An AI-driven intelligent metadata analysis method

Abstract

The application discloses an AI-driven intelligent metadata analysis method, which comprises the following steps: multi-source non-invasive discovery, AI semantic automatic labeling, sensitive data deep sniffing, intelligent blood relationship splicing, data quality preliminary evaluation, incremental perception and version comparison, open API and low-code integration. On the basis of fusing AI technology, the system is comprehensively constructed from the aspects of database structure design, algorithm module development, user interface development, system deployment and integration, and the like, so that the efficiency, stability and expansibility of the system are ensured.

Description

Technical Field

[0001] This invention relates to the field of digital transformation, and in particular to an AI-driven intelligent metadata analysis method. Background Technology

[0002] In the current wave of digital transformation, AI-driven data governance is at a crucial turning point. This turning point is of extraordinary significance, clearly indicating that data governance is moving rapidly from the traditional "tool-assisted" model to a more advanced and intelligent "cognitive-driven" stage. In its continued development, the role of this field will undergo profound changes. In the past, it was largely seen as a set of automated tools for tasks such as data cleaning and metadata management, playing a relatively basic and limited role in data governance. However, in the future, it will gradually evolve and upgrade, ultimately transforming into a management system with powerful cognitive and decision-making capabilities.

[0003] Existing metadata management technical solutions can be summarized into a five-step architecture: "collection → storage → processing → service → application," as follows: I. Acquisition Layer Active data collection: Retrieves database dictionaries, API definitions, log header information, etc. periodically via JDBC / Agent; Passive reception: Message queue (Kafka / Pulsar) or Hook method, pushed in real time by the application system when changes occur; II. Storage and Modeling Central Metadata Repository: The logical model adopts the CWM / XMI standard, and the physical database can be MySQL, PostgreSQL or distributed NewSQL; III. Treatment and Management Cleaning and standardization: unify units, naming, and data element definitions, and automatically remove invalid fields; Calculation of the influence of bloodline: Extracting field-level lineage based on SQL parsing and AST syntax tree; Combine with DAG scheduling systems (Airflow, DolphinScheduler) to generate end-to-end data flow graphs; Quality inspection: PDCA closed loop, regularly benchmarking against data standards, and outputting scores and correction suggestions.

[0004] IV. Service Layer Metadata service API: RESTful / gRPC provides unified interfaces for directory, search, permissions, and version control. Real-time change broadcasting: Utilize an event bus (such as Spring Cloud Bus) to synchronize structural changes to downstream consumer systems in real time; Access Control and Auditing: Integrates with Enterprise IAM to achieve field-level masking and access log auditing.

[0005] Disadvantages of existing technology: Existing technologies have significant shortcomings in many aspects of bank indicator configuration and analysis, which seriously restricts banks' comprehensive and in-depth understanding and effective management of their own operations.

[0006] 1) Limited coverage, unable to span the entire data lifecycle. Problem: Traditional metadata management systems often focus on downstream systems such as ODS and data warehouses, with insufficient monitoring of source systems, edge devices, and streaming data. As a result, problems often only become apparent when the data enters the analysis stage, leading to high repair costs.

[0007] 2) Heavy reliance on manual labor, resulting in both low efficiency and low accuracy. Metadata lineage analysis relies on manual code analysis, which is time-consuming and prone to omissions.

[0008] The cleaning process is automated: manual correction of outliers accounts for 30%, with an error rate of 5%.

[0009] Consequences: Human resource costs account for over 60% of the total investment in the metadata management system, and it is unable to cope with the exponential growth of data volume. 3) Out of touch with business needs, becoming an "IT island" Low business involvement: The DQM rules designed by the IT department focus on technical dimensions (such as NULL value ratio <1%), but the business department is more concerned about business logic errors (such as "high-value customer orders are submitted repeatedly").

[0010] Lack of feedback loop: After data quality issues were fixed, business rules were not updated synchronously, leading to repeated occurrences of similar problems.

[0011] Consequence: After a retail company's DQM system went live, the stockout rate increased by 15% due to incorrect inventory data caused by the failure to integrate with the procurement process. Summary of the Invention

[0012] In view of the above problems, the present invention is proposed to provide an AI-driven intelligent metadata analysis method that overcomes or at least partially solves the above problems.

[0013] According to one aspect of the present invention, an AI-driven intelligent metadata analysis method is provided, the data analysis method comprising: Multi-source, non-intrusive detection; AI-powered automatic semantic annotation; Deep sniffing of sensitive data; Intelligent bloodline matching; Preliminary data quality assessment; Incremental awareness and version comparison; Open API & low-code integration.

[0014] Optionally, the multi-source non-intrusive discovery specifically includes: Probe types: JDBC direct connection, log CDC, agentless network segment sniffing, object storage event subscription; Data source; Discovery granularity: Database → Table → Field → Partition → Storage path → API endpoint.

[0015] Optionally, the data source specifically includes: Oracle, SQL Server, MySQL, PostgreSQL, Hive, Spark, Kafka, S3, OSS, API, and Excel.

[0016] Optionally, the AI ​​semantic automatic annotation specifically includes: A domain-fine-tuned large model is used to perform three-modal fusion inference of field name, sample value, and annotation, and output the business meaning, label, and sensitivity level. It has a built-in dictionary of 2000+ industry terms and supports the unification of synonyms / abbreviations.

[0017] Optionally, the deep sniffing of sensitive data specifically includes: Dual-layer scanning: shallow layer completes in 30 seconds; deep layer + regularization + NER model, recognizing various entities, including mobile phone numbers, ID cards, bank cards, addresses, and medical records; Automatically match desensitization algorithms and output desensitization strategy recommendations.

[0018] Optionally, the intelligent bloodline splicing specifically includes: Real-time parsing of SQL / ETL scripts, supporting stored procedures, views, temporary tables, and multi-segment SQL; Log side: Reverse completion of Query-log solves black-box ETL scenarios; Graph database storage.

[0019] Optionally, the initial data quality assessment specifically includes: Built-in 200+ rule templates; Large-scale model anomaly pattern detection: Identify latent anomalies such as sales = 0 and inventory decline without writing rules.

[0020] Optionally, the incremental awareness and version comparison specifically include: Based on dual verification of timestamp and CRC, it automatically identifies four types of changes: addition, deletion, renaming, and type change. Changes are stored in a Git-style version chain, supporting one-click rollback of historical metadata versions.

[0021] Optionally, the open API & low-code integration specifically includes: It uses both REST and GraphQL protocols and provides a complete set of interfaces for "scan task creation - status query - result callback". Output formats: OpenMetadataStandard, DataHubAVRO, and custom JSON, facilitating integration with Collibra, DataWorks, and Purview; Low-code Flow: Scan complete → Automatically trigger quality scoring → Automatically enter asset catalog → Automatically notify responsible person.

[0022] This invention provides an AI-driven intelligent metadata analysis method, which includes: multi-source non-intrusive discovery; AI semantic automatic annotation; deep sniffing of sensitive data; intelligent lineage splicing; initial data quality assessment; incremental awareness and version comparison; and open API & low-code integration. Based on the integration of AI technology, it comprehensively constructs the system from aspects such as database structure design, algorithm module development, user interface development, and system deployment and integration, ensuring the system's efficiency, stability, and scalability.

[0023] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart illustrating an AI-driven intelligent metadata analysis method provided in an embodiment of the present invention. Detailed Implementation

[0026] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0027] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0028] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0029] The technical problems to be solved by this invention include: Intelligent metadata scanning Based on NLP and machine learning models, AI can automatically scan data lakes, data warehouses, code, and log files, proactively discovering and associating technical and business metadata to build a dynamically updated "data asset map." Multimodal large models understand unstructured data such as text, images, and voice, achieving more accurate metadata association.

[0030] Intelligent bloodline analysis Leveraging natural language processing (NLP) technology and machine learning algorithms, the system can automatically parse stored procedures. During this process, the program deeply analyzes the stored procedure's code structure, logical relationships, and data operation patterns to identify key information. Subsequently, based on the extracted content, the system further constructs the relationships between data, thereby generating a complete lineage analysis report that clearly shows the data's source, flow path, and interdependencies.

[0031] Data Asset Map A data asset catalog or data asset panorama is a tool for visualizing, managing, and governing an enterprise's data assets. It displays key information such as the distribution, relationships, quality, responsible parties, and usage permissions of all data assets within an enterprise in a graphical or structured manner, helping enterprises achieve the goal of making their data assets "visible, locatable, manageable, and effectively used."

[0032] Metadata Intelligent Standardization AI automatically constructs data standards and generates quality rules by understanding structured and unstructured data, reducing the time-consuming process of manually defining standards. For example, the Dataphin platform uses AI to automatically generate standard attributes for fields such as ID card numbers, making the failure to meet data standards a "barometer" of quality.

[0033] like Figure 1 As shown, intelligent metadata is primarily a three-dimensional system based on "AI-driven, cloud-native, and real-time lineage." It undergoes intelligent upgrades around the five stages of "collection-storage-computation-application-governance," with the following core components and technology selections: I. Overall Architecture (Four Layers, Five Repositories) 1) Access Layer Multi-connector: Supports 30+ data sources (Oracle, SQL Server, Hive, Kafka, S3, OSS, API).

[0034] Real-time probe: JDBC / Log-based CDC + Agentless crawler, discovering new databases and tables in minutes.

[0035] 2) Intelligent Parsing Layer (AI-Parser) Large model semantic understanding: Automatically maps order_amt to "order amount", with a label accuracy of ≥95%.

[0036] Fingerprint lineage analysis: SQL / ETL / stored procedure syntax tree reconstruction, automatic field-level lineage concatenation, supporting stored procedures, views, and temporary tables.

[0037] Metadata "Five Repositories" Technical metadata database: Neo4j / GraphScope lineage; OpenMetadata standard API.

[0038] Business terminology database: knowledge graph + word vectors, supporting automatic normalization of synonyms and domain terms.

[0039] Management strategy library: data classification, data anonymization, and compliance strategies (GDPR / China PIA).

[0040] Quality rule library: 200+ built-in rule templates, supporting the generation of personalized rules from large models.

[0041] AI Feature Library: Records sample labels and feature derivation logic for reuse in subsequent model training.

[0042] 3) Service and Consumption Layer Semantic search: Vector retrieval + inverted index, enabling "I want the sales figures for the last 7 days" → automatic location of tables and fields.

[0043] Impact Analysis: Before the change, assess downstream reports / APIs / models and return an impact list within seconds.

[0044] Data Asset Mall: One-click subscription to data products, with automatic configuration of permissions and de-identification strategies.

[0045] II. Key Technologies of AI + Automation Automatic data cataloging Unsupervised clustering + NER identifies sensitive fields and automatically generates triples of "column name - business meaning - sensitivity level" with a recall rate of 99.9%.

[0046] Intelligent Quality Engine Large-scale model anomaly detection: By using joint temporal and semantic modeling, we discovered hidden quality issues such as "sales volume is 0 but inventory is reduced", improving detection efficiency by 8 times.

[0047] Dynamic lineage tracking Based on dual-channel completion using query logs and ETL scripts, it supports fuzzy matching for field renaming, with a lineage integrity rate of >98%.

[0048] Chatbot (self-service governance robot) Natural language interaction completes the entire process of "adding new business terms - approval - release", reducing the approval time from days to hours.

[0049] III. Deployment and Expansion Modes Cloud-native: Kubernetes + Serverless index (such as Alibaba Cloud TablestoreSearchIndex), storage and computing can be horizontally scaled on demand, and a single instance supports 1 billion metadata records.

[0050] Hybrid cloud: Core metadata is hot-backed locally, and public cloud is used for computing acceleration to meet the compliance requirements of finance and government.

[0051] SaaS Zero Maintenance: Provides multi-tenant isolation, OAuth2 / SSO, API gateway rate limiting, and is ready to use out of the box.

[0052] IV. Typical Implementation Path (90 days) D1-30 Connection + Discovery: Connect to major business systems and output the first data asset list; D31-60 Semantics + Lineage: Complete the semantic annotation of core table fields and generate field-level lineage graphs; D61-90 Governance + Operation: Launch quality scoring, impact analysis, and self-service search, and incorporate them into SLA assessment.

[0053] V. Optional Commercial / Open Source Combinations Commercial: Collibra + Google Dataplex Universal Catalog, Alibaba Cloud DataWorks Intelligent Meta Module.

[0054] Open source: ApacheAtlas + DataHub + Airflow + Neo4j, combined with self-developed AI-Parser.

[0055] Hybrid approach: Smartbi metadata suite serves as the semantic layer, while the underlying Atlas handles lineage storage.

[0056] Through the integration of "AI semantic understanding + graph computing + cloud-native elasticity", enterprises can achieve a sustainable and evolving intelligent metadata management system that enables data to be "findable, understandable, trustworthy, and quick to use".

[0057] 1. Design of AI-driven metadata analysis The intelligent block data management functionality of this invention aims to create a highly intelligent, user-friendly, and flexibly adaptable platform by integrating AI technology to meet comprehensive metadata management needs and unique business characteristics. Specifically, it covers the following aspects: Multi-source non-intrusive detection Probe types: JDBC direct connection, log CDC, agentless network segment sniffing, object storage event subscription.

[0058] Data sources: 40+ types (Oracle, SQL Server, MySQL, PostgreSQL, Hive, Spark, Kafka, S3, OSS, API, Excel, etc.).

[0059] Discovery granularity: Database → Table → Field → Partition → Storage path → API endpoint.

[0060] Metrics: Full scan of 10,000 tables and 100,000 fields in a single instance ≤ 8 minutes; Incremental event detection ≤ 30 seconds.

[0061] AI-generated semantic annotation The domain-fine-tuned large model (BERT-Fin / GPT-4-Fin) is used to perform trimodal inference of "field name + sample value + annotation" and output "business meaning + label + sensitivity level".

[0062] It has a built-in dictionary of 2000+ industry terms (finance, telecommunications, manufacturing, government affairs) and supports the unification of synonyms / abbreviations.

[0063] Metrics: Top-1 business meaning prediction accuracy ≥ 95%; sensitive field recall rate ≥ 99.9% (PII, GDPR, national cryptography).

[0064] Deep sniffing of sensitive data Dual-layer scanning: Shallow layer (column names only) completes in 30 seconds; Deep layer (sampling 1000 rows of content) + regular expression + NER model, recognizing 30 types of entities such as mobile phone numbers, ID cards, bank cards, addresses, and medical records.

[0065] Automatically match de-identification algorithms (mask, hash, mock, k-anonymity) and output a "De-identification Strategy Recommendation".

[0066] Metrics: Deep scan of 100 million rows of data ≤ 6 minutes; false alarm rate ≤ 1%.

[0067] Intelligent bloodline splicing Real-time SQL / ETL script parsing (Antlr4 syntax tree), supporting stored procedures, views, temporary tables, and multi-segment SQL.

[0068] Log side: Reverse completion of Query-log solves the "black box ETL" scenario.

[0069] Graph database storage (Neo4j / GraphScope) supports second-level impact analysis of 1 billion relationships.

[0070] Metrics: Field-level lineage integrity rate ≥ 98%; Change impact query time ≤ 2 seconds.

[0071] Preliminary assessment of data quality Built-in 200+ rule templates (completeness, uniqueness, consistency, timeliness, business logic).

[0072] Large-scale model anomaly pattern discovery: latent anomalies such as "sales = 0 and inventory decrease" can be identified without writing rules.

[0073] Metrics: Quality score can be output on the first scan; rule automatic generation coverage ≥70%.

[0074] Automatic association of business terms By combining the Business Glossary with semantic similarity and knowledge graph reasoning, automatic binding of "table / field → term" is achieved with a binding success rate of ≥90%.

[0075] Supports multiple languages ​​(Chinese, English, French) and industry abbreviations (such as AML = Anti-Money Laundering).

[0076] Incremental awareness and version comparison Based on timestamp + CRC dual verification, it automatically identifies four types of changes: "addition, deletion, renaming, and type change".

[0077] Changes are stored in a Git-style version chain, supporting one-click rollback of historical metadata versions.

[0078] Metrics: Incremental event capture latency ≤15s; Change notification API reachability 99.9%.

[0079] Open API & Low-code Integration It uses both REST and GraphQL protocols and provides a complete set of interfaces for "scan task creation - status query - result callback".

[0080] Output formats: OpenMetadataStandard, DataHubAVRO, and custom JSON, facilitating integration with Collibra, DataWorks, Purview, etc.

[0081] Low-code Flow: Scan complete → Automatically trigger quality scoring → Automatically enter asset catalog → Automatically notify responsible person.

[0082] The technical solution of this invention, based on the integration of AI technology, is comprehensively constructed from aspects such as database structure design, algorithm module development, user interface development, and system deployment and integration, ensuring the system's efficiency, stability, and scalability. Database structure design: Utilizing Gravitino's three-level namespaces, MOF metamodel extensions, and Serverless multidimensional indexes, it achieves a complete closed loop encompassing "AI semantics + field-level lineage + quality / compliance / version". Adopting the "Meta-Meta-Model (MOF)" approach, all attributes are broken down into EAV structures, supporting zero-code extensions of any new type. The three-level namespace structure—Catalog→Schema→Entity—is consistent with the GravitinoFileset concept, facilitating future integration with multi-cloud object storage.

[0083] Algorithm Module Development: This algorithm module is built upon RAG (Retrieval Augmentation) technology and LLM (Large Language Model, specifically the GPT-4-turbo version). Its main function is to implement the tasks related to the metadata chatbot. Specifically, RAG technology combines the retrieval capabilities of external knowledge sources with the text generation capabilities of the language generation model, enabling the model to better utilize information from external knowledge bases when generating responses, thereby improving the accuracy, richness, and relevance of the response content to the input question. GPT-4-turbo, as a currently advanced large language model, possesses powerful natural language understanding and generation capabilities, providing high-quality language processing support for the metadata chatbot. The combination of these two technologies ensures that this metadata chatbot can accurately understand the user's input question, retrieve relevant information from a wide range of metadata resources, and generate responses in a natural and fluent manner to meet the user's query needs.

[0084] User Interface Development: Utilizing cutting-edge web front-end technologies (such as Vue and Element-UI) combined with AI interaction technologies, we develop powerful and user-friendly user interfaces. We leverage computer vision technology to achieve natural interaction modes such as gesture recognition and facial expression recognition, enhancing the user experience. Based on AI image generation and visualization technologies, we automatically generate high-quality, personalized visualization charts and data display interfaces according to user needs, enhancing the intuitiveness and readability of data. Simultaneously, by conducting real-time monitoring and analysis of user interface operations, we use machine learning algorithms to optimize interface layout and interaction processes, improving user efficiency and satisfaction. Furthermore, we integrate AI customer service functions, using NLP technology to answer user questions in real time, providing operational guidance and assistance.

[0085] System Deployment and Integration: Utilizing containerization technologies (such as Docker) and microservice architecture, the system is deployed on cloud platforms (such as Alibaba Cloud and Tencent Cloud) to achieve elastic scaling and high availability. Simultaneously, automated deployment tools (such as Kubernetes) enable rapid deployment, updates, and rollbacks of the system.

[0086] System integration solution: Interface Design and Development: Conduct interface design and development work for interfacing with existing systems, including data transmission interfaces, authentication and authorization interfaces, etc., to achieve seamless integration with existing systems and ensure the security and reliability of data exchange between systems.

[0087] Deployment and Testing: During deployment, it is essential to ensure compatibility of the system's interfaces and data formats with other systems to enable data sharing and interoperability. Subsequently, the system will be deployed to a test environment for comprehensive testing, including functional testing, performance testing, and security testing, to ensure the system's stability and reliability.

[0088] Deployment and Maintenance: After testing is completed, deploy the system to the production environment, monitor the system's operation, promptly handle and fix faults, and regularly update and optimize system functions.

[0089] Beneficial effects: A flexible and adaptable interface and a rich and diverse indicator system: A meticulously designed, highly flexible user interface and a comprehensive indicator selection library form the cornerstone of achieving universal derivative indicator configuration functionality, and represent a significant difference between this invention and existing technologies. Leveraging AI-driven interface interaction optimization, the system intelligently adjusts the interface layout based on user behavior analysis and machine learning algorithms, prominently displaying frequently used functions and indicators, greatly enhancing ease of use. Simultaneously, utilizing natural language processing technology, users can describe their business needs in natural language, and the system accurately understands and recommends suitable indicator options, achieving a personalized and intelligent indicator configuration experience that comprehensively meets the diverse needs of different banks in terms of management models, business priorities, and development stages.

[0090] Intelligent metadata management: It provides a comprehensive suite of methods for generating various types of derived indicators, ranging from basic methods based on simple arithmetic operations to advanced methods utilizing complex statistical and risk assessment models. In simple arithmetic operations, AI-driven dynamic precision control and anomaly prediction mechanisms are introduced to ensure high accuracy and stability of calculation results. For complex models, machine learning algorithms are used to achieve intelligent model selection, automatic parameter optimization, and in-depth interpretation of results, enabling the system to generate more in-depth and insightful derived indicators tailored to various business scenarios. For example, through deep learning of massive amounts of historical and real-time business data, the system can automatically match the most suitable complex model and adjust model parameters in real time according to business changes, providing banks with comprehensive and accurate indicator support for risk assessment and business decision-making.

[0091] A scientifically efficient technical implementation solution: Based on NLP and machine learning models, AI can automatically scan data lakes, data warehouses, code, and log files, proactively discovering and associating technical metadata with business metadata, and building a dynamically updated "data asset map." Multimodal large models understand unstructured data such as text, images, and voice, achieving more accurate metadata association.

[0092] This invention, by providing an intelligent block data management solution, successfully overcomes the bottlenecks of existing technologies, offering bank management a powerful, accurate, and efficient metadata management tool. The key features and mutual protection points of this invention fully demonstrate its innovation and practicality. In the context of the rapid development of financial technology, it is of great significance for promoting business innovation and improving management efficiency in the banking industry, and therefore deserves full legal protection.

[0093] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-driven intelligent metadata analysis method, characterized in that, The data analysis methods include: Multi-source, non-intrusive detection; AI-powered automatic semantic annotation; Deep sniffing of sensitive data; Intelligent bloodline matching; Preliminary data quality assessment; Incremental awareness and version comparison; Open API & low-code integration.

2. The AI-driven intelligent metadata analysis method according to claim 1, characterized in that, The multi-source non-intrusive detection specifically includes: Probe types: JDBC direct connection, log CDC, agentless network segment sniffing, object storage event subscription; Data source; Discovery granularity: Database → Table → Field → Partition → Storage path → API endpoint.

3. The AI-driven intelligent metadata analysis method according to claim 2, characterized in that, The data sources specifically include: Oracle, SQL Server, MySQL, PostgreSQL, Hive, Spark, Kafka, S3, OSS, API, and Excel.

4. The AI-driven intelligent metadata analysis method according to claim 1, characterized in that, The AI-based automatic semantic annotation specifically includes: A domain-fine-tuned large model is used to perform three-modal fusion inference of field name, sample value, and annotation, and output the business meaning, label, and sensitivity level. It has a built-in dictionary of 2000+ industry terms and supports the unification of synonyms / abbreviations.

5. The AI-driven intelligent metadata analysis method according to claim 1, characterized in that, The deep sniffing of sensitive data specifically includes: Dual-layer scanning: shallow layer completes in 30 seconds; deep layer + regularization + NER model, recognizing various entities, including mobile phone numbers, ID cards, bank cards, addresses, and medical records; Automatically match desensitization algorithms and output desensitization strategy recommendations.

6. The AI-driven intelligent metadata analysis method according to claim 1, characterized in that, The intelligent bloodline splicing specifically includes: Real-time parsing of SQL / ETL scripts, supporting stored procedures, views, temporary tables, and multi-segment SQL; Log side: Reverse completion of Query-log solves black-box ETL scenarios; Graph database storage.

7. The AI-driven intelligent metadata analysis method according to claim 1, characterized in that, The initial data quality assessment specifically includes: Built-in 200+ rule templates; Large-scale model anomaly pattern detection: Identify latent anomalies such as sales = 0 and inventory decline without writing rules.

8. The AI-driven intelligent metadata analysis method according to claim 1, characterized in that, The incremental awareness and version comparison specifically include: Based on dual verification of timestamp and CRC, it automatically identifies four types of changes: addition, deletion, renaming, and type change. Changes are stored in a Git-style version chain, supporting one-click rollback of historical metadata versions.

9. The AI-driven intelligent metadata analysis method according to claim 1, characterized in that, The open API & low-code integration specifically includes: It uses both REST and GraphQL protocols and provides a complete set of interfaces for "scan task creation - status query - result callback". Output formats: OpenMetadataStandard, DataHubAVRO, and custom JSON, facilitating integration with Collibra, DataWorks, and Purview; Low-code Flow: Scan complete → Automatically trigger quality scoring → Automatically enter asset catalog → Automatically notify responsible person.

Images