels, constructs a core policy engine, and comprises a privacy desensitization sub-engine, an access control sub-engine and a privacy risk prediction sub-engine, so as to realize automatic generation and dynamic adaptation of protection policies. And the privacy management module defines an execution sequence and a collaboration mechanism of each sub-engine according to a pre-configuration process, generates a privacy protection method covering the full life cycle of the data, ensures end-to-end collaboration management of links such as data acquisition, storage, use and sharing, and remarkably improves the compliance, efficiency and risk prevention and control capability of data privacy protection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data privacy protection, and particularly relates to a data privacy protection management system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of information technology and the deepening of digital transformation, data has become a core production factor driving enterprise innovation. However, the increasing value of data is accompanied by increasingly severe data privacy and security challenges. Data leakage incidents occur frequently, and user privacy rights and interests are damaged, which poses a major threat to personal information security and enterprise compliance operation. At the same time, higher requirements are put forward for the data privacy governance capability of enterprises, and it is urgent to build an intelligent privacy protection system covering the whole life cycle of data.

[0003] Traditional data privacy protection methods mainly rely on manual strategy formulation, static rule matching and fragmented management, and are difficult to cope with the complexity and dynamics of multi-source heterogeneous data. Specifically, the existing solutions have the following pain points: first, data privacy resources are scattered in heterogeneous data sources such as laws and regulations, industry standards and technical solutions, and lack of systematic knowledge integration and structured management, making it difficult to quickly obtain complete compliance basis when formulating strategies; second, the identification and classification of data privacy information rely on manual annotation or simple rule matching, which cannot efficiently process unstructured data, and the dynamic assessment capability of data sensitivity level is insufficient; third, the generation and execution of privacy protection strategies lack intelligent collaboration, and it is difficult to dynamically adjust desensitization rules, access control strategies and risk prediction models according to data types, sensitivity levels and application scenarios, resulting in a mismatch between protection strength and business needs; fourth, the privacy protection processes of each link in the whole life cycle of data are fragmented, and lack of end-to-end collaborative management mechanism, making it difficult to realize real-time monitoring of risks and automatic adaptation of strategies.

[0004] The above content is only used to assist in understanding the technical solutions of the present application, and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide a data privacy protection management system based on artificial intelligence, which aims to solve the technical problems that traditional data privacy protection methods are difficult to cope with complex compliance requirements due to resource fragmentation, rough classification, static strategy and fragmented management with the development of information technology and the increase of data value.

[0006] To achieve the above purpose, the present application provides a data privacy protection management system based on artificial intelligence, which comprises a resource database module, a privacy classification module, a privacy protection module and a privacy management module. The resource library module is configured to construct a data privacy protection library based on privacy protection knowledge and information extracted from data privacy protection related materials; The privacy classification module is configured to identify and classify different types of data privacy information and corresponding sensitive levels by intelligently identifying and analyzing the data privacy protection library, wherein different sensitive levels correspond to different intensity privacy protection strategies; The privacy protection module is configured to fuse the different types of data privacy information and sensitive levels to construct a core strategy engine of the artificial intelligence-based data privacy protection management system, wherein the core strategy engine includes a privacy desensitization sub-engine, an access control sub-engine, and a privacy risk prediction sub-engine. The privacy management module is configured to generate an artificial intelligence-based data privacy protection management method according to a preconfigured privacy protection process and the core strategy engine, wherein the privacy protection process is used to determine the execution order and coordination mode between the sub-engines in the core strategy engine to complete the privacy protection of the data throughout the life cycle.

[0007] Optionally, the data privacy protection library is constructed based on the privacy protection knowledge and information extracted from the data privacy protection related materials, and includes: The obtained initial data privacy protection related materials are parsed, and the parsing results of the initial data privacy protection related materials are standardized in format based on a data format conversion tool to generate the data privacy protection related materials; The privacy protection constituent elements in the data privacy protection related materials are located through a pre-defined privacy element classification rule and a risk index extraction mode, and the core privacy index set of the data privacy protection related materials is determined from the constituent elements based on the appearance frequency, correlation relationship, and application scenario information of the privacy elements in the data privacy protection related materials; The privacy entities and relationship rules included in the data privacy protection related materials are obtained by using knowledge graph construction technology, and the features of the privacy entities and the relationship rules are deeply mined based on semantic web model according to the attribute information and relationship information of the privacy entities, so as to map the privacy information in the data privacy protection related materials to corresponding knowledge graph nodes and edges; The original data of the data privacy protection related materials, the determined core privacy index set, and the generated knowledge graph are associated and integrated by means of privacy data association technology through the unique identifiers of each material in the data privacy protection related materials to generate the data privacy protection library.

[0008] Optionally, the different types of data privacy information and corresponding sensitive levels are identified and classified by intelligently identifying and analyzing the data privacy protection library, and include: Based on the knowledge graph and the core privacy indicator set in the data privacy protection library, semantic similarity values and attribute matching scores of each data item in a data sample and a known privacy entity are calculated, and data items in the data sample are divided into different privacy candidate sets based on the semantic similarity values and the attribute matching scores; The features and association relationships of the data items included in each of the privacy candidate sets are parsed to determine the attribute features and data types of different fields in each of the data items, and key privacy information is extracted by identifying identifiers, sensitive attributes, and associated data constituent elements in each of the data items, the key privacy information being used to explicitly determine the privacy sensitivity of different fields in each of the data items; Based on the attribute features, the data types, and the privacy sensitivity, data items in each of the privacy candidate sets are summarized to generate privacy type definitions of each of the privacy candidate sets, the privacy type definitions being used to reflect the category positioning and protection requirements of the data items in the privacy candidate sets in data privacy protection; Based on the privacy type definitions in each of the privacy candidate sets, the specific composition of the data items in each of the privacy candidate sets, and a preset sensitive level scoring rule, typical data instances in each of the privacy candidate sets are determined as privacy samples of each of the privacy candidate sets; According to the privacy type definitions and the privacy samples in each of the privacy candidate sets, privacy protection level standards of each of the privacy candidate sets are determined from preset privacy protection regulations and standards, the privacy protection level standards being used to explicitly determine the sensitive levels, protection measures, and processing requirements of the data items in the privacy candidate sets; The consistency of the privacy type definitions, the privacy samples, and the privacy protection level standards corresponding to each data item in each of the privacy candidate sets is referred to to determine different types of data privacy information and corresponding sensitive levels of the data privacy protection library in each of the privacy candidate sets.

[0009] Optionally, the different types of data privacy information and sensitive levels are fused to construct a core strategy engine of an artificial intelligence-based data privacy protection management system, including: According to the types and sensitive levels of each of the data privacy information, privacy attributes, data features, and processing requirements of each of the data privacy information are obtained, and the privacy attributes, the data features, and the processing requirements are input into an intelligent strategy generation engine to drive the intelligent strategy generation engine to generate a privacy protection strategy template corresponding to each of the data privacy information and the sensitive level; Based on the privacy protection policy template and the privacy samples of the data privacy information, secondary development and configuration are performed on an initial privacy protection tool to generate a privacy desensitization sub-engine and an access control sub-engine corresponding to each data privacy information and sensitive level; The privacy protection level standards of each data privacy information and sensitive level are converted into a calculable risk assessment index and model to construct a risk prediction algorithm corresponding to each privacy protection level standard, and a corresponding privacy risk prediction sub-engine is generated based on each risk prediction algorithm; By deeply mining the privacy requirements of each stage of the data life cycle and the association between each data privacy information, and based on the full life cycle privacy requirements and the association, a collaborative workflow of each privacy desensitization sub-engine, access control sub-engine, and privacy risk prediction sub-engine is determined, and a strategy coordination engine related to the collaborative workflow is generated; According to the strategy coordination engine, the privacy desensitization sub-engine, the access control sub-engine, and the privacy risk prediction sub-engine corresponding to each data privacy information and sensitive level, the core strategy engine of the artificial intelligence-based data privacy protection management system is constructed.

[0010] Optionally, the artificial intelligence-based data privacy protection management method is generated according to the preconfigured privacy protection process and the core strategy engine, comprising: By formalizing the privacy protection process, the privacy protection stage nodes, the inter-stage dependency relationships, and the conversion rules between the stage nodes in the privacy protection process are obtained, the privacy protection stage nodes represent specific privacy protection tasks, the inter-stage dependency relationships represent task order, and the conversion rules represent triggering conditions for task start or completion; According to the privacy protection stage nodes, the inter-stage dependency relationships, and the conversion rules, the calling order and data interaction mode of each privacy desensitization sub-engine, access control sub-engine, privacy risk prediction sub-engine, and strategy coordination engine are determined; Based on the calling order and the data interaction mode, the strategy calling interface and the data flow path in the initial privacy protection management platform are configured to generate the artificial intelligence-based data privacy protection management method.

[0011] Optionally, the semantic similarity value and attribute matching degree score of each data item in the data sample and the known privacy entity are calculated based on the knowledge graph and the core privacy index set in the data privacy protection resource library, comprising: extracting semantic feature vectors of each known privacy entity from the knowledge graph and extracting attribute feature vectors of each known privacy entity from the core privacy indicator set, and aggregating the semantic feature vectors and the attribute feature vectors of each known privacy entity into a privacy entity comprehensive feature vector through a feature fusion algorithm; extracting semantic description vectors and attribute description vectors of each data item from the data sample, and aggregating them into data item comprehensive feature vectors through the same feature fusion algorithm; calculating semantic similarity values between each data item comprehensive feature vector and each privacy entity comprehensive feature vector based on a cosine similarity algorithm, and calculating attribute matching scores between each data item and each known privacy entity based on attribute field matching and value range comparison algorithms.

[0012] Optionally, after calculating the semantic similarity values based on the cosine similarity algorithm and calculating the attribute matching scores based on the attribute field matching, the method further comprises: weighting and fusing the semantic similarity values and the attribute matching scores according to a preset weight allocation strategy to generate comprehensive association degree scores between each data item and each known privacy entity; constructing a data-privacy entity association matrix according to the comprehensive association degree scores, and performing unsupervised clustering analysis on the data-privacy entity association matrix based on a density clustering algorithm to generate a data privacy clustering result; According to the data density distribution characteristics and the difference characteristics between clusters in the data privacy clustering result, dynamically adjusting the clustering parameters and re-dividing the data clusters until the comprehensive association degree score variance of the data items in each cluster is lower than a preset threshold, to generate the privacy candidate set.

[0013] Optionally, inputting the privacy attribute, the data feature, and the processing requirement into an intelligent policy generation engine to drive the intelligent policy generation engine to generate a privacy protection policy template corresponding to each data privacy information and sensitivity level, comprises: converting the privacy attribute into a privacy type description statement, converting the data feature into a data format constraint list, and converting the processing requirement into a compliance control specification; organizing the privacy type description statement, the data format constraint list, and the compliance control specification into an initial policy definition sequence in a logical order based on a preset policy template generation rule; performing syntax analysis and rule verification on the initial policy definition sequence through the intelligent policy generation engine to generate a candidate privacy protection policy template set; The privacy sample is used as verification data, and the intelligent strategy generation engine generates a simulated privacy protection scheme based on the candidate privacy protection strategy template and calculates the compliance matching degree and protection strength index between the simulated privacy protection scheme and the privacy protection level standard; The candidate privacy protection strategy template with a compliance matching degree higher than a first preset threshold and a protection strength index higher than a second preset threshold is selected as the privacy protection strategy template corresponding to the data privacy information and the sensitive level.

[0014] Optionally, when the privacy protection constituent elements in the data privacy protection related information are located through the pre-defined privacy element classification rule and risk index extraction mode, the method further includes: Identifying a combination of privacy elements with a fixed association relationship in the data privacy protection related information, and extracting a core element and an associated element relationship in the combination of privacy elements as a privacy element hierarchical label; Analyzing the attribute characteristics and risk parameters of each element in the combination of privacy elements, and associating the attribute characteristics and the risk parameters to the privacy element hierarchical label to form a structured privacy element attribute table; Detecting the influence relationship and dependency type between the privacy elements in the data privacy protection related information, and dividing a risk transmission chain of the privacy protection constituent elements according to the influence relationship and the dependency type; Mapping the structured privacy element attribute table and the risk transmission chain to a pre-set privacy protection framework classification system to generate a multi-level risk relationship network of the privacy protection constituent elements; Based on the multi-level risk relationship network, the similar but differently sourced privacy elements described in the data privacy protection related information are normalized to eliminate the repeatedly defined privacy elements.

[0015] In this invention, an AI-based data privacy protection management system integrates heterogeneous data from multiple sources, including laws and regulations, technical solutions, and case studies, through a resource database module. This constructs a structured privacy protection resource database, solving the problems of fragmented privacy knowledge and difficulty in efficient reuse in traditional solutions. It provides comprehensive and accurate compliance basis and technical support for subsequent strategy formulation. The privacy classification module performs in-depth analysis of the resource database based on AI technology, achieving intelligent identification, accurate classification, and sensitivity level assessment of data privacy information. Compared with traditional manual annotation or simple rule matching, it significantly improves the processing efficiency of unstructured data and can dynamically adapt to changes in data type and sensitivity, providing precise input for differentiated protection strategies. The privacy protection module integrates data privacy types and sensitivity levels to construct a core strategy engine that includes privacy desensitization, access control, and risk prediction, achieving automated generation and intelligent adaptation of protection strategies. Each sub-engine can dynamically adjust its protection strength based on data characteristics, solving the problem of the disconnect between traditional static strategies and business scenarios, and balancing the strength of privacy protection with data utilization efficiency. The privacy management module defines the execution order and collaboration mechanism of each sub-engine through pre-configured processes, incorporating the entire lifecycle of data collection, storage, use, and sharing into a unified management framework, forming a closed loop of privacy protection covering the entire data flow chain. This effectively solves the problems of fragmented links and lagging risk monitoring in traditional solutions, improving the systematicness and real-time nature of data privacy protection. By integrating regulatory standards, dynamic risk prediction, and full-process strategy collaboration, the solution ensures that data processing complies with the latest compliance requirements. At the same time, it uses historical cases and risk indicator libraries to identify potential risks in advance, reducing the probability of privacy leakage. This provides enterprises with an intelligent and automated comprehensive solution to meet regulatory requirements and actual business needs. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of the first embodiment of the data privacy protection management system based on artificial intelligence of the present invention; Figure 2 This is a flowchart illustrating the specific steps of the data privacy protection management system based on artificial intelligence according to the present invention; Figure 3 This is a flowchart illustrating the specific steps involved in building the core policy engine in the AI-based data privacy protection management system of this invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0018] This invention provides an artificial intelligence-based data privacy protection management system, referring to...Figure 1 As shown, Figure 1 This is a structural block diagram of the first embodiment of the AI-based data privacy protection management system of the present invention. The AI-based data privacy protection management system of the present invention includes: a resource database construction module, a privacy classification module, a privacy protection module, and a privacy management module.

[0019] The resource library module is used to build a data privacy protection resource library based on privacy protection knowledge and information extracted from data privacy protection related materials. Data privacy protection related materials include at least one of the following: laws and regulations and standards, historical privacy breach cases, privacy protection technical solutions, and privacy risk assessment indicator library.

[0020] A data privacy protection resource repository can be a structured knowledge set used to verify and evaluate data privacy protection-related knowledge and information. It can provide compliance basis and technical reference for the identification, classification, and policy generation of data privacy information. The data privacy protection resource repository can use natural language processing and information extraction technologies to extract entities, relationships, and rules from unstructured text and map them to a predefined ontology model for structured storage. The data privacy protection resource repository can be a compliance knowledge base, a technical solution knowledge base, a risk case knowledge base, etc.

[0021] The privacy classification module is used to intelligently identify and analyze the data privacy protection resource library to identify and classify different types of data privacy information and their corresponding sensitivity levels. Different sensitivity levels are used to correspond to different levels of privacy protection strategies.

[0022] In this embodiment, the privacy classification module can perform semantic parsing on unstructured text in the resource library, extract privacy entities, associate them with a preset classification system, and output type labels and sensitivity scores. The privacy classification module can use a BERT+CRF model to perform entity recognition and relation extraction on regulatory texts, combine expert annotation rules to construct a sensitivity mapping function, or use a graph neural network to model event elements in the case library and infer the sensitivity level of unknown data through similarity matching.

[0023] The privacy protection module is used to integrate different types of data privacy information and sensitivity levels to build the core strategy engine of an AI-based data privacy protection management system. The core strategy engine includes a privacy de-identification sub-engine, an access control sub-engine, and a privacy risk prediction sub-engine.

[0024] In this embodiment, the core policy engine can use privacy type labels and sensitivity levels as input features to drive the parameter configuration and decision logic initialization of the three sub-engines. The core policy engine can construct a multi-task learning framework, jointly training the desensitization rule generator, access policy generator, and risk predictor to share an encoding layer, or adopt a combination of rule engine and deep reinforcement learning to optimize the combined weights of the sub-engines based on historical policy performance feedback. The privacy desensitization sub-engine can rely on the sensitivity level output by the privacy classification module to respond to the process triggering instructions of the privacy management module. The privacy desensitization sub-engine can be a generalized desensitization engine, a perturbation desensitization engine, a replacement desensitization engine, etc. The access control sub-engine can receive sensitivity labels from the privacy classification module and share risk scores with the privacy risk prediction sub-engine to adjust access thresholds. The access control sub-engine can be a role-based access control engine, an attribute-based access control engine, a context-aware access control engine, etc. The privacy risk prediction sub-engine can provide dynamic risk weights for the access control sub-engine and provide enhanced desensitization triggering conditions for the privacy desensitization sub-engine. The privacy risk prediction sub-engine can be a behavior pattern analysis engine, an association risk inference engine, a compliance deviation detection engine, etc.

[0025] The privacy management module is used to generate an AI-based data privacy protection management method based on the pre-configured privacy protection process and the core policy engine. The privacy protection process is used to determine the execution order and collaboration method among the sub-engines in the core policy engine in order to complete the privacy protection of the entire data lifecycle.

[0026] In this embodiment, the privacy management module can sequentially call each sub-engine of the core policy engine according to the preset flowchart definition, and inject control commands at each stage of data flow. The privacy management module can define the process nodes of collection → storage → use → sharing based on the workflow engine (such as Camunda), bind the corresponding sub-engine trigger conditions, or adopt an event-driven architecture to listen for data operation events, automatically match the process context and activate the corresponding sub-engine policy.

[0027] Taking privacy compliance processing in medical data sharing as an example, when the system detects that a medical dataset contains patient names, diagnosis records, and medical insurance numbers, the privacy classification module identifies it as personal health information with a high sensitivity level. The core policy engine automatically activates and replaces the de-identification sub-engine to hash the names, the access control sub-engine to restrict access to authorized doctors only, and the risk prediction sub-engine to assess the risk of third-party leakage and trigger secondary approval. The privacy management module calls the three sub-engines in sequence according to the data sharing process to ensure that the policy deployment is completed before data transmission, without the need for manual intervention throughout the process.

[0028] This embodiment constructs a data privacy protection resource library through a resource library building module, intelligently identifies and analyzes the resource library through a privacy classification module, integrates different types of data privacy information and sensitivity levels to build a core strategy engine through a privacy protection module, and generates a privacy protection management method covering the entire data lifecycle based on a pre-configured privacy protection process and the core strategy engine. This achieves the goal of acquiring semantic representations of multi-source compliance and technical knowledge by constructing a structured data privacy protection resource library. The privacy classification module performs deep semantic analysis of the resource library based on artificial intelligence, enabling automated classification and dynamic sensitivity level assessment of unstructured data. The core strategy engine integrates the classification results, driving the three sub-engines of de-identification, access control, and risk prediction to form a collaborative decision chain, allowing the protection strength to adaptively adjust according to data characteristics. The privacy management module uniformly schedules the execution sequence and interaction logic of each sub-engine in the data collection, storage, use, and sharing stages through a pre-configured process, forming an end-to-end policy closed loop. This mechanism transforms static compliance criteria into dynamic execution instructions, realizing a shift in privacy protection from passive response to proactive adaptation, ensuring that data processing behavior always matches the latest compliance requirements and business scenario risks, thereby reducing the risk of systemic leakage while ensuring data availability.

[0029] In one embodiment, a data privacy protection resource library is constructed based on privacy protection knowledge and information extracted from data privacy protection-related materials, including: The process involves parsing the initial data privacy protection information obtained and standardizing the parsing results using a data format conversion tool to generate data privacy protection information. The initial data privacy protection information comes from at least one of the following sources: government regulatory databases, industry standard documents, academic research literature, and corporate privacy policies. The initial data privacy protection information includes at least one of the following: text files, tabular data, image data, and audio / video data. Text files are parsed using natural language processing technology, tabular data is parsed using structured data parsing technology, and image data and audio / video data are parsed using multimedia content recognition technology. This involves standardizing the format of privacy-related materials from multimodal initial data using natural language processing, structured data parsing, and multimedia content recognition technologies. This can be achieved by employing corresponding parsing techniques to output an intermediate representation with a unified semantic structure. In this embodiment, text files can be extracted using a BERT+Span extraction model to identify privacy clause entities and constraints; tabular data can be extracted using a schema-aware parser to retrieve field-compliance item mappings and converted to JSON-LD format; and audio / video data can be extracted using ASR+OCR joint recognition to retrieve speech text and on-screen subtitles, followed by the extraction of privacy-related scenarios from the visual semantic model. This approach achieves semantic alignment and structural unification of heterogeneous data sources, eliminating the interference of format differences on subsequent analysis.

[0030] By using predefined privacy element classification rules and risk indicator extraction patterns, the privacy protection constituent elements in data privacy protection-related materials are located. Based on the frequency of occurrence, correlation, and application scenario information of privacy elements in data privacy protection-related materials, the core privacy indicator set of data privacy protection-related materials is determined from the constituent elements. The core privacy indicator set can be used to quantify key dimensions of privacy protection, serving as semantic anchors for knowledge graph construction and strategy generation. In this embodiment, the core privacy indicator set can calculate the frequency of occurrence, co-occurrence relationship, and application scenario matching degree of privacy elements in text, tables, audio, and video based on predefined classification rules and risk indicator extraction patterns, thus filtering out high-frequency and highly relevant elements. In this way, reusable and inferable key dimensions of privacy protection can be extracted from massive amounts of redundant information, improving the accuracy and generalization ability of knowledge extraction.

[0031] Knowledge graph construction technology is used to obtain privacy entities and relation rules included in data privacy protection materials. Based on the semantic web model, the features of privacy entities and relation rules are deeply mined according to the attribute information and relation information of privacy entities, so as to map the privacy information in data privacy protection materials into corresponding knowledge graph nodes and edges. In this context, the knowledge graph can be used to explicitly express the semantic relationships and reasoning paths between privacy elements, supporting the automated derivation of complex compliance logic. In this embodiment, the knowledge graph can be based on a semantic web model to extract attributes and model relationships of the parsed privacy entities, constructing a semantic network of nodes and edges. In this way, implicit compliance logic can be transformed into a queryable and reasonable explicit semantic network, supporting the automatic derivation of complex rules.

[0032] By leveraging privacy data association technology, the original data, the determined core privacy indicator set, and the generated knowledge graph of data privacy protection-related materials are linked and integrated through the unique identifiers of each material in the data privacy protection-related materials to generate a data privacy protection resource library; In this embodiment, the data privacy protection resource library may include a raw data layer, an indicator set layer, and a knowledge graph layer. Based on privacy data association technology, the data privacy protection resource library binds the raw data, core privacy indicator set, and knowledge graph with unique identifiers, forming a dynamically updatable triple mapping structure. As the integrated carrier of the core privacy indicator set and knowledge graph, the data privacy protection resource library provides input sources for the privacy classification module and supports the semantic reasoning of the policy engine.

[0033] In one embodiment, by intelligently identifying and analyzing the data privacy protection resource library, different types of data privacy information and their corresponding sensitivity levels are identified and classified, including: Based on the knowledge graph and core privacy indicator set in the data privacy protection resource library, the semantic similarity value and attribute matching score of each data item in the data sample with known privacy entities are calculated, and the data items in the data sample are divided into different privacy candidate sets based on the semantic similarity value and attribute matching score. Knowledge graphs can be graph-structured knowledge representations used for constructing rules for privacy entities and relationships. They can provide reasoning basis for entity relationships in semantic matching of data items and support the identification of cross-source privacy features. Knowledge graphs can extract attributes and model relationships of privacy entities based on semantic web models, forming a semantic network of nodes and edges. These can be compliance rule graphs, risk pattern graphs, technical solution graphs, etc. Core privacy indicator sets can provide quantifiable dimensions for privacy identification, serving as a benchmark for attribute matching scores and sensitivity level determination. Based on predefined classification rules and risk indicator extraction patterns, they can calculate the frequency of occurrence, co-occurrence relationships, and application scenario matching degree of privacy elements in text, tables, audio, and video, filtering out high-frequency, highly correlated elements. These can be compliance obligation indicators, risk exposure indicators, technical control indicators, etc.

[0034] In this embodiment, each field in the data sample is semantically vectorized, and cosine similarity and attribute consistency scores are performed with knowledge graph nodes and core indicator set attributes. Candidate sets are then formed by clustering based on thresholds. Furthermore, Sentence-BERT is used to generate semantic vectors for the field text, and similarity is calculated with knowledge graph entity description vectors. Weighted matching is then performed using attribute keywords from the core indicator set, or a graph embedding model is used to map the field structure and attribute patterns of the core indicator set to the same vector space. The candidate sets are then divided using K-means clustering, achieving automatic clustering from raw data fields to semantic privacy categories, replacing keyword matching and manual classification.

[0035] The characteristics and relationships of the data items included in each privacy candidate set are analyzed to determine the attribute characteristics and data types of different fields in each data item. Key privacy information is extracted by identifying the identifiers, sensitive attributes and related data components in each data item. The key privacy information is used to clarify the privacy sensitivity of different fields in each data item. In this embodiment, the semantic roles of fields within the privacy candidate set are analyzed to identify whether they are unique identifiers, sensitive attributes, or components of associated data combinations. Furthermore, based on regular expression patterns and contextual semantics, it detects whether a field is a unique identifier such as an ID card number or mobile phone number, or identifies whether multiple non-sensitive fields together constitute a combination that can infer sensitive information (such as date of birth + postal code + gender) through graph path analysis. This achieves fine-grained evaluation of field-level privacy sensitivity, breaking through the traditional coarse-grained judgment based on field names or table structures.

[0036] Based on attribute characteristics, data types, and privacy sensitivity, the data items in each privacy candidate set are summarized and a privacy type definition for each privacy candidate set is generated. The privacy type definition is used to reflect the category positioning and protection needs of the data items in the privacy candidate set in data privacy protection. In this embodiment, cluster analysis is performed on the field attributes, data structure, and associated elements of data items within the privacy candidate set to extract common patterns and form standardized type descriptions, generating privacy type definitions for each privacy candidate set. Furthermore, cluster analysis is performed on the field attributes, data structure, and associated elements of data items within the privacy candidate set to extract common patterns and form standardized type descriptions, generating privacy type definitions for each privacy candidate set. This abstracts the original data items into manageable privacy categories, providing a semantic foundation for subsequent sensitivity level mapping.

[0037] Based on the privacy type definitions in each privacy candidate set, the specific composition of the data items in each privacy candidate set, and the preset sensitivity level scoring rules, typical data instances in each privacy candidate set are determined as privacy samples for each privacy candidate set. In this embodiment, the privacy type definition is semantically aligned with the applicable conditions in regulatory clauses. The sensitivity level and processing requirements are output by combining the privacy sample distribution and scoring rules. Typical data instances in each privacy candidate set are then identified as privacy samples for that set. Furthermore, a legal clause matching model is used to map the privacy type definition to specific clauses, or a scoring function is constructed to dynamically calculate the sensitivity level weighted score based on the density, correlation strength, and scenario frequency of sensitive attributes in the privacy samples. This transforms the sensitivity level from empirical setting to compliant semantic reasoning, ensuring that the evaluation results are traceable and verifiable.

[0038] Based on the privacy type definitions and privacy samples in each privacy candidate set, the privacy protection level standards for each privacy candidate set are determined from the preset privacy protection laws and standards. The privacy protection level standards are used to clarify the sensitivity level, protection measures, and processing requirements of each data item in the privacy candidate set. In this embodiment, a semantic matching algorithm is used to align privacy type definitions with compliance clauses in regulatory texts. Sensitivity levels are calculated by combining the distribution characteristics of privacy samples with scoring rules, thus determining the privacy protection level standard for each privacy candidate set. Furthermore, the semantic matching algorithm is used to align privacy type definitions with compliance clauses in regulatory texts. The sensitivity levels are calculated by combining the distribution characteristics of privacy samples with scoring rules, thus determining the privacy protection level standard for each privacy candidate set. This provides the privacy classification module with executable sensitivity level determination criteria, supporting the generation of differentiated protection strategies.

[0039] By referencing the consistency of privacy type definitions, privacy samples, and privacy protection level standards for each data item in each privacy candidate set, different types of data privacy information and their corresponding sensitivity levels are determined for each privacy candidate set corresponding to the data privacy protection resource library.

[0040] In this embodiment, as the final output of privacy classification, its definition is used by the privacy protection module to drive the parameter configuration of the de-identification, access control and risk prediction sub-engines. By referring to the privacy type definition, privacy sample and privacy protection level standard corresponding to each data item in each privacy candidate set, the different types of data privacy information and their corresponding sensitivity levels corresponding to the data privacy protection resource library are determined in each privacy candidate set. Taking e-commerce platform user behavior data processing as an example, the AI-based data privacy protection management system in this embodiment can analyze fields such as delivery address, browsing history, and payment method in the user's order table. Through knowledge graphs, it discovers a geographically related path between the delivery address and identity identifier, and marks the payment method as a financially sensitive item using core indicator sets. Based on semantic similarity and attribute matching, it categorizes these into a candidate privacy set for personal identity and behavioral trajectory. Further, it identifies a combination of delivery address and mobile phone number that can be used to infer a home address, extracting it as key privacy information. The privacy type of this set is defined as the behavioral trajectory class that can be associated with an identity. By comparing the provisions in regulations concerning user behavior data involving personal life characteristics, and combining high-frequency association patterns in the samples, its sensitivity level is determined to be high, and desensitization requirements and access approval processes are bound. Finally, the privacy protection level standard of this set is output for the policy engine to call and execute dynamic desensitization and permission interception.

[0041] In one embodiment, different types of data privacy information and sensitivity levels are integrated to construct the core policy engine of an AI-based data privacy protection management system, including: Based on the type and sensitivity level of each data privacy information, the privacy attributes, data characteristics, and processing requirements of each data privacy information are obtained. The privacy attributes, data characteristics, and processing requirements are then input into the intelligent policy generation engine to drive the intelligent policy generation engine to generate privacy protection policy templates corresponding to each data privacy information and sensitivity level.

[0042] Privacy protection policy templates can be parameterized protection logic frameworks generated based on the type, sensitivity level, privacy attributes, data characteristics, and processing requirements of data privacy information. They can be used to provide standardized and reusable policy structure definitions for the privacy desensitization and access control sub-engines. These templates may include generalized desensitization templates, role-based access templates, and conditional approval templates.

[0043] The intelligent policy generation engine can extract field attributes, data structures, and compliance constraints from privacy type definitions, map them to a pre-defined policy template library, and generate parameterizable policy configuration units. The template engine can bind field types and sensitivity levels to a set of de-identification functions (such as masking, generalization, and hashing) to generate structured JSON policy descriptions. It can also use rule induction algorithms to automatically extract access control logic from historical compliance cases, generating attribute-based policy templates. This transforms policy generation from manual writing to automatic generation, ensuring policy semantics are consistent with data and compliance is traceable.

[0044] Based on the privacy protection policy template and privacy samples of data privacy information, the initial privacy protection tool is further developed and configured to generate privacy de-identification sub-engines and access control sub-engines corresponding to each data privacy information and sensitivity level.

[0045] The privacy desensitization sub-engine can be based on the privacy protection policy template, allowing for secondary development and configuration of the initial privacy protection tool to generate privacy desensitization sub-engines corresponding to each data privacy information and sensitivity level. Similarly, the access control sub-engine can be based on the privacy protection policy template, allowing for secondary development and configuration of the initial privacy protection tool to generate access control sub-engines corresponding to each data privacy information and sensitivity level.

[0046] The privacy protection level standards for each data privacy information and sensitivity level are transformed into computable risk assessment indicators and models to construct risk prediction algorithms corresponding to each privacy protection level standard, and generate corresponding privacy risk prediction sub-engines based on each risk prediction algorithm.

[0047] Risk prediction algorithms can transform privacy protection level standards into computable risk assessment models to predict the potential leakage probability of specific data processing behaviors. They can construct classification or regression models based on the sensitivity levels, processing requirements, and historical risk indicators defined in the privacy protection level standards, outputting risk scores. Privacy risk prediction sub-engines can be sub-engines used to predict data leakage risks and may include risk prediction strategy execution units. Corresponding privacy risk prediction sub-engines can be generated based on risk prediction algorithms.

[0048] By deeply exploring the privacy requirements at each stage of the data lifecycle and the relationships between various data privacy information, and based on the privacy requirements and relationships throughout the entire lifecycle, the collaborative workflow of each privacy de-identification sub-engine, access control sub-engine, and privacy risk prediction sub-engine is determined, and a policy coordination engine related to the collaborative workflow is generated.

[0049] A policy coordination engine can be a coordination system that dynamically orchestrates the execution order and triggering conditions of each sub-engine based on privacy requirements and semantic relationships between data items at each stage of the data lifecycle. It can include process-driven coordinators, event-triggered coordinators, context-aware coordinators, etc.

[0050] In this embodiment, taking the cross-departmental access of financial customer data in the analysis platform as an example, when customer data enters the analysis platform from the repository, the strategy coordination engine detects that the data type is a highly sensitive financial transaction record, triggering a risk prediction algorithm to assess that its leakage probability under multi-user concurrent access is high; then, the access control sub-engine is activated, restricting access to authorized analysts only, and linking with the privacy protection policy template to start field-level masking and desensitization; at the same time, the strategy coordination engine monitors subsequent operations, and if it detects that the data is associated with external IP access behavior, it immediately enhances the risk score and triggers a secondary approval process; the entire process does not require manual configuration, and the sub-engines automatically coordinate according to the semantic flow path to achieve dynamic protection throughout the entire process from data entry to use.

[0051] In one embodiment, an AI-based data privacy protection management method is generated based on a pre-configured privacy protection process and a core policy engine, including: The privacy protection process is formally described to obtain the privacy protection stage nodes, inter-stage dependencies, and transition rules between stage nodes. The privacy protection stage nodes represent specific privacy protection tasks, the inter-stage dependencies represent the task order, and the transition rules represent the triggering conditions for task initiation or completion.

[0052] The formal description of the privacy protection process can be a logical framework for structured modeling of stage nodes, inter-stage dependencies, and transformation rules within the privacy protection process. This framework can provide a computable execution logic foundation for sub-engine scheduling, transforming the process from a natural language description to a machine-executable model. In this embodiment, a DAG or state machine model can be used, abstracting privacy protection stages as nodes, defining dependencies as directed edges, and encoding transformation rules as conditional trigger expressions. The formal description of the privacy protection process can include a set of stage node definitions, a set of sequential dependencies, and a set of event trigger rules.

[0053] Based on the privacy protection stage nodes, inter-stage dependencies, and conversion rules, the calling order and data interaction method of each privacy de-identification sub-engine, access control sub-engine, privacy risk prediction sub-engine, and policy coordination engine are determined.

[0054] In this embodiment, the calling order and data interaction method of each sub-engine are determined based on transformation rules and dependencies. The formal flowchart can be traversed, and the activation timing and input / output parameters of each sub-engine at each stage can be derived based on dependency edges and triggering conditions. A topological sorting algorithm can be used to determine the execution sequence of the sub-engines and bind the input data source and output target of each node.

[0055] Based on the call order and data interaction method, configure the policy call interface and data flow path in the initial privacy protection management platform to generate an AI-based data privacy protection management method.

[0056] In this embodiment, a privacy protection management method is generated by configuring the policy call interface and data flow path according to the call order and data interaction method. API endpoints, message topics, or data pipelines can be registered in the platform, and the call parameters and data formats of the sub-engines can be bound. Microservice call chains and event routing rules can be automatically generated using service orchestration tools (such as KubernetesOperator), or a plug-in architecture can be used to dynamically load sub-engine adapters and automatically configure input-output mapping relationships according to the process definition. Taking the cross-institutional sharing process of medical data as an example, the system loads a formalized process: data collection → anonymization → risk assessment → access authorization → sharing transmission; the transformation rule is defined as blocking sharing and triggering secondary approval if the risk prediction value > 0.8; the dependency relationship requires that access control must be executed after risk assessment; the system automatically configures accordingly: data is output to the anonymization sub-engine during the collection phase, and after anonymization, it is passed to the risk prediction sub-engine; if the score is high, the access control sub-engine is triggered to reduce the privilege and block the transmission channel; otherwise, transmission is allowed through the sharing interface; the entire process is driven by the policy call interface according to the DAG order, and data is passed through the memory context without manual intervention, achieving fully automated closed-loop execution of the compliance process.

[0057] In one embodiment, based on the knowledge graph and core privacy indicator set in the data privacy protection resource library, the semantic similarity value and attribute matching score between each data item in the data sample and known privacy entities are calculated, including: The semantic feature vectors of each known privacy entity are extracted from the knowledge graph, and the attribute feature vectors of each known privacy entity are extracted from the core privacy index set. The semantic feature vectors and attribute feature vectors of each known privacy entity are aggregated into a comprehensive privacy entity feature vector through a feature fusion algorithm.

[0058] The privacy entity comprehensive feature vector can be a high-dimensional joint representation generated by aggregating the semantic feature vectors of privacy entities in the knowledge graph with the corresponding attribute feature vectors in the core privacy indicator set through a feature fusion algorithm. It can be used as a semantic-attribute joint reference model for known privacy entities, used for semantic matching and attribute consistency comparison with data items. In this embodiment, the privacy entity comprehensive feature vector can be formed by concatenating or weighting the semantic embedding vectors of knowledge graph nodes with the attribute description vectors in the core privacy indicator set to form a representation in a unified vector space. The privacy entity comprehensive feature vector serves as a reference benchmark for semantic similarity calculation; its generation depends on the knowledge graph and the core privacy indicator set, and its output is used in the matching process of the data item comprehensive feature vectors.

[0059] Semantic description vectors and attribute description vectors of each data item are extracted from the data sample and aggregated into a comprehensive feature vector of the data item using the same feature fusion algorithm.

[0060] The comprehensive feature vector of a data item can be a high-dimensional joint representation generated by aggregating the semantic description vectors and attribute description vectors of each data item in the data sample using the same feature fusion algorithm. It can be used as a semantic-attribute expression of the data item to be identified and compared with the comprehensive feature vector of the privacy entity to determine privacy attribution. In this embodiment, the comprehensive feature vector of a data item can semantically encode the text content of the data item to generate a semantic description vector, encode structural attributes such as field names, data types, and value distributions into attribute description vectors, and then merge them using a fusion algorithm.

[0061] The semantic similarity value between the comprehensive feature vector of each data item and the comprehensive feature vector of each privacy entity is calculated based on the cosine similarity algorithm, and the attribute matching score between each data item and each known privacy entity is calculated based on the attribute field matching and value range comparison algorithm.

[0062] In this embodiment, based on the cosine similarity algorithm and the attribute field matching and value range comparison algorithm, the cosine similarity of two comprehensive feature vectors can be calculated as the semantic matching score. Simultaneously, the consistency of field names and the conformity of values ​​within a preset range are compared to generate an attribute matching score. Semantic similarity uses cosine similarity, while attribute matching uses exact field name matching plus value regularization validation (e.g., ID card number format) and value range distribution consistency (e.g., age 0-120). A weighted fusion score is introduced: semantic score accounts for 60%, attribute score accounts for 40%, and a threshold combination is set to determine whether a valid match is constituted.

[0063] Taking privacy identification before anonymizing medical data as an example, the system analyzes the chief complaint field in the electronic medical record. Its semantic description vector is encoded by BERT into semantic vectors for persistent chest pain and palpitations, and the attribute vector is text-based and non-unique. It is then fused with the semantic vector of cardiovascular disease symptom entities in the knowledge graph and the symptom description attribute vector in the core indicator set to generate a comprehensive feature vector of the privacy entity. The cosine similarity between the two is calculated to be 0.87, the attribute field matches the symptom description class, and the value is within the range of medical terminology, which is judged as a high-confidence match. Finally, the field is included in the health and medical privacy candidate set, and subsequent sensitivity level assessment and anonymization strategy activation are triggered.

[0064] In one embodiment, after calculating the semantic similarity value based on the cosine similarity algorithm and calculating the attribute matching score based on attribute field matching, the method further includes: Based on a preset weighting strategy, the semantic similarity value and attribute matching score are weighted and fused to generate a comprehensive correlation score between each data item and each known privacy entity.

[0065] The data-privacy entity association matrix (DATA-PRIMARY Matrix) can be used to transform multidimensional scoring relationships into a topological data structure suitable for cluster analysis, supporting the input space for unsupervised clustering. DATA-PRIMARY Matrix can include high-density association matrix sub-regions, sparse association matrix sub-regions, and cross-domain association matrix sub-regions. The operational principle of the DATA-PRIMARY Matrix can be explained in context: it calculates a weighted and fused comprehensive association score between each data item and each privacy entity, and organizes it into a matrix form, setting missing matching items to zero or a low confidence value. The DATA-PRIMARY Matrix can serve as a direct input to density-based clustering algorithms; its structure determines the granularity and coverage of the clustering results, and its score distribution influences the adaptive adjustment direction of the clusters.

[0066] A data-privacy entity association matrix is ​​constructed based on the comprehensive association score, and unsupervised clustering analysis is performed on the data-privacy entity association matrix based on the density clustering algorithm to generate data privacy clustering results.

[0067] The privacy candidate set can be a grouping of data items with high internal correlation and low boundary ambiguity, output by a density clustering algorithm, representing a potential privacy type. Privacy candidate sets can include identity-based candidate sets, health behavior-based candidate sets, financial transaction-based candidate sets, etc. The operating principle of the privacy candidate set can be explained in context: it uses density clustering to perform unsupervised partitioning of the correlation matrix, aggregating data items with a comprehensive correlation score higher than a local density threshold into independent clusters, each cluster being a privacy candidate set.

[0068] Based on the data density distribution characteristics and inter-cluster differences of each cluster in the data privacy clustering results, the clustering parameters are dynamically adjusted and the data clusters are re-divided until the variance of the comprehensive correlation score of the data items within each cluster is lower than the preset threshold, thereby generating a privacy candidate set.

[0069] In this embodiment, semantic similarity values ​​and attribute matching scores are weighted and fused according to a preset weighting strategy to generate a comprehensive relevance score. This can be achieved using a fixed weighting strategy, such as a semantic weight of 0.7 and an attribute weight of 0.3, suitable for compliant scenarios with clearly defined rules; or it can employ a dynamic weighting strategy based on confidence levels, increasing the attribute weight to above 0.6 when the attribute field matching degree is higher than a threshold, and vice versa. Furthermore, this achieves a flexible fusion of semantic and attribute-based dual-dimensional evaluation, avoiding misjudgments caused by a single indicator and improving the robustness of the comprehensive judgment.

[0070] In one specific embodiment, taking the privacy identification of user behavior data in a financial app as an example, the system calculates a comprehensive correlation score for fields such as login time, device model, consumption amount, and browsed pages in the user behavior log, and constructs a data-privacy entity correlation matrix. Unsupervised clustering is performed using DBSCAN, forming three clusters under initial parameters. However, the variance within the login time + device model cluster is 0.12, higher than the threshold of 0.08. The system automatically reduces the neighborhood radius and, after re-clustering, separates a new cluster of device fingerprint correlation, whose members are high-frequency device change records. The comprehensive correlation variance of data items within this cluster decreases to 0.06, satisfying the convergence condition and forming an independent privacy candidate set. This set is not defined in the historical rules but is autonomously identified by the system as a new privacy combination, used to trigger subsequent sensitivity level assessments and enhanced access control.

[0071] In one embodiment, privacy attributes, data characteristics, and processing requirements are input into an intelligent policy generation engine to drive the engine to generate privacy protection policy templates corresponding to each data privacy information and sensitivity level, including: Convert privacy attributes into privacy type description statements, convert data characteristics into data format constraint lists, and convert processing requirements into compliance control specifications; Based on the preset policy template generation rules, the privacy type description statement, data format constraint list, and compliance control specifications are organized into an initial policy definition sequence in a logical order; The intelligent policy generation engine performs syntax parsing and rule validation on the initial policy definition sequence to generate a set of candidate privacy protection policy templates. Using privacy samples as verification data, a simulated privacy protection scheme is generated based on candidate privacy protection policy templates through an intelligent policy generation engine, and the compliance matching degree and protection strength index between the simulated privacy protection scheme and the privacy protection level standard are calculated. Candidate privacy protection strategy templates with compliance matching higher than the first preset threshold and protection strength index higher than the second preset threshold are selected as privacy protection strategy templates corresponding to data privacy information and sensitivity levels.

[0072] The intelligent policy generation engine can be a semantic parsing and verification system that automatically transforms privacy attributes, data characteristics, and processing requirements into executable policy templates. It can be used to automate the conversion of compliance requirements into computable policies and to perform quality screening. The intelligent policy generation engine can receive structured input sequences, perform syntax parsing, rule validation, and empirical evaluation, and output policy templates that meet dual thresholds. The engine relies on compliance control specifications, data format constraint lists, and privacy type description statements to generate an initial sequence, calls privacy samples for verification, and outputs the final privacy protection policy template. The intelligent policy generation engine may include a syntax parser, a rule validator, and an empirical evaluator.

[0073] In this embodiment, privacy attributes are converted into privacy type description statements, data features are converted into data format constraint lists, and processing requirements are converted into compliance control specifications. Privacy attributes can be data attributes containing personally identifiable information, biometric information, financial information, etc., and can be used to describe the sensitivity of the data. Privacy attributes can be converted into privacy type description statements, such as "Data containing personally identifiable information should meet the encryption requirements of the Personal Information Protection Standard." Data features can be attributes such as the structure, length, and format of the data, and can be used to describe the organization of the data. Data features can be converted into a data format constraint list, such as "Telephone numbers must conform to the International Telephone Number Format Standard and cannot be stored in plaintext." Processing requirements can be requirements regarding the operation, purpose, and scope of data processing, and can be used to describe the legal boundaries of data processing. Processing requirements can be converted into compliance control specifications, such as "If the data is used for marketing, it must contain explicit user consent."

[0074] Based on preset policy template generation rules, privacy type description statements, data format constraint lists, and compliance control specifications are logically organized into an initial policy definition sequence. The policy template generation rules can be a preset set of rules used to guide the intelligent policy generation engine in generating policy templates. These rules may include mapping rules for privacy type description statements, mapping rules for data format constraint lists, and mapping rules for compliance control specifications, among others.

[0075] An intelligent policy generation engine performs syntax parsing and rule validation on the initial policy definition sequence to generate a set of candidate privacy protection policy templates. Syntax parsing involves lexical analysis, syntax tree construction, and logical consistency checks on the policy definition sequence to eliminate conflicting or invalid policy combinations. Rule validation checks the consistency of rules within the policy definition sequence to ensure there are no conflicts between them.

[0076] Using privacy samples as validation data, a simulated privacy protection scheme is generated based on candidate privacy protection policy templates through an intelligent policy generation engine. The compliance matching degree and protection strength indicators between the simulated privacy protection schemes and privacy protection level standards are then calculated. The privacy samples can represent typical data instances from a specific set of privacy candidates, used to verify the effectiveness of the policy templates on real data. Privacy samples can include highly sensitive data instances, moderately sensitive data instances, and low-sensitivity data instances.

[0077] Candidate privacy protection strategy templates with a compliance matching degree higher than a first preset threshold and a protection strength index higher than a second preset threshold are selected as privacy protection strategy templates corresponding to data privacy information and sensitivity levels. The compliance matching degree can be a quantitative assessment of the degree of matching between the strategy template and the privacy protection level standard, while the protection strength index can be a quantitative assessment of the strategy template's ability to protect privacy.

[0078] In one embodiment, when locating privacy protection components in data privacy protection-related materials using predefined privacy element classification rules and risk indicator extraction patterns, the method further includes: Identify combinations of privacy elements with fixed relationships in data privacy protection-related materials, and extract the core elements and related elements in the privacy element combinations as privacy element hierarchical labels; Here, the privacy element combination can be a set of multiple interrelated privacy elements describing data privacy protection-related materials. In this embodiment, fixed tripartite combinations such as data subject + consent + purpose limitation are identified through pattern matching and dependency parsing, and these are marked as consent control levels. This abstracts scattered compliance clauses into reusable semantic hierarchical units, realizing the modular deconstruction of complex rules.

[0079] Analyze the attribute characteristics and risk parameters of each element in the privacy element combination, and associate the attribute characteristics and risk parameters with the privacy element hierarchical labels to form a structured privacy element attribute table; Among them, attribute features can be specific attributes describing privacy elements, such as applicable subjects, processing purposes, and storage periods; risk parameters can be descriptions of the risk level of privacy elements, such as exposure scope, dissemination range, and depth of impact. In this embodiment, attributes are extracted from personal identifier elements: subject type (natural person / child), identifier type (ID / biometrics), and storage period (permanent / limited-time), and risk parameters are bound to them: re-identifiability (high) and traceability (medium), thereby transforming vague compliance descriptions into comparable and computable structured attributes to support cross-document consistency verification.

[0080] The study examines the relationships and dependencies among privacy elements in data privacy protection materials and identifies risk transmission chains based on these relationships and dependencies. In this context, the influence relationship can describe the path of interaction between privacy elements, such as triggering, dependency, enhancement, or inhibition relationships; the dependency type can describe the type of mutual dependency between privacy elements, such as triggering dependency, enhancement dependency, etc. In this embodiment, a semantic implication model is used to determine whether data minimization is dependent on the objective, constructing a dependency transmission chain to reveal the non-local propagation mechanism of risk among elements, thus overcoming the limitations of single-point assessment.

[0081] The structured privacy element attribute table and risk transmission chain are mapped to a pre-defined privacy protection framework classification system to generate a multi-level risk relationship network; The preset privacy protection framework classification system can be, for example, GDPR chapters or the NIST framework. In this embodiment, consent control level nodes are mapped to the personal information control chapter of the ISO framework, and trigger transmission chains are mapped to control failure sub-paths, thereby achieving semantic integration of privacy elements from heterogeneous sources under a unified framework and supporting the construction of a global risk view.

[0082] Based on a multi-level risk relationship network, privacy elements described similarly but from different sources in data privacy protection-related materials are normalized to eliminate duplicate definitions. The normalization process can be achieved by calculating semantic similarity and aligning hierarchical levels to merge cross-source element definitions with the same attributes and propagation paths. In this embodiment, a graph embedding model is used to calculate the vector distance between sensitive personal information nodes in different documents. If the distance is below a threshold, they are merged into the same entity, thereby eliminating redundancy and semantic ambiguity in multi-source documents and ensuring consistency and reasonability within the knowledge base. Taking the unified compliance management of multinational enterprises as an example, the system identifies combinations of elements such as sensitive personal information, data subject rights, and cross-border transmission from different regional regulations, extracts consent control levels and cross-border control levels, binds attribute tables to each level (e.g., re-identifiability of biometric information = high), constructs a transmission chain such as no consent → illegal processing → regulatory penalties, maps the three to the ISO27701 framework to generate a multi-level risk network, and finds that although the descriptions of genetic data in the three regulations are different, their attributes and transmission paths are consistent, automatically normalizing them into the same entity. Ultimately, the privacy classification module can identify regional data with unified semantics and trigger consistent desensitization and access control strategies.

[0083] This embodiment provides an AI-based data privacy protection and management system. It constructs hierarchical tags by identifying combinations of privacy elements, structurally binds attribute features and risk parameters to form an attribute table, and generates risk transmission chains by mining the dependencies and influence relationships between elements. Finally, it integrates these two elements within a unified framework to form a multi-level risk relationship network, achieving semantic unification and logical integration of privacy elements with cross-source, heterogeneous, and ambiguous descriptions. This process transforms fragmented and ambiguous compliance clauses into a computable knowledge system with hierarchical structure, attribute quantification, and propagation paths. It upgrades sensitivity level assessment from keyword matching to dynamic inference based on relational reasoning, providing a highly consistent and interpretable semantic foundation for privacy classification and policy engines. This fundamentally supports the system's accurate adaptation and intelligent response in multi-source compliance environments.

[0084] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An artificial intelligence-based data privacy protection management system, characterized by, The system comprises a resource library module, a privacy classification module, a privacy protection module and a privacy management module; The resource library module is configured to construct a data privacy protection resource library based on privacy protection knowledge and information extracted from data privacy protection related materials; The privacy classification module is configured to identify and classify different types of data privacy information and corresponding sensitive levels by intelligently identifying and analyzing the data privacy protection resource library, wherein different sensitive levels correspond to different intensity privacy protection strategies; The privacy protection module is configured to fuse the different types of data privacy information and sensitive levels to construct a core strategy engine of the artificial intelligence-based data privacy protection management system, wherein the core strategy engine comprises a privacy desensitization sub-engine, an access control sub-engine and a privacy risk prediction sub-engine; The privacy management module is configured to generate an artificial intelligence-based data privacy protection management method according to a preconfigured privacy protection process and the core strategy engine, wherein the privacy protection process is used to determine the execution order and coordination mode between each sub-engine in the core strategy engine to complete the privacy protection of the data throughout the life cycle.

2. The artificial intelligence-based data privacy protection management system of claim 1, wherein, The data privacy protection resource library is constructed based on the privacy protection knowledge and information extracted from the data privacy protection related materials, comprising: parsing the obtained initial data privacy protection related materials, and performing format standardization processing on the parsing results of the initial data privacy protection related materials based on a data format conversion tool to generate the data privacy protection related materials; locating privacy protection constituent elements in the data privacy protection related materials through a pre-defined privacy element classification rule and a risk index extraction mode, and determining a core privacy index set of the data privacy protection related materials from the constituent elements based on the appearance frequency, correlation relationship and application scenario information of the privacy elements in the data privacy protection related materials; obtaining privacy entities and relationship rules included in the data privacy protection related materials by using knowledge graph construction technology, and deeply mining the features of the privacy entities and the relationship rules based on semantic web model according to the attribute information and relationship information of the privacy entities, so as to map the privacy information in the data privacy protection related materials to corresponding knowledge graph nodes and edges; associating and integrating the original data of the data privacy protection related materials, the determined core privacy index set and the generated knowledge graph through the unique identifiers of each material in the data privacy protection related materials by means of privacy data association technology, to generate the data privacy protection resource library. 3.The artificial intelligence-based data privacy protection management system of claim 2, wherein The different types of data privacy information and corresponding sensitive levels are identified and classified by intelligently identifying and analyzing the data privacy protection resource library, comprising: based on the knowledge graph and the core privacy index set in the data privacy protection resource library, calculating the semantic similarity value and attribute matching degree score of each data item in the data sample with the known privacy entities, and dividing the data items in the data sample into different privacy candidate sets based on the semantic similarity value and attribute matching degree score; The characteristics and associations of the data items included in each of the privacy candidate sets are analyzed to determine the attribute characteristics and data types of different fields in each of the data items, and key privacy information is extracted by identifying identifiers, sensitive attributes, and associated data elements in each of the data items, the key privacy information being used to determine the privacy sensitivity of different fields in each of the data items; Based on the attribute characteristics, the data types, and the privacy sensitivity, the data items in each of the privacy candidate sets are summarized to generate privacy type definitions of each of the privacy candidate sets, the privacy type definitions being used to reflect the category positioning and protection requirements of the data items in the privacy candidate sets in data privacy protection; Based on the privacy type definitions in each of the privacy candidate sets, the specific composition of the data items in each of the privacy candidate sets, and a preset sensitive level scoring rule, a typical data instance in each of the privacy candidate sets is determined as a privacy sample of each of the privacy candidate sets; According to the privacy type definitions and the privacy samples in each of the privacy candidate sets, a privacy protection level standard of each of the privacy candidate sets is determined from preset privacy protection regulations and standards, the privacy protection level standard being used to determine the sensitive level, protection measures, and processing requirements of each data item in the privacy candidate sets; The consistency of the privacy type definitions, the privacy samples, and the privacy protection level standards corresponding to each data item in each of the privacy candidate sets is referred to to determine different types of data privacy information and corresponding sensitive levels in the data privacy protection resource library corresponding to each of the privacy candidate sets. 4.The artificial intelligence-based data privacy protection management system of claim 3, wherein The different types of data privacy information and sensitive levels are fused to construct a core strategy engine of an artificial intelligence-based data privacy protection management system, including: According to the types and sensitive levels of each of the data privacy information, the privacy attributes, data characteristics, and processing requirements of each of the data privacy information are obtained, and the privacy attributes, data characteristics, and processing requirements are input into an intelligent strategy generation engine to drive the intelligent strategy generation engine to generate a privacy protection strategy template corresponding to each of the data privacy information and sensitive levels; Based on the privacy protection strategy template and the privacy samples of the data privacy information, an initial privacy protection tool is developed and configured again to generate a privacy desensitization sub-engine and an access control sub-engine corresponding to each of the data privacy information and sensitive levels; The privacy protection level standards of each of the data privacy information and sensitive levels are converted into calculable risk assessment indicators and models to construct a risk prediction algorithm corresponding to each of the privacy protection level standards, and a corresponding privacy risk prediction sub-engine is generated based on each of the risk prediction algorithms; By deeply mining the privacy requirements at each stage of the data life cycle and the associations between each of the data privacy information, and based on the full life cycle privacy requirements and the associations, a collaborative workflow of each of the privacy desensitization sub-engine, the access control sub-engine, and the privacy risk prediction sub-engine is determined to generate a strategy coordination engine related to the collaborative workflow; According to the policy coordination engine, the privacy desensitization sub-engine corresponding to each data privacy information and sensitivity level, the access control sub-engine, and the privacy risk prediction sub-engine, the core policy engine of the artificial intelligence-based data privacy protection management system is constructed. 5.The artificial intelligence-based data privacy protection management system of claim 4, wherein According to the pre-configured privacy protection process and the core policy engine, an artificial intelligence-based data privacy protection management method is generated, which includes: By formally describing the privacy protection process, the privacy protection phase nodes, the inter-phase dependency relationships, and the conversion rules between the phase nodes in the privacy protection process are obtained, the privacy protection phase nodes represent specific privacy protection tasks, the inter-phase dependency relationships represent task order, and the conversion rules represent triggering conditions for task initiation or completion; According to the privacy protection phase nodes, the inter-phase dependency relationships, and the conversion rules, the calling order and data interaction mode of each privacy desensitization sub-engine, access control sub-engine, privacy risk prediction sub-engine, and policy coordination engine are determined; Based on the calling order and the data interaction mode, the policy calling interface and the data flow path in the initial privacy protection management platform are configured, and the artificial intelligence-based data privacy protection management method is generated. 6.The artificial intelligence-based data privacy protection management system of claim 3, wherein Based on the knowledge graph and the core privacy indicator set in the data privacy protection resource library, the semantic similarity values and attribute matching scores of each data item in the data sample and the known privacy entities are calculated, which includes: The semantic feature vectors of each known privacy entity are extracted from the knowledge graph, and the attribute feature vectors of each known privacy entity are extracted from the core privacy indicator set. The semantic feature vectors and attribute feature vectors of each known privacy entity are aggregated into a privacy entity comprehensive feature vector through a feature fusion algorithm; The semantic description vectors and attribute description vectors of each data item in the data sample are extracted, and are aggregated into a data item comprehensive feature vector through the same feature fusion algorithm; Based on the cosine similarity algorithm, the semantic similarity values between each data item comprehensive feature vector and each privacy entity comprehensive feature vector are calculated, and based on the attribute field matching and value range comparison algorithm, the attribute matching scores between each data item and each known privacy entity are calculated.

7. The artificial intelligence-based data privacy protection management system of claim 6, wherein, After calculating the semantic similarity values based on the cosine similarity algorithm and the attribute matching scores based on the attribute field matching, the following steps are further included: According to a preset weight allocation strategy, the semantic similarity values and the attribute matching scores are weighted and fused to generate a comprehensive correlation degree score between each data item and each known privacy entity; According to the comprehensive correlation degree score, a data-privacy entity association matrix is constructed, and unsupervised clustering analysis is performed on the data-privacy entity association matrix based on a density clustering algorithm to generate a data privacy clustering result; According to the data density distribution characteristics and inter-cluster difference characteristics of each cluster in the data privacy clustering result, the clustering parameters are dynamically adjusted and the data clusters are re-divided until the comprehensive correlation degree score variance of the data items in each cluster is below a preset threshold, and the privacy candidate set is generated. 8.The artificial intelligence-based data privacy protection management system of claim 4, wherein The privacy attribute, the data feature, and the processing requirement are input into an intelligent policy generation engine to drive the intelligent policy generation engine to generate a privacy protection policy template corresponding to each data privacy information and sensitive level. The privacy attribute is converted into a privacy type description statement, the data feature is converted into a data format constraint list, and the processing requirement is converted into a compliance control specification; Based on a preset policy template generation rule, the privacy type description statement, the data format constraint list, and the compliance control specification are organized in a logical order into an initial policy definition sequence; The initial policy definition sequence is parsed and checked by the intelligent policy generation engine to generate a candidate privacy protection policy template set; The privacy sample is used as verification data, and a simulated privacy protection scheme is generated based on the candidate privacy protection policy template by the intelligent policy generation engine, and the compliance matching degree and protection strength index between the simulated privacy protection scheme and the privacy protection level standard are calculated; A candidate privacy protection policy template with a compliance matching degree higher than a first preset threshold and a protection strength index higher than a second preset threshold is selected as the privacy protection policy template corresponding to the data privacy information and sensitive level. 9.The artificial intelligence-based data privacy protection management system of claim 2, wherein In the process of locating the privacy protection constituent elements in the data privacy protection related materials through the pre-defined privacy element classification rule and risk index extraction mode, the following steps are further included: Identify the combination of privacy elements with fixed association relationship in the data privacy protection related materials, extract the relationship between the core elements and the associated elements in the combination of privacy elements as the privacy element hierarchical label; Analyze the attribute characteristics and risk parameters of each element in the combination of privacy elements, and associate the attribute characteristics and the risk parameters to the privacy element hierarchical label to form a structured privacy element attribute table; Detect the influence relationship and dependency type between the privacy elements in the data privacy protection related materials, and divide the risk transmission chain of the privacy protection constituent elements according to the influence relationship and dependency type; Map the structured privacy element attribute table and the risk transmission chain to a preset privacy protection framework classification system to generate a multi-level risk relationship network of the privacy protection constituent elements; Based on the multi-level risk relationship network, normalize the privacy elements described similarly but differently in the data privacy protection related materials to eliminate the repeatedly defined privacy elements.

Images