CN-121980581-A - Easy-to-count model management system based on dynamic authority control and automatic deployment

CN121980581ACN 121980581 ACN121980581 ACN 121980581ACN-121980581-A

Abstract

A system for managing an easily-counted model based on dynamic authority control and automatic deployment belongs to the field of data management and artificial intelligence intersection, and comprises a front-end interaction layer, a core service layer, a data persistence layer and an external integration layer, wherein the front-end interaction layer is used for providing an interaction interface between a user and the system, the core service layer is used for realizing model classification, version control, metadata management, dynamic allocation of access authorities, automatic model deployment and data management model adaptation, the data persistence layer is used for realizing storage of model files and metadata and storage and quick retrieval of unstructured models, and the external integration layer is used for realizing unified management of model assets and automatic triggering of model deployment in a butt joint mode with a data management platform. The invention improves the model deployment efficiency, reduces the sensitive data leakage risk, improves the system expandability and the model multiplexing rate by integrating dynamic authority control, automatic deployment and model classification storage technologies, and is suitable for the scenes of financial management, government data sharing, medical main data management and the like.

Inventors

REN HAO
LIANG XUESONG
LIU ZHANZHU
BAI XIN
ZHAO SHUANG

Assignees

长春市万易科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251226

Claims (10)

1. An easy-to-count model management system based on dynamic rights control and automated deployment, comprising: A front-end interaction layer for providing an interaction interface between a user and a system; The core service layer is used for realizing model classification, version control and metadata management, dynamically distributing access rights, realizing automatic deployment of the model and realizing adaptation of a data management model; The data persistence layer is used for realizing the storage of the model file and the metadata and realizing the storage and the quick retrieval of the unstructured model; and the external integration layer is used for interfacing with the data management platform to realize unified management of model assets and realizing automatic triggering of model deployment.
2. The system is characterized in that the front-end interaction layer comprises a model management panel, a permission configuration workbench and a monitoring instrument panel, wherein the model management panel is used for displaying model classification tree diagrams and screening data management models according to different dimensions, the permission configuration workbench is used for visually configuring mapping relations between user roles and data sensitivity, and the monitoring instrument panel is used for displaying model deployment states in real time and generating index reports.
3. The system for managing the easy-to-count model based on dynamic authority control and automatic deployment according to claim 1, wherein the core service layer comprises a model classification and storage engine module, a dynamic authority control engine module, an automatic deployment service module and a multi-mode data processing adapter module, wherein the model classification and storage engine module is used for classifying, version controlling and metadata managing a data management model, the dynamic authority control engine module is used for dynamically distributing access authorities according to user roles, data sensitivity and user behaviors, the automatic deployment service module is used for integrating with a data management flow to achieve automatic deployment after model update, and the multi-mode data processing adapter module is used for achieving management model adaptation of structured data and unstructured data.
4. The dynamic authority control and automation deployment-based easy-number model management system according to claim 3, wherein the model classification and storage engine module comprises a model classification module, a model version control module and a metadata management module, the model classification module automatically classifies models based on metadata by adopting a decision tree algorithm, the model version control module realizes model version rollback and branch management by adopting a Git-like version control mechanism, and the metadata management module stores model metadata by using an elastic search and supports full-text retrieval and association analysis.
5. The system for managing the easy-to-count model based on dynamic authority control and automatic deployment according to claim 3, wherein the dynamic authority control engine module comprises a hybrid control model and a real-time decision engine module, the hybrid control model combines RBAC and ABAC, the RBAC defines a basic role and distributes default access authorities, the ABAC dynamically adjusts the access authorities according to data sensitivity and user behaviors, and the real-time decision engine module adopts a Redis cache authority strategy to evaluate access authority requests in real time through a rule engine.
6. The system for managing the easy-to-count model based on dynamic authority control and automatic deployment according to claim 3, wherein the automatic deployment service module comprises an API interface and an automatic deployment pipeline, the API interface adopts a RESTful API for interfacing with an external system, and the automatic deployment pipeline integrates Jenkins or GitLab CI to realize full-flow automation of model verification, environment configuration and online release.
7. The system for managing the easy-to-count model based on dynamic authority control and automatic deployment according to claim 3, wherein the multi-mode data processing adapter module comprises a structured data processing module and an unstructured data processing module, the structured data processing module is connected with a relational database through JDBC/ODBC to automatically generate SQL scripts to execute data cleaning and classification, and the unstructured data processing module integrates OCR and NLP models to extract text key information and generate structured metadata.
8. The system for managing the easy-to-count model based on dynamic authority control and automatic deployment according to claim 1 is characterized in that the data persistence layer comprises a model storage module and an authority policy storage module, wherein the model storage module is used for storing model files and metadata by adopting MongoDB, supporting storage and quick retrieval of unstructured models, and the authority policy storage module is used for storing mapping relations and audit logs between user roles and data sensitivity by adopting MySQL, so that ACID transaction consistency is ensured.
9. The system for managing the easily-counted model based on dynamic authority control and automatic deployment according to claim 1, wherein the external integration layer comprises a data management tool integration module and a CI/CD pipeline integration module, the data management tool integration module is in butt joint with a data management platform through an API to realize unified management of model assets, and the CI/CD pipeline integration module is integrated with Jenkins and GitLab CI to support automatic triggering of a model automatic deployment pipeline.
10. An easy-to-count model management method based on dynamic authority control and automatic deployment realized by adopting the easy-to-count model management system based on dynamic authority control and automatic deployment as claimed in any one of claims 1-9, which is characterized by comprising the following steps: Step S1, uploading a model; Uploading a client box-dividing model by a data engineer through a front-end interaction layer, screening a data treatment model by a model management panel, automatically classifying the data treatment model by a model classification module by adopting a decision tree algorithm, and generating a version number; S2, configuring authorities; Defining a basic role and distributing default access rights through a hybrid control model, and dynamically adjusting the access rights according to the data sensitivity and user behaviors; step S3, deployment triggering; after the model passes the unit test, the automatic deployment service module calls an API interface to push the model to the production environment data management cluster through an automatic deployment pipeline; s4, treating in real time; after the client data enter the system, the model automatically executes the box-dividing operation, the real-time decision engine module monitors the access behavior in real time, and the access permission request is evaluated in real time by adopting the Redis cache permission strategy through the rule engine, so that the large-batch data export request in irregular time is blocked.

Description

Easy-to-count model management system based on dynamic authority control and automatic deployment Technical Field The invention belongs to the technical field of data management and artificial intelligence intersection, and particularly relates to an easy-to-count model management system based on dynamic authority control and automatic deployment. Background With the advancement of digital economy wave, data has become an enterprise core asset. The traditional data management mode is difficult to cope with complex scenes with unstructured data accounting for over 80 percent due to lack of intelligent means. For example, the financial industry needs to analyze customer behavior data in real time to prevent fraud, while the medical industry needs to implement cross-institution data collaboration while protecting patient privacy. In this context, AI-driven data management techniques (e.g., automated cleaning, intelligent blood-margin analysis) are becoming the most demanding, and standardization and dynamics of model management are key to achieving data value release. The current data management system has the following major core defects: (1) Model management fragmentation, namely focusing on data integration and cleaning by a traditional data management system, and not establishing a unified framework of model classification storage, dynamic authority control and automatic deployment. For example, data management models such as financial wind control rules and medical diagnosis algorithms are distributed in different systems, so that updating and maintaining costs are high. (2) The deployment flow is low-efficiency and error frequent, and the manual deployment is easy to cause environmental configuration errors. For example, a certain e-commerce enterprise rolls back a data management model due to version confusion, so that service is interrupted for a plurality of hours, and the workload of authority configuration is increased. (3) Rights control statization most data governance systems employ role-based access control (RBAC) and cannot dynamically adjust rights based on data sensitivity (e.g., public/secret) or user behavior. On the one hand, permission solidification easily causes security holes, and a static RBAC strategy cannot adapt to a dynamic scene. For example, when a bank analyst derives customer data in batches at night, the system does not trigger secondary authentication, resulting in data leakage. Meanwhile, when an analyst processes highly sensitive customer data, the system may still give the data export authority to the analyst, which causes leakage risk. In addition, GDPR and other regulations require that data governance should achieve full life cycle traceability, and traditional data governance systems lack dynamic authority audit functions. Compliance audit is costly, and static rights logs are difficult to meet real-time audit requirements of GDPR and other regulations. For example, a national business may employ a full-time team for data access record analysis, costing millions of dollars each year. (4) In the conventional data management system, the model update needs manual triggering deployment and cannot be integrated with a CI/CD pipeline. For example, a certain bank data management team needs to take 2 hours to complete model verification, environment configuration and online operation, and the service response speed is seriously affected. (5) The multi-mode data processing capability is insufficient, the traditional data management system only supports structured data management, and the non-structured data (such as contract text and medical images) is lack of classification and quality detection means. For example, some legal technology company needs to manually review the key terms of the contract, which is inefficient. (6) The model multiplexing rate is low, namely the traditional data management system strongly binds the data management model with the service system, so that the model multiplexing rate is low. For example, a government platform needs to maintain a data quality detection model for each department separately, which is costly. Disclosure of Invention The invention provides an easy-to-count model management system based on dynamic authority control and automatic deployment, which aims to solve the problems of low model management and deployment efficiency, static authority control and manual dependency of deployment flow in the existing data management system. The invention improves the model deployment efficiency, reduces the sensitive data leakage risk, improves the expandability of the system, improves the model multiplexing rate and is suitable for scenes requiring high safety and high availability such as financial wind control, government data sharing, medical main data management and the like by integrating dynamic authority control, automatic deployment and model classification storage technologies. The data management technology comprises data classif