Intelligent file management system based on RAG and metadata full life cycle management

Abstract

The invention relates to an intelligent file management system based on RAG and metadata full-life-cycle management, which is characterized by comprising a front-end application layer, a back-end service layer, an RAG knowledge base analysis layer and a data storage layer, and all the layers work cooperatively through standardized interfaces; the RAG knowledge base analysis layer is constructed by taking a self-built RAG knowledge base as a core, and the back-end service layer encapsulates a metadata management center. According to the method, through a multi-dimensional retrieval mode of'metadata screening + semantic retrieval + knowledge graph association ', the retrieval range is narrowed by metadata, then a knowledge unit is accurately positioned through RAG semantic retrieval, and finally associated document information is supplemented, so that compared with traditional full-text retrieval, the retrieval precision is improved, and the retrieval response time is shortened to millisecond level; automatically mining a document association relationship through a knowledge graph construction module, and supporting a user-defined association rule; and through a document-knowledge integrated view and a scenarized knowledge aggregation function, cross-department and cross-project knowledge rapid circulation is realized, and the knowledge reuse efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of file management technology, specifically to an intelligent file management system based on RAG and metadata lifecycle management. Background Technology

[0002] In current enterprise document management scenarios, traditional document management systems generally suffer from three major pain points: First, low retrieval efficiency. Relying on a "folder hierarchy + full-text keyword matching" retrieval model, users need to sift through massive amounts of documents to find the target content, especially when faced with complex formats such as PDF academic papers and Word contracts, making it difficult to accurately locate key knowledge units. Second, difficulty in knowledge reuse. Documents lack effective connections, forming "document silos." Cross-departmental and cross-project knowledge sharing requires manual screening and transfer, making it impossible to achieve rapid flow of structured knowledge. Third, fragmented metadata management. Metadata relies heavily on manual annotation, which is not only labor-intensive and error-prone but also lacks full lifecycle tracking, leading to chaotic document classification and failure to promptly clean up expired documents, further exacerbating the bloated knowledge base and retrieval difficulty. While existing RAG technology enables semantic retrieval, it is largely limited to semantic parsing of single documents and lacks deep integration with metadata management. This prevents the narrowing of search scope through metadata filtering, leading to an excessive number of irrelevant results. Furthermore, current metadata management solutions only support basic field annotation, lacking a linkage mechanism with the RAG knowledge base index and failing to achieve full lifecycle management of metadata from extraction, annotation, association to expiration. Therefore, there is an urgent need for an intelligent file management system that integrates RAG semantic retrieval capabilities with full metadata lifecycle management and enables seamless document correlation, thereby addressing the aforementioned shortcomings of traditional systems. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an intelligent file management system based on RAG and metadata lifecycle management. This system solves the problems of low retrieval efficiency, difficulty in knowledge reuse, and fragmented metadata management in traditional file management systems, enabling intelligent processing of document resources, structured knowledge transformation, and efficient correlation retrieval.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent file management system based on RAG and metadata full lifecycle management, characterized in that it includes a front-end application layer, a back-end service layer, a RAG knowledge base parsing layer, and a data storage layer, with each layer working collaboratively through standardized interfaces; The RAG knowledge base parsing layer is built around the DEEP DOC engine and the Auto Chunk method, while the backend service layer encapsulates the metadata management hub.

[0005] Furthermore, the front-end application layer includes a document operation interface, a knowledge base retrieval interface, and a metadata configuration interface; the document operation interface supports users to upload documents in multiple formats and view an integrated "document-knowledge" view, which simultaneously displays basic document information, knowledge tags, and related documents; the knowledge base retrieval interface supports dual-mode linkage operation of "metadata filtering + semantic retrieval"; the metadata configuration interface allows users to view metadata automatically extracted by the system and supplement custom metadata.

[0006] Furthermore, the backend service layer includes a metadata management hub and a RAG knowledge base scheduling module. The metadata management hub has the capabilities of automatic metadata extraction, manual annotation, association mapping, and full lifecycle tracking. It can call the RAG knowledge base parsing layer interface to extract document titles, formats, core keywords, generate text summaries and keywords, and also supports user configuration of custom metadata such as "business department", "project affiliation", "confidentiality level", and "validity period", and records the entire process log of metadata "creation-modification-deletion".

[0007] Furthermore, the RAG knowledge base parsing layer includes an automatic document parsing module, an adaptive semantic segmentation module, a vector index generation module, and a knowledge graph construction module. The adaptive semantic segmentation module can automatically match differentiated segmentation strategies such as segmentation by chapter, segmentation by clause, and segmentation by table dimension for different document types such as PDF, Word, and Excel, avoiding retrieval accuracy problems caused by overly coarse or overly fine segmentation.

[0008] Furthermore, the data storage layer adopts a hybrid storage mode of "structured + unstructured"; the structured storage unit is used to store metadata, knowledge base configuration, user operation logs and knowledge graph relationship data, and supports fast filtering and querying by "document type", "creation time" and "confidentiality level"; the unstructured storage unit is used to store the original document content and RAG vector index data, and the original document, metadata and RAG vector index are strongly bound through "document unique ID".

[0009] Furthermore, the knowledge graph construction module of the RAG knowledge base parsing layer can automatically build relationships between documents based on extracted document keywords, topics, and entities (such as project names, customer names, and product models), including "same project" and "same time period" relationships. It also supports users to manually add custom relationship rules such as "supporting documents" and "version iterations" based on business needs.

[0010] Furthermore, the RAG knowledge base scheduling module in the backend service layer can schedule tasks such as document parsing, semantic segmentation, and vector index generation. After the document is uploaded, it first performs format verification and security checks. If the verification passes, the DEEPDOC engine is started to automatically parse the document and provides real-time feedback on the processing progress to the front end. If the indexing fails, it records error logs and supports one-click reprocessing.

[0011] Furthermore, it also includes a metadata management hub, which identifies expired documents based on the "expiration date" field in the metadata, and classifies files by department based on the "business affiliation" field, performs permission classification, and triggers reminders to notify administrators to clean up or archive them, so as to avoid the knowledge base becoming bloated.

[0012] Furthermore, the front-end application layer supports scenario-based knowledge aggregation, which can automatically aggregate relevant documents and knowledge based on user roles or business scenarios, and assign different viewing and editing permissions for knowledge base files to different R&D groups; it can aggregate requirement documents, design documents, and test reports of the projects they are responsible for for "R&D engineers", and aggregate product manuals, competitor analysis, and marketing plans for "marketing personnel", without requiring users to filter across folders.

[0013] Furthermore, it also includes an asynchronous processing mechanism for the system, which asynchronously processes time-consuming operations such as document parsing, RAG index generation, and batch metadata annotation to avoid blocking user operations; at the same time, it caches the knowledge base vector index for high-frequency retrieval and indexes the metadata segments for high-frequency filtering to ensure that the retrieval response time is in the millisecond range, thereby improving the system's interactive experience.

[0014] Compared with the prior art, the technical solution of this application has the following beneficial effects: 1. This invention employs a multi-dimensional retrieval model combining "metadata filtering + semantic retrieval + knowledge graph association." First, it narrows the retrieval scope using metadata, then accurately locates knowledge units through RAG semantic retrieval, and finally supplements related document information. Compared to traditional full-text retrieval, this improves retrieval accuracy and reduces retrieval response time to milliseconds. At the same time, the adaptive semantic segmentation strategy avoids the problem of "overly coarse / overly fine segmentation," further ensuring retrieval accuracy. It can directly perform the most relevant information retrieval for large amounts of text information and can achieve document-level support for text content. 2. This invention automatically mines document relationships through a knowledge graph construction module and supports custom association rules, breaking down "document silos"; the integrated "document-knowledge" view and scenario-based knowledge aggregation function enable rapid knowledge flow across departments and projects, improving knowledge reuse efficiency. On the other hand, it can provide reference and document support for knowledge generated in reports. 3. This invention achieves automatic extraction of basic metadata through a metadata management hub, reducing the workload of manual annotation; reduces knowledge base maintenance costs based on automatic metadata governance; improves system interaction experience and reduces user waiting time by asynchronously processing time-consuming operations such as document parsing and index generation; enables hierarchical access to the knowledge base based on the metadata "confidentiality level" field; encrypts and stores sensitive metadata; records audit logs for key operations throughout the process, meeting the data security and compliance auditing needs of enterprises; and supports efficient aggregation and intelligent processing of a large number of documents, achieving a comprehensive upgrade from precise field-level retrieval to semantic-level knowledge retrieval. Attached Figure Description

[0015] Figure 1 This is a block diagram of an intelligent file management system based on RAG and metadata lifecycle management according to the present invention; Figure 2 This is a flowchart of an intelligent file management system based on RAG and metadata lifecycle management according to the present invention. Figure 3 This is a hierarchical interaction diagram of an intelligent file management system based on RAG and metadata lifecycle management, as described in this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1-3 This embodiment presents an intelligent file management system based on RAG and metadata lifecycle management, comprising a front-end application layer, a back-end service layer, a data storage layer, and a RAG knowledge base parsing layer. Each layer collaborates through standardized interfaces. The specific structure and functions are as follows: Front-end application layer: Provides a visual operation interface, including a document operation interface, a knowledge base retrieval interface, and a metadata configuration interface; the document operation interface supports users to upload documents and view a document-knowledge integrated view; the knowledge base retrieval interface supports multi-dimensional retrieval operations of "metadata filtering + semantic retrieval + knowledge graph association", and synchronously displays related documents of the retrieval results; the metadata configuration interface allows users to view metadata automatically extracted by the system, supplement custom metadata, and configure metadata templates. Backend service layer: Built with Django + REST Framework, the core encapsulates the metadata management hub and RAG knowledge base scheduling module. The metadata management hub is used to realize automatic extraction, manual annotation, association mapping, and full lifecycle tracking of metadata. Specifically, it includes: receiving document uploads or metadata annotation requests from the frontend, triggering the automatic metadata extraction function of the RAG knowledge base parsing layer to obtain automatically extracted metadata such as document title, format, and core keywords; feeding back the automatically extracted metadata to the frontend for users to supplement custom metadata such as "business department," "confidentiality level," and "validity period"; and storing the annotated structured metadata in the data storage layer after associating it with the document's unique ID and RAG index ID, and recording the metadata's "creation-modification-deletion" process. The system features a full-process log; it triggers expired document reminders based on the "expiration date" field in the metadata; the RAG knowledge base scheduling module coordinates document parsing, semantic segmentation, vector index generation, and retrieval tasks within the RAG knowledge base parsing layer, specifically including: upon receiving a document upload request from the front end, performing format verification and security checks on the document; if the verification passes, initiating the automatic parsing process of the RAG knowledge base parsing layer; real-time monitoring of the progress of document parsing, semantic segmentation, and vector index generation, and synchronizing this information to the front end; recording error logs and supporting one-click reprocessing when indexing fails; and upon receiving a retrieval request from the front end, narrowing the retrieval scope based on metadata filtering conditions, then calling the semantic retrieval interface of the RAG knowledge base parsing layer to return accurate retrieval results. RAG Knowledge Base Parsing Layer: Built upon a self-built RAG knowledge base, this layer provides the system with core AI processing capabilities, including: an automatic document parsing module that supports content extraction from multiple document formats such as PDF, Word, and Excel; an adaptive semantic segmentation module that matches segmentation strategies based on document type, such as segmenting PDF academic papers into "abstract-chapter-references" blocks and Word contracts into "clauses-attachments-signing information" blocks; a vector index generation module that converts the semantically segmented content into document semantic vector indexes; a knowledge graph construction module that automatically builds relationships between documents (such as associating documents within the same project or time period) based on extracted document keywords, topics, and entities (such as project names and client names), and supports users manually adding custom association rules (such as associating documents with "supporting documents" or "version iterations"); and a semantic retrieval module that receives retrieval requests from the backend service layer, performs semantic matching based on vector indexes, and returns retrieval results.

[0018] Data storage layer: Adopting a hybrid storage mode of "structured + unstructured", including: structured storage units for storing metadata, knowledge base configuration, user operation logs, and knowledge graph relationship data, supporting fast filtering and querying by dimensions such as "document type", "creation time", and "confidentiality level"; unstructured storage units for storing original document content and RAG vector index data, ensuring document integrity and semantic retrieval accuracy; and data association units, using "document unique ID" as the core association key, to achieve strong binding of "original document - metadata - RAG vector index - knowledge graph relationship", ensuring the consistency and traceability of data flow. III. Work Process Knowledge base construction process S1: Users upload documents through the document operation interface of the front-end application layer, and the front-end sends the document data to the back-end service layer; S2: The backend service layer performs format validation and security checks on the document. If the validation fails, it returns an error message (format not supported / content risk) to the frontend. If the validation passes, it triggers the automatic document parsing process of the RAG knowledge base parsing layer and synchronizes the processing progress (parsing / blocking / indexing) to the frontend in real time. S3: The RAG knowledge base parsing layer's automatic document parsing module extracts document content, and the adaptive semantic segmentation module matches segmentation strategies based on document type to generate semantic segments; S4: The vector index generation module transforms semantic blocks into document semantic vector indexes, while the knowledge graph construction module extracts document keywords and entities and automatically builds document relationships. S5: The metadata management hub of the backend service layer calls the metadata automatic extraction interface of the RAG knowledge base parsing layer to obtain the metadata automatically extracted by the system and feed it back to the frontend for users to supplement custom metadata; S6: Users submit the annotated metadata through the metadata configuration interface. The backend service layer associates the structured metadata with the document's unique ID and RAG index ID, and then stores it in the structured storage unit of the data storage layer; the RAG vector index and the original document content are stored in the unstructured storage unit respectively; the knowledge graph association is stored in the structured storage unit. S7: The backend service layer updates the document processing status to "complete" and synchronizes it to the frontend to complete the knowledge base construction. Multi-dimensional search process T1: Users initiate a search request by entering metadata filtering conditions (such as "business department = R&D department" and "confidentiality level = public") and semantic search keywords (such as "cost budget") through the knowledge base search interface of the front-end application layer. T2: After receiving the search request, the backend service layer first filters out the target document range based on the metadata filtering conditions through the structured storage unit of the data storage layer. T3: The backend service layer calls the semantic retrieval module of the RAG knowledge base parsing layer to perform semantic matching based on the RAG vector index within the scope of the target document and obtain the semantic retrieval results; T4: The knowledge graph construction module supplements the related document information (such as project documents and similar topic documents) based on the relationship between the target documents. T5: The backend service layer integrates "semantic search results + associated document information" and feeds it back to the frontend application layer; T6: The front-end application layer displays search results in an integrated "document-knowledge" view, including basic document information, knowledge tags, and related documents, and supports users to click on related document nodes to jump to view them. Metadata lifecycle management process U1: Metadata Creation: In the knowledge base construction process, structured metadata is generated through "automatic extraction by the system + manual annotation by the user" to complete metadata creation; U2: Metadata Update: Users initiate metadata modification requests through the metadata configuration interface of the front-end application layer. The back-end service layer records the modified content, operator, and operation time, updates the metadata to the data storage layer, and synchronously updates the associated RAG index and knowledge graph relationships. U3: Metadata Tracking: The backend service layer records the entire process log of metadata "creation-modification-deletion" and stores it in the structured storage unit of the data storage layer, allowing users to query the metadata change history through the front-end interface; U4: Metadata Governance: The backend service layer periodically checks for expired documents based on the "expiration date" field in the metadata and triggers expiration reminders; based on the "business ownership" field, it identifies documents without ownership and reminds administrators to clean up or archive them to avoid bloated knowledge base.

[0019] Example 1: Knowledge Base Management in the R&D Department of a Technology Company 1. System Deployment and Configuration Hardware environment: The front-end user terminal uses a regular office computer (CPU i5-12400, memory 16GB); the back-end service is deployed on an enterprise private cloud server (CPU 2 AMD EPYC7763, memory 256GB, hard disk 2TB SSD); the RAG knowledge base parsing layer is deployed on a GPU server (2 NVIDIA A10, video memory 24GB / card) to meet the requirements of multi-document parallel parsing and semantic retrieval. Software environment: The front-end is built based on Vue.js 3.0 + ElementPlus, and integrates ECharts 5.3 to visualize the knowledge graph; the back-end uses Django + REST Framework to build a microservice architecture, and uses Nacos to implement service registration and configuration management; the RAG knowledge base parsing layer integrates a self-built RAG knowledge base 1.2, and uses LangChain 0.1.10 to optimize semantic block logic; the data storage layer uses MySQL 8.0 (structured data, master-slave backup) + MinIO 2023.10 (unstructured data, 3-replica storage) to ensure data security. 2. Function Implementation Process Step 1: Batch import of R&D documents and initialization of the knowledge base The R&D department needs to manage four core document categories: project requirements documents, technical design documents, test reports, and patent application materials, totaling over 500 documents (including Word, PDF, and Excel formats). Administrators can import all documents at once using the front-end "batch document upload" function, and the system will automatically process them. 1) Format verification: Two corrupted PDFs were identified (with the message "File corrupted, cannot be parsed"). The administrator replaced them and re-uploaded them. 2) Security check: Three Word documents containing external links were scanned and automatically marked as "pending review". Once approved, they will proceed to the parsing process. 3) Adaptive semantic segmentation: The technical design document (PDF) is segmented into "overall design - module design - interface definition - test plan" (15-20 segments per document on average), and the test report (Excel) is segmented into "test case number - test steps - expected results - actual results" (5-8 segments per sheet). 4) Automatic metadata extraction: Extract basic metadata such as "document title, creation time, core keywords (such as 'artificial intelligence algorithm', 'distributed architecture'), and document format", and simultaneously generate a unique document ID (such as "RD-20240508-001") and RAG index ID. Step 2: Metadata Annotation and Knowledge Association Construction Developers can add custom metadata to each document through the front-end "metadata configuration interface": Business Dimension: "Project Affiliation (e.g., 'Intelligent Driving Project', 'Industrial AI Platform Project')" and "Research and Development Phase (Requirements Review / Development / Testing / Acceptance)"; Security dimension: "Confidentiality level (public / internal / confidential)" (120 core design documents are marked "confidential"); Lifecycle dimension: "Validity period (e.g., '2024.05-2026.05')" "Archiving conditions (3 months after project acceptance)". The system automatically builds knowledge associations based on metadata and document content: Link with other projects: Automatically link the requirements documents, design documents, and test reports of the "Intelligent Driving Project" to generate a "Project Document Cluster"; Technical association: Identify the keyword "distributed architecture" and associate it with all design documents containing this keyword to form a "technical theme association chain"; R&D personnel manually added "version associations": marking "V1.0 design document" and "V2.0 optimization document" as "iteration versions" to ensure clear version traceability. Step 3: Develop scenario-based retrieval and knowledge reuse Scenario 1: An algorithm engineer needs to query the "LiDAR data processing solution for an intelligent driving project". 1) Metadata filtering: In the front-end search interface, select "Project Affiliation = Intelligent Driving Project", "Confidentiality Level = Internal", and "Research and Development Stage = Development"; 2) Semantic retrieval: Input the keyword "LiDAR data processing solution", the system first filters out 20 target documents, and then locates the "LiDAR module processing logic" section in 3 core design documents through RAG semantic retrieval; 3) Related Display: Automatically displays related "LiDAR Test Cases (Excel)" and "Data Interface Documents (Word)". Engineers can view them by clicking on the related nodes without having to search across folders. The search time is reduced from 15 minutes in the traditional system to 30 seconds. Scenario 2: The patent specialist needs to compile a list of "patent technology points related to the industrial AI platform project". 1) Metadata filtering: Select "Project Affiliation = Industrial AI Platform Project" and "Document Type = Patent Application Materials"; 2) Semantic retrieval: Input "patent technology point extraction", and the system will automatically extract 8 core technology points such as "AI model optimization method" and "data desensitization technology" from 15 patent documents and generate a "technology point-document" association list. Patent specialists can directly reuse this list to complete the patent summary, improving work efficiency by 70%.

[0020] Step 4: Metadata Lifecycle Governance The system automatically performs metadata governance tasks on the 1st of each month: 1) Expiration Reminder: If the "Intelligent Driving Project V1.0 Requirements Document" (valid from May 2024 to November 2024) is about to expire, send an email to the R&D manager reminding him that "the document will expire in 30 days, would you like to extend the validity period?" After the manager confirms, the validity period will be updated to November 2025. 2) Cleaning up unassigned documents: Five test reports with "unassigned project" were found and automatically assigned to the test department administrator. The administrator then added metadata such as "Project Assignment = Industrial AI Platform Project" to prevent the accumulation of "zombie documents" in the knowledge base. 3) Confidentiality level audit: If an unauthorized user (ordinary R&D personnel) attempts to retrieve a "confidential" design document, the request will be blocked immediately and logged. The information security department will be notified simultaneously to meet the company's data security compliance requirements. 3. Application Results Search efficiency: The average time for R&D personnel to search documents has been reduced from 15 minutes to 45 seconds, and the search accuracy has been improved from 40% of traditional keyword searches to 92%. Knowledge reuse: Cross-project knowledge sharing frequency increased by 80%, and the time for new employees to become familiar with project documents was shortened from 1 month to 2 weeks; Management costs: The workload of manual metadata annotation was reduced by 65%, and the number of knowledge base maintenance personnel was reduced from 3 to 1, resulting in annual savings of approximately 150,000 yuan in management costs. Example 2: Compliance Document Management in the Risk Control Department of Financial Enterprises 1. System deployment and adaptation In response to the financial industry's demand for "high security and high compliance", an "encryption module" has been added to the data storage layer: sensitive metadata (such as "customer ID number" and "risk control model parameters") are stored using AES-256 encryption, and the original document transmission process is encrypted using SSL / TLS1.3. The RAG knowledge base parsing layer has added a "compliance semantic filtering" function: when parsing documents, it automatically identifies "non-compliant statements" (such as content that does not comply with the "Core Indicators for Risk Supervision of Commercial Banks"), marks them, and reminds risk control specialists to review them. 2. Core Application Scenarios Compliance document retrieval and verification: Risk control specialists need to query the "2024 Personal Credit Risk Control Model Compliance Report". By filtering the metadata ("Document Type = Compliance Report", "Time Range = 2024.01-2024.12", "Business Line = Personal Credit"), and semantically searching for "Risk Control Model Compliance", the system not only returns the target report, but also automatically extracts core indicators such as "Model Pass Rate" and "Bad Debt Rate" from the report, compares them with regulatory threshold requirements, and generates "Compliance Verification Result" (e.g., "Bad Debt Rate 2.1% < Regulatory Threshold 3%, Meets Requirements"), reducing manual verification time. Metadata Compliance Tracking: For "customer credit files", the system automatically extracts metadata such as "customer name, credit amount, and approval time", and records a full log of "file viewer, viewing time, and viewing purpose", which meets the "data use traceability" requirement of the Personal Information Protection Law and avoids compliance risks. Example 3: Process Document Management in the Production Department of a Manufacturing Enterprise 1. System deployment and adaptation Supports "Workshop Terminal Adaptation": The front-end interface is adapted to industrial tablets (touchscreen operation) to meet the needs of workshop workers to view process documents on the production site; The RAG knowledge base parsing layer has added a "drawing parsing" function: it supports content extraction and semantic segmentation of CAD drawings (.dwg format), segmented by "part size - assembly process - precision requirements", making it easier for process engineers to retrieve key parameters.

[0021] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0022] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent file management system based on RAG and metadata full life cycle management, characterized in that, It includes a front-end application layer, a back-end service layer, a RAG knowledge base parsing layer, and a data storage layer, with each layer working together through standardized interfaces; The RAG knowledge base parsing layer is built around a self-built RAG knowledge base, and the backend service layer encapsulates the metadata management hub. 2.The intelligent file management system based on RAG and metadata full life cycle management according to claim 1, characterized in that, The front-end application layer includes a document operation interface, a knowledge base retrieval interface, and a metadata configuration interface. The document operation interface allows users to upload documents in multiple formats and view an integrated "document-knowledge" view, which simultaneously displays basic document information, knowledge tags, and related documents. The knowledge base retrieval interface supports dual-mode operation of "metadata filtering + semantic retrieval". The metadata configuration interface allows users to view metadata automatically extracted by the system and supplement custom metadata. 3.The intelligent file management system based on RAG and metadata full life cycle management according to claim 1, characterized in that, The backend service layer includes a metadata management hub and a RAG knowledge base scheduling module. The metadata management hub has the capabilities of automatic metadata extraction, manual annotation, association mapping, and full lifecycle tracking. It can call the RAG knowledge base parsing layer interface to automatically extract metadata such as document title, format, core keywords, and chapter structure, and generate text summaries and keywords. It also supports user configuration of custom metadata such as "business department", "project affiliation", "confidentiality level", and "validity period", and records the entire process log of metadata "creation-modification-deletion", with the aim of realizing document traceability and lineage analysis.

4. The intelligent file management system based on RAG and metadata lifecycle management as described in claim 1, characterized in that, The RAG knowledge base parsing layer includes an automatic document parsing module, an adaptive semantic segmentation module, a vector index generation module, and a knowledge graph construction module. The adaptive semantic segmentation module can automatically match differentiated segmentation strategies such as segmentation by chapter, segmentation by clause, and segmentation by table dimension for different document types such as PDF, Word, and Excel, avoiding retrieval accuracy problems caused by overly coarse or overly fine segmentation.

5. The intelligent file management system based on RAG and metadata lifecycle management according to claim 1, characterized in that, The data storage layer adopts a hybrid storage mode of "structured + unstructured", including structured storage units and unstructured storage units. The structured storage units are used to store metadata, knowledge base configuration, user operation logs and knowledge graph relationship data, and support fast filtering and querying by "document type", "creation time" and "confidentiality level". The unstructured storage units are used to store the original document content and RAG vector index data, and a strong binding between the original document, metadata and RAG vector index is achieved through "unique document ID".

6. The intelligent file management system based on RAG and metadata lifecycle management according to claim 1, characterized in that, The knowledge graph construction module of the RAG knowledge base parsing layer can automatically build the relationship between documents based on the extracted document keywords, topics, and entities (such as project name, customer name, product model), including the "same project" and "same time period" associations. It also supports users to manually add custom association rules such as "supporting documents" and "version iteration" based on business needs.

7. The intelligent file management system based on RAG and metadata lifecycle management according to claim 3, characterized in that, The RAG knowledge base scheduling module in the backend service layer can schedule tasks such as document parsing, semantic chunking, and vector index generation. After the document is uploaded, it first performs format verification and security checks. If the verification passes, the DEEP DOC engine is started to automatically parse the document and provides real-time feedback on the processing progress to the front end. If the indexing fails, it records error logs and supports one-click reprocessing.

8. The intelligent file management system based on RAG and metadata lifecycle management according to claim 1, characterized in that, It also includes a metadata management hub, which identifies expired documents based on the "expiration date" field in the metadata and classifies files by department based on the "business affiliation" field, assigns permissions, and triggers reminders to administrators to clean up or archive the files, thus preventing the knowledge base from becoming bloated.

9. The intelligent file management system based on RAG and metadata lifecycle management according to claim 1, characterized in that, The front-end application layer supports scenario-based knowledge aggregation, which can automatically aggregate relevant documents and knowledge based on user roles or business scenarios. It can aggregate requirement documents, design documents, and test reports of the projects they are responsible for for "R&D engineers", and aggregate product manuals, competitor analysis, and marketing plans for "marketing personnel", without requiring users to filter across folders.

10. The intelligent file management system based on RAG and metadata lifecycle management according to claim 1, characterized in that, It also includes an asynchronous processing mechanism for the system, which asynchronously processes time-consuming operations such as document parsing, RAG index generation, and batch metadata annotation to avoid blocking user operations; at the same time, it caches the knowledge base vector index for high-frequency retrieval and indexes the metadata segments for high-frequency filtering to ensure that the retrieval response time is in the millisecond range and improve the system's interactive experience.

Images