Digital document management system and method and storage medium

Abstract

The invention relates to the technical field of computer information processing, in particular to a digital document management system and method and a storage medium, and the management method comprises the following steps: obtaining a to-be-sorted file and a file sorting instruction during file sorting; identifying a file sorting instruction based on a large language model, carrying out file content integration on a to-be-sorted file, generating an abstract corresponding to the to-be-sorted file, and generating a file sorting scheme according to a set sorting rule; when a confirmation instruction of the file arrangement scheme is received, the file arrangement scheme is executed to store the to-be-arranged file, and file arrangement is completed; when the file is retrieved, obtaining a file retrieval condition; and retrieving in the stored file according to the file retrieval condition, and returning a file information list meeting the file retrieval condition. According to the system and the method, file retrieval universality and accuracy can be improved, and compared with manual operation of traditional arrangement, the system and the method have the advantage that manual intervention links are reduced through workflows.

Description

Technical Field

[0001] The present invention relates to the field of computer information processing technology, and in particular to a digital document management system, method and storage medium. Background Art

[0002] Current digital document organization tools primarily rely on manual classification mechanisms based on file names or metadata, and archive through user-defined tags or folder structures. However, these tools have the following technical drawbacks:

[0003] Low intelligence: Excessive reliance on manual intervention prevents automatic identification of document content and generation of classification tags. Experiments have shown insufficient archiving accuracy. Traditional software (such as Windows Explorer) relies solely on file extensions or manual tags for classification, lacking the ability to automatically generate multi-dimensional semantic tags based on document content. The resulting classification granularity is coarse and difficult to meet diverse needs. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the prior art and provide a digital document management system, method and storage medium to solve the problem that existing digital document organization tools only rely on file extensions or manual tags for classification, lack the ability to automatically generate multi-dimensional semantic tags based on document content, have rough classification granularity, and are difficult to meet diverse needs.

[0005] The technical solution to achieve the above purpose is:

[0006] The present invention provides a digital document management method, comprising the following steps:

[0007] When organizing files, obtain the files to be organized and the file organization instructions;

[0008] Identify file organization instructions based on a large language model, integrate the content of the files to be organized, generate summaries corresponding to the files to be organized, and then generate a file organization plan based on the set organization rules;

[0009] Upon receiving a confirmation instruction of the file arrangement plan, executing the file arrangement plan to store the files to be arranged, thereby completing the file arrangement;

[0010] When searching for a file, obtain the file search conditions;

[0011] Search the stored files according to the file search conditions and return a list of file information that meets the file search conditions.

[0012] A further improvement of the digital document management method of the present invention is that, when generating a file organization plan, key parameters are obtained according to the identified file organization instructions, and the obtained key parameters include classification basis, classification dimension, and target storage location;

[0013] Obtain metadata of the file to be organized, the metadata obtained including the creation time, modification time, and access time of the file to be organized;

[0014] According to the metadata and summary of each file, and based on the key parameters obtained, a storage path and a new name are assigned to each file, thereby completing the generation of a file organization plan.

[0015] A further improvement of the digital document management method of the present invention is that it further comprises:

[0016] Get user feedback;

[0017] Store the acquired user feedback records;

[0018] When identifying file organization instructions and file retrieval conditions based on the large language model, the file organization instructions and file retrieval conditions are understood and identified based on recorded user feedback.

[0019] A further improvement of the digital document management method of the present invention is that it further comprises:

[0020] Get configuration parameters;

[0021] Generate corresponding sorting rules based on the obtained configuration parameters.

[0022] A further improvement of the digital document management method of the present invention is that, when searching for a file, it also includes obtaining a file search method;

[0023] According to the obtained file retrieval method, the retrieval is performed, and a list of retrieved file information is returned.

[0024] The present invention also provides a storage medium on which a program of the digital document management method is stored. When the program of the digital document management method is executed by a processor, the steps of the digital document management method are implemented.

[0025] The present invention further provides a digital document management system, comprising:

[0026] The acquisition module is used to obtain the files to be organized and the file organization instructions when organizing files, and is also used to obtain the file search conditions when searching files;

[0027] A large language model decision engine, connected to the acquisition module, configured to identify file organization instructions based on the large language model, integrate the contents of the files to be organized, generate summaries corresponding to the files to be organized, and then generate a file organization plan according to the set organization rules;

[0028] Data storage module;

[0029] an execution module connected to the acquisition module, the large language model decision engine, and the data storage module, and configured to, upon receiving a confirmation instruction of the file organization plan, execute the file organization plan to store the files to be organized in the data storage module, thereby completing the file organization;

[0030] A file retrieval module is connected to the acquisition module and the data storage module. The file retrieval module is used to search the data storage module according to the file retrieval conditions and return a list of file information that meets the file retrieval conditions.

[0031] A further improvement of the digital document management system of the present invention is that the large language model decision engine is further used to extract key parameters from the identified file organization instructions, the key parameters including classification basis, classification dimension and target storage location;

[0032] The large language model decision engine is further used to obtain metadata of the files to be sorted, and the obtained metadata includes the creation time, modification time and access time of the files to be sorted;

[0033] The large language model decision engine is also used to assign a storage path and a new name to each file based on the metadata and summary of each file and the key parameters obtained, thereby completing the generation of a file organization plan.

[0034] A further improvement of the digital document management system of the present invention is that the acquisition module is also used to obtain user feedback;

[0035] The large language model decision engine is further used to store user feedback in the data storage module, and is further used to understand and identify file organization instructions and file retrieval conditions based on the stored user feedback.

[0036] A further improvement of the digital document management system of the present invention is that the acquisition module is also used to acquire configuration parameters;

[0037] The large language model decision engine is further configured to generate corresponding sorting rules according to the acquired configuration parameters.

[0038] The digital document management system, method, and storage medium of the present invention have the following beneficial effects:

[0039] The present invention generates a file organization solution based on a large language model, reducing manual reading and operation time. The large language model retrieves files based on semantics, which can improve the breadth and accuracy of file retrieval. Compared with traditional manual organization operations, the system and method of the present invention reduce manual intervention links through workflow.

[0040] The digital document management system and method of this invention improves the efficiency of automated file organization. When a user clicks a button to trigger the automated execution module, the system executes the file operation instruction set according to the generated organization plan. This reduces the time required to organize 100 documents from 35 minutes (using traditional methods such as manual resource management systems) to 8 minutes, reducing manual clicks by 93% and significantly improving operational efficiency.

[0041] The digital document management system and method of the present invention allows users to complete complex operations through natural language instructions, lowering the barrier to entry. The system also displays processes and results in natural language, making them easier for users to understand and comprehend, lowering the barrier to entry.

[0042] The digital document management system and method of the present invention support multi-terminal access and unified management in a local area network, bring about multi-host collaboration, and enhance the scope of asset management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is an architectural diagram of the digital document management system of the present invention.

[0044] Figure 2 This is a data flow chart of the digital document management method of the present invention.

[0045] Figure 3 This is a step diagram of the workflow of the large language model decision engine in the digital document management system or method of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0047] See Figure 1The present invention provides a digital document management system, method, and storage medium to address the shortcomings of existing file management solutions in terms of integrated organization and retrieval, intelligent interaction, multi-dimensional data utilization, automated execution, multi-dimensional retrieval, and multi-host collaboration. The system and method of the present invention have a natural language interactive interface, allowing users to complete file organization and retrieval through natural language, and the system returns natural language to present the organization process and results. The system and method of the present invention include a decision engine based on a large language model, which analyzes user needs in combination with the large language model to generate a file organization plan. The system and method of the present invention include multi-dimensional metadata automatic extraction and association, realizing automatic recording of file content and other metadata (such as summary, path, modification time) to support reasonable and efficient intelligent organization and multi-dimensional retrieval. The system and method of the present invention include automated execution, and after user confirmation and permission, automatically parses the organization plan and executes the corresponding file system operation instruction set (such as copy, move, rename), reducing manual intervention. The system and method of the present invention also include multi-dimensional file retrieval, using the large language model to understand the user's search intent and find appropriate file metadata and content summary. Supporting more search conditions and methods, the system and method of the present invention improves the usability and universality of the search function. The digital document management system, method and storage medium of the present invention are described below with reference to the accompanying drawings.

[0048] See Figure 1 , shows the architecture diagram of the digital document management system of the present invention. Figure 1 , the digital document management system of the present invention is described.

[0049] like Figure 1 As shown, the digital document management system of the present invention includes an acquisition module, a large language model decision engine, a data storage module, an execution module and a file retrieval module, wherein the large language model decision engine is connected to the acquisition module, the execution module is connected to the acquisition module, the large language model decision engine and the data storage module, the file retrieval module is connected to the acquisition module and the data storage module, and the file retrieval module is also connected to the large language model decision engine. The acquisition module is used to obtain the files to be sorted and the file sorting instructions when sorting files. The acquisition module is also used to obtain file retrieval conditions when searching for files; the large language model decision engine is used to identify the file sorting instructions based on the large language model, integrate the file contents of the files to be sorted, and generate a summary corresponding to the files to be sorted, and then generate a file sorting plan according to the set sorting rules; the execution module is used to execute the file sorting plan when receiving the confirmation instruction of the file sorting plan to store the files to be sorted in the data storage module, thereby completing the file sorting; the file retrieval module is used to search in the data storage module according to the file retrieval conditions and return a list of file information that meets the file retrieval conditions.

[0050] Furthermore, the digital document management system of the present invention also includes an interactive interface module, which is used to display an operation interface for the user. Text blocks and buttons are displayed on the operation interface to facilitate the user to perform corresponding operations.

[0051] Specifically, the interactive interface module displays a file upload text block and button on the operation interface. Clicking the file upload button opens a local program to select a file. After selecting a file, click OK to upload the file. The system's acquisition module then receives the uploaded file as the file to be sorted. When a file is selected through the local program, the local program can obtain the absolute path of the file storage location and locate the file based on the absolute path. This allows the acquisition module to obtain the metadata of the uploaded file.

[0052] The metadata of the uploaded files obtained by the acquisition module includes creation time, modification time and access time, as well as old location, old name, file type, file size and upload time. The obtained metadata is used to provide data support for subsequent file management.

[0053] The operation interface displayed by the interactive interface module is provided with a natural language input and output box. When in use, the user can input corresponding instructions in the natural language input and output box, and then call other modules of the system to implement corresponding functions. The user can input natural language in the natural language input and output box, that is, a language that is consistent with speaking habits. The digital document management system of the present invention has a semantic understanding function, which can understand the meaning of the natural language input by the user and then perform corresponding operations. The user can input file sorting requirements in the natural language input and output box to form file sorting instructions, and can also input search requirements to form search conditions, and can also input feedback information to form user feedback. The above information input by the user is obtained by the acquisition module, and then understood by the large language model decision engine, and then executed by the corresponding model.

[0054] The operation interface displayed by the interactive interface model has an operation execution and result display box, which is used to display the file organization plan given by the large language model decision engine for user confirmation, and is also used to display the execution results of the execution module and the file information list retrieved by the file retrieval module.

[0055] The interactive interface module of the present invention uses a standard web interface, namely HTML, CSS, and JavaScript, which can be displayed in various browsers. Users can directly observe and operate the interactive interface module, thereby scheduling other modules of the system. The interactive interface module's input is the user's natural language and interactive operations, and its output is the calls, process outputs, and results of each module in the system. The interactive interface module has the advantages of being lightweight and universal. It uses text blocks and buttons to make functions and results easy to read and use.

[0056] In a specific embodiment of the present invention, the large language model decision engine is designed through a workflow, and can receive the file sorting instructions input by the user and the uploaded files to be sorted obtained by the acquisition module, and then generate a sorting plan (including an operation instruction set for execution by the execution module) based on the received file sorting instructions input by the user and the uploaded files to be sorted, and then display the generated sorting plan on the operation interface through the interactive interface module for user confirmation.

[0057] The large language model decision engine of the present invention is capable of understanding user intent, analyzing file content, generating file summaries, designing file system operation instructions, and the like. The large language model decision engine includes configurable parameters and prompt word templates, which are used to constrain the output of the large language model to improve decision stability and controllability. Preferably, during the design phase of the digital document management system of the present invention, the large language model decision engine can be configured with set parameters and set prompt word templates so that the large language model decision engine can achieve the above functions. When in use, the user can also configure the functional parameters and prompt word templates of the large language model decision engine according to their own needs so that the large language model decision engine can meet the user's usage requirements.

[0058] Furthermore, the large language model decision engine is also used to extract key parameters from the identified file sorting instructions, including classification basis, classification dimension, and target storage location;

[0059] The large language model decision engine is also used to obtain metadata of the files to be sorted, including the creation time, modification time, and access time of the files to be sorted;

[0060] The large language model decision engine is also used to assign a storage path and a new name to each file based on the metadata and summary of each file and the key parameters obtained, thereby completing the generation of a file organization plan.

[0061] Specifically, through multi-stage workflow design, the output of large language models can be better planned and constrained, such as Figure 3As shown, the multi-stage workflow of the large language model decision engine includes stage 1-input decomposition stage, stage 2-file content integration stage, stage 3-summary generation stage, and stage 4-solution generation stage. This multi-stage workflow is used to implement file organization. In stage 1-input decomposition stage, the large language model is guided by prompt words to identify key parameters in the file organization instructions (that is, the natural language input by the user, the file organization requirements). The key parameters include classification basis, classification dimension, target storage location, etc. The identified key parameters are used for calls in subsequent stages. In stage 2-file content integration stage, for the uploaded files to be organized, the structured text content is extracted, and the unstructured data content is also extracted to generate the corresponding content text. The unstructured data content includes PDF, DPCX, image OCR and other format content, and finally the file content data is completely in text format. In stage 3-summary generation stage, according to the content of each file, a content summary is generated through prompt words, and the content length and summary method of the summary are constrained. The following is an example of a prompt word template:

[0062] #Prompt word example

[0063] """

[0064] You are a file information organization assistant, and you are responsible for organizing the obtained file information into the specified format.

[0065] Only output custom format content, no other content or explanation is required

[0066] The file name includes the extension

[0067] """

[0068] #User Question Example

[0069] """

[0070] According to the {document parsing results}, sort out the file name, file type, and file summary of each file, and simplify the summary to less than 100 words.

[0071] Output in the following format:

[0072] File name: File1

[0073] File type: type

[0074] Document Summary: A text

[0075] """

[0076] In Phase 4 - Scheme Generation, a sorting scheme is generated based on the metadata and file content summary of each file through restrictive prompt words to constrain the depth of the resource path and classification basis of the sorting.

[0077] Furthermore, the sorting plan generated by the large language model decision engine is displayed on the operation interface through the interactive interface module for user confirmation.

[0078] In the digital document management system of the present invention, the acquisition module is further configured to acquire configuration parameters; and the large language model decision engine is further configured to generate corresponding sorting rules based on the acquired configuration parameters. The configuration parameters acquired by the acquisition module are the prompt word template and the user question entered by the user through the operation interface.

[0079] The sorting scheme generated by the large language model decision engine is controllable, that is, when the user is not satisfied with the sorting scheme it generates, the user can use prompt words and user questions to constrain the output method of the large language model, thereby obtaining a more controllable and expected sorting scheme. The user can configure different prompt word templates and user questions to achieve the storage and rapid switching of sorting rules, thereby adapting to different document management scenarios and needs. Specifically, the user can input the prompt word template and user question through the operation interface, and then the large language model decision engine can obtain the prompt word template and user question input by the user, and then set it as the corresponding sorting rule storage for the user to select when using it later, and then give a sorting scheme based on the sorting rule.

[0080] The following template example shows a collation rule:

[0081] #Prompt word example

[0082] """

[0083] You are a file renaming and classification management assistant. You are responsible for proposing organization plans based on file information and user needs, and outputting file names and paths.

[0084] Requirements must be strictly followed:

[0085] 1. The original file name and the new file name must include the extension

[0086] 2. One record per line, only the content of the custom output format is output, no additional content or explanation is required.

[0087] """

[0088] #User Question Example

[0089] """

[0090] Utilize {document parsing results} to group and rename these files in {user-specified location}.

[0091] Refer to the {Organization Rules} in {System Configuration} and follow the path depth, classification basis, and renaming style.

[0092] Only output the content of the custom output format, the structure is as follows:

[0093] Original file name->new path\new file name

[0094] """

[0095] Furthermore, the acquisition module is also used to obtain user feedback; the large language model decision engine is also used to store user feedback in the data storage module, and is also used to understand and identify file organization instructions and file retrieval conditions based on the stored user feedback.

[0096] The large language model decision engine of the present invention can record user feedback to build a dedicated knowledge base, thereby optimizing the large language model's ability to understand the user's context.

[0097] Specifically, users can score the sorting solutions generated by the large language model each time and give modification suggestions. These modification suggestions will be recorded in the feedback knowledge base (a database within the data storage module) to supplement the large language model's understanding of the current situation and specific conditions. For example, a user creates a new suggestion: "XXX department or team" specifically refers to the "certain business department" where the user "I" is located. When analyzing and generating solutions, the large language model will use the content in the feedback knowledge base as a reference to understand user needs and file data. Continuing with the previous example, the large language model can understand the user's needs: "Find the files under my department" means finding the files belonging to "certain business department". Users can adjust all modification suggestions, which will take effect at the next governance, so that the large language model can dynamically adjust the sorting strategy based on the user's acceptance / modification behavior of historical solutions.

[0098] The large language model decision engine of the present invention establishes a large language model workflow, enabling the large language model to better help users solve file organization needs. The large language model decision engine breaks down the file organization task into multiple detailed steps, scheduling, planning, and constraining the use of the large language model's capabilities in each step. The large language model decision engine uses information from the data storage module for analysis and planning, ultimately deriving a file organization plan. The input of the large language model decision engine is the user's natural language, configuration parameters, and file data, and the output is the file organization plan and operation instructions.

[0099] Furthermore, the sorting scheme generated by the large language model decision engine includes operation instructions, which are used by the execution module to execute and complete the sorting of the corresponding files. A custom separation identification string (such as "---RENAME_START---") is added to the large language model workflow output to distinguish different content blocks output by the large language model workflow, such as file information blocks and operation instruction blocks, so that the execution module can correctly identify the operation instructions in one output.

[0100] In a specific embodiment of the present invention, the execution module is used to execute the operating instructions in the file organization plan, and is also used to execute the organization instructions input by the user from the operation interface. The execution module can automatically execute the organization plan. It can parse the operating instructions in the organization plan generated by the large language model decision engine and perform corresponding organization operations, such as copying and renaming files. After the user quickly calls the execution module through the button on the operation interface, the execution module will also execute the corresponding instructions, including renaming and moving the file. The input of the execution module is the user instructions and the file organization plan of the large language model decision engine, and the output is the change in the actual location and name of the file in the system resource manager.

[0101] Furthermore, the execution module executes file movement and renaming according to the arrangement plan, dynamically generates the target path, and automatically creates a directory to ensure that the file can be moved to the directory given by the plan and that the directory and file are correctly named. Dynamically generating the target path includes the following steps:

[0102] Step 1 - Path parsing: Use regular expressions to extract the target path string from the collation scheme and decompose it into a hierarchical structure.

[0103] Step 2 - Path Verification: Check whether the path complies with the file system specifications, including illegal character filtering (such as "\", " / ", ":", "*", "?", "|", etc.) and length restrictions (such as the maximum path length in Windows is 260 characters).

[0104] Step 3 - Directory creation: Check whether the path exists level by level, and create it recursively if it does not exist.

[0105] Step 4 - Conflict handling: If a file with the same name already exists in the target path, a new file name must be generated to avoid overwriting. According to the user's preset fault tolerance rules, a specified rename conflict identification character suffix is ​​added to distinguish them.

[0106] In a specific embodiment of the present invention, the data storage module adopts a layered architecture, including an object storage system, a relational database, and a vector database, wherein the object storage system is used for persistent storage of unstructured file data (PDF, images, etc.). When the large language model decision engine generates a sorting scheme, it can set the storage address of the file to be sorted to the address in the object storage system, so that when the subsequent execution module is executed, the file data can be stored in the object storage system. The relational database is used to store structured metadata, which includes basic file information, operation logs, and semantic summaries. When the large language model decision engine generates a sorting scheme, it can set the storage address of the metadata corresponding to the file obtained by the acquisition module to the address in the relational database, so that when the subsequent execution module is executed, the metadata can be stored in the relational database. The vector database is used to store the knowledge base, including vectorized file summaries and user feedback records. When the large language model decision engine generates a sorting scheme, it sets the storage address of the generated file summary to the address in the vector database, so that when the subsequent execution module is executed, the file summary can be stored in the vector database. After obtaining user feedback, the large language model decision engine stores it in the vector database for use in subsequent semantic understanding. The data storage module forms the data support basis for the large language model decision-making and file retrieval module. The input of the data storage module is the information of user selected files and user feedback, and the output is file metadata, file content, vectorized information, etc.

[0107] Furthermore, after the execution module executes the file organization plan, the metadata of the file changes. The file metadata also includes the new location, new name, file summary, and access control. The large language model decision engine stores the file metadata information in relational data.

[0108] The file metadata stored in relational data is shown in Table 1 below:

[0109] File metadata Data Type effect Old location text Record the location of the file before organization old name text Record the name of the file before editing New Location text Record the location of the organized files New name text The name of the file after recording File Type text Record file type File size Plastic surgery The size of the storage space occupied by the record file Document Summary text A summary text of the log file contents Upload time time Record the time when the file is uploaded to the database Creation time time Record the time the file was created Modification time time Record the time when the file was last modified Access time time Record the time when the file was last accessed Access Control text Role-based access control

[0110] Table 1 - File metadata table.

[0111] In a specific embodiment of the present invention, the acquisition module is further used to acquire the file retrieval mode during file retrieval;

[0112] The file retrieval module is used to perform retrieval according to the acquired file retrieval method, and then return a list of retrieved file information.

[0113] The file retrieval module provides users with a search function, allowing them to find organized files. It filters files that meet user requirements and sorts them according to rules. The module includes a three-level retrieval system, namely three search methods. Users can select the appropriate search method based on the complexity of the search situation and enter search criteria (which can be in natural language). The module then returns a corresponding list containing file metadata information. The module's input is the user-selected search method and search criteria, and its output is a file list containing file information.

[0114] The file retrieval module of the present invention includes three retrieval methods, namely keyword matching retrieval, semantic expansion search and summary RAG (retrieval enhancement generation) search. Keyword matching retrieval is a common retrieval method, and semantic expansion search and summary RAG (retrieval enhancement generation) search are intelligent retrieval methods. Among them, the semantic expansion search understands the user's semantics and analyzes the associated intentions through a large language model, generates a semantic graph for file information (including a semantic graph of associated entities, time ranges, and document metadata), and then converts it into a retrieval scope with multiple condition constraints, that is, a combined query SQL statement. This retrieval scope can contain the user's query purpose with a greater probability. This solution has the advantage of a high recall rate, as shown in Table 2:

[0115] index Traditional keyword search Semantic expansion search Recall 62％ 89％ False Positive 28％ 15％ Long-tail query coverage 41％ 79％

[0116] Table 2 - Performance comparison of traditional keyword search and semantic expansion search.

[0117] Abstract RAG (Retrieval-Augmented Generation) search builds a document summary knowledge base by encoding document summaries into vectors using a specialized text embedding model. HNSW, an approximate nearest neighbor search algorithm (ANN), is then used to quickly retrieve these vectors, finding and ranking the summaries that most closely match the content description. Finally, a large language model interprets, verifies, and filters the found document summaries. Compared to directly searching the full text, this summary vectorization solution reduces the index size by 87% while maintaining over 95% semantic integrity.

[0118] The query rule for semantic expansion search is that the large language model semantically expands the search requirements, forming a search intent scope and generating a set of SQL statements. The query rule for summary RAG search is to filter out a batch of document summaries from the knowledge base that are most similar to the content description provided by the user.

[0119] Users can choose the appropriate search method based on their search needs, balancing time and performance. Keyword matching is implemented using an inverted index, with the core goal of quickly and accurately matching literal keywords. Semantic expansion search is implemented using a large-scale model for semantic understanding and SQL generation, with the core goal of expanding user potential intent and improving recall. Summary RAG search is implemented using vectorized retrieval and semantic filtering, with the core goal of matching semantic similarity and accurately filtering content.

[0120] A model performance comparison experiment was conducted on the three retrieval methods and direct processing of large models, and the performance data shown in Table 3 below was obtained:

[0121]

[0122]

[0123] Table 3 - Performance comparison experimental data of four retrieval methods.

[0124] like Figure 2 As shown, the data flow of the digital governance system of the present invention is divided into four stages, namely: file upload and processing, intelligent interaction and organization, automated execution and file retrieval. The process of file upload and processing includes: the user selects the file through the local program → the local program obtains the absolute path of the file storage location → locates the file according to the absolute path → obtains the file metadata and stores it in the relational database → uploads the file to the object storage database. The process of intelligent interaction and organization includes: the user sets various functional parameters → the user inputs the organization requirements in natural language → the large language model workflow receives the user input and file content and generates an organization plan (including a file system operation instruction set) → the organization plan is displayed on the interactive interface. The process of automated execution includes: the user approves the organization plan → the user clicks the button to confirm the adoption → the system executes the file move and rename operation → the database records the file metadata changes → clears the cache. The process of file retrieval includes: the user selects the retrieval mode and enters the retrieval conditions → searches for files according to the corresponding rules and returns a sorted file list.

[0125] The following is an example of the technical framework selection of each module of the digital document management system of the present invention.

[0126] The interactive interface module implements server-side logic using a lightweight web framework (such as Flask). Its rapid development and modular extensibility make it suitable for building user-friendly natural language interfaces. The data storage module employs a layered architecture, including an object storage system (such as MinIO) for file content persistence, a relational database (such as MySQL) for storing and managing file metadata, and a vector database (such as PostgreSQL-PGvector) for storing vectorized content. MinIO was chosen for its distributed storage and high-concurrency support, while MySQL's transaction processing capabilities and support for complex queries make it suitable for the multidimensional data management needs of file management systems. PostgreSQL-PGvector supports the storage of algorithm-generated vector data and enables efficient retrieval through indexing techniques and vector similarity distance queries. Fastgpt is used to configure the large language model workflow. The decision engine uses Qwen2.5 as the pre-trained language model for semantic parsing and bge-m3 as the pre-trained index model for knowledge base retrieval. The retrieval module uses Qwen2.5-Coder as the pre-trained language model for semantic parsing and generates MySQL queries. Fastgpt was chosen for its open source nature, attractive interface, and comprehensive functionality. The Qwen series was chosen because it is an industry-leading large language model and has good support for Chinese.

[0127] The following is an example of a user usage scenario of the digital document management system of the present invention.

[0128] Scenario Background: A business department needs to organize and retrieve 500 project documents (including PDF scans, Word revisions, and Excel balance sheets) from the past three years. Traditional methods, such as manual operation of Windows Explorer, have the following problems: file classification relies on manual reading of content, which takes about 5 people per day; searching for "Requirement analysis of xx function in December 2024" requires opening each file for viewing.

[0129] The operating procedures of this system are as follows:

[0130] File upload and processing: Users upload 500 files in batches through the web interface. The system automatically extracts metadata (creation time, modification history) and stores it in MySQL, and the file content is stored in MinIO.

[0131] Natural language interactive organization: The user inputs the command: "Classify by customer name, find the file containing the demand analysis, and then classify by the modification month." Large language model decision engine analysis: Phase 1: Extract key parameters (classification dimensions: customer name, demand analysis, modification month); Phase 2: Parse the content, including unstructured PDF documents (OCR to identify the terms and conditions text) and Word revisions (extract revision comments); Phase 3: Generate a summary (such as "Customer: Company A, Requirement Analysis: bb, Modification Month: 2024.12, Requirement Content: ccc"); Phase 4: Output the organization plan (xx functional requirements.docx → / Company A / Requirement Analysis / 202412 / Company A_Requirement Analysis

[0132] _202412_bbRequirements.docx).

[0133] Automated Execution and Feedback: After the user confirmed the solution, the system automatically performed file renaming and moving, which took 12 minutes. A discrepancy was discovered in the identification of two key steps in the requirements analysis. After the user corrected the discrepancy, the system updated the feedback knowledge base ("Step xx in Company A's bb requirement specifically requires a certain architecture to achieve a specific software goal.").

[0134] File Search: If a user enters the search criteria "Find the requirements analysis document of Company A last uploaded in December 2024," selecting Semantic Extension Search will analyze the semantics (upload date 2024.12, file name: Requirements Analysis) and generate a search scope (e.g., file names containing: Requirements, Requirements Analysis, Requirements Document, Requirements Description, Requirements List, etc.), returning one file. If a user enters descriptive text (e.g., "Requirement: Complete specified software functionality according to a certain architecture to achieve high-performance transformation"), selecting Abstract RAG Search will perform a vectorized search of abstracts, finding files whose requirements descriptions are similar to the user's description, and returning four files.

[0135] The comparison of the effects of the digital document management system of the present invention and the traditional manual processing method is shown in Table 4:

[0136]

[0137]

[0138] Table 4 - Comparison of the effects of the digital document management system of the present invention and the traditional method.

[0139] The present invention also provides a digital document management method, which is described below.

[0140] The digital document management method of the present invention comprises the following steps:

[0141] When organizing files, obtain the files to be organized and the file organization instructions;

[0142] Identify file organization instructions based on a large language model, integrate the content of the files to be organized, generate summaries corresponding to the files to be organized, and then generate a file organization plan based on the set organization rules;

[0143] Upon receiving a confirmation instruction of the file arrangement plan, executing the file arrangement plan to store the files to be arranged, thereby completing the file arrangement;

[0144] When searching for a file, obtain the file search conditions;

[0145] Search the stored files according to the file search conditions and return a list of file information that meets the file search conditions.

[0146] In a specific embodiment of the present invention, when generating a file organization plan, key parameters are obtained according to the identified file organization instruction, and the obtained key parameters include classification basis, classification dimension, and target storage location;

[0147] Obtain metadata of the file to be organized, the metadata obtained including the creation time, modification time, and access time of the file to be organized;

[0148] According to the metadata and summary of each file, and based on the key parameters obtained, a storage path and a new name are assigned to each file, thereby completing the generation of a file organization plan.

[0149] In a specific embodiment of the present invention, it also includes:

[0150] Get user feedback;

[0151] Store the acquired user feedback records;

[0152] When identifying file organization instructions and file retrieval conditions based on the large language model, the file organization instructions and file retrieval conditions are understood and identified based on recorded user feedback.

[0153] In a specific embodiment of the present invention, it also includes:

[0154] Get configuration parameters;

[0155] Generate corresponding sorting rules based on the obtained configuration parameters.

[0156] In a specific embodiment of the present invention, when searching for a file, the method further includes obtaining a file search method;

[0157] According to the obtained file retrieval method, the retrieval is performed, and a list of retrieved file information is returned.

[0158] The present invention also provides a storage medium on which a program of the digital document management method is stored. When the program of the digital document management method is executed by a processor, the steps of the digital document management method are implemented.

[0159] The beneficial effects of the digital document management system and method of the present invention include:

[0160] Multi-stage workflow design based on large language models: Through a multi-stage workflow (input decomposition, content analysis, summary generation, solution generation), the output of the large language model is planned and constrained to generate a file organization solution that meets user needs.

[0161] Multi-dimensional metadata collection method based on local agent: obtain file metadata (such as creation time, modification time, access time, etc.) through local programs to provide data support for subsequent governance.

[0162] Dynamic directory creation mechanism of the automated execution module: Dynamically parses the path information in the organization plan and automatically creates the target directory to ensure that the file can be moved to the specified location.

[0163] Semantic extension search method based on large language model: parse user natural language queries through large language model to generate semantic graph containing related entities, time range, and document attributes; convert the semantic graph into a group of SQL query statements with multiple condition combinations.

[0164] RAG retrieval method based on summary vectorization: Encode the document summary into a vector through a dedicated embedding model; Retrieval implementation process: Perform vector similarity retrieval in the knowledge base → Use a large language model to semantically filter the results.

[0165] Dynamic adjustment strategy based on user feedback: Build a feedback knowledge base to record users' ratings and modification suggestions on the sorting plan, and use it to optimize the sorting strategy of the large language model.

[0166] A collaborative architecture model that supports multi-terminal access within the LAN: Build network services through a Web framework, support multi-host access within the LAN, and achieve resource integration and governance.

[0167] The present invention has been described in detail above with reference to the embodiments of the accompanying drawings. A person skilled in the art can make various modifications to the present invention based on the above description. Therefore, certain details in the embodiments should not be construed as limiting the present invention. The scope of protection of the present invention shall be determined by the scope defined in the appended claims.

Claims

1. A digital document management method, characterized in that: The steps include: When organizing files, obtain the files to be organized and the file organization instructions; Identify file organization instructions based on a large language model, integrate the content of the files to be organized, generate summaries corresponding to the files to be organized, and then generate a file organization plan based on the set organization rules; Upon receiving a confirmation instruction of the file arrangement plan, executing the file arrangement plan to store the files to be arranged, thereby completing the file arrangement; When searching for a file, obtain the file search conditions; Search the stored files according to the file search conditions and return a list of file information that meets the file search conditions.

2. The digital document management method according to claim 1, wherein: When generating a file organization plan, key parameters are obtained according to the identified file organization instructions. The key parameters obtained include classification basis, classification dimension, and target storage location. Obtain metadata of the file to be organized, the metadata obtained including the creation time, modification time, and access time of the file to be organized; According to the metadata and summary of each file, and based on the key parameters obtained, a storage path and a new name are assigned to each file, thereby completing the generation of a file organization plan.

3. The digital document management method according to claim 1, wherein: Also includes: Get user feedback; Store the acquired user feedback records; When identifying file organization instructions and file retrieval conditions based on the large language model, the file organization instructions and file retrieval conditions are understood and identified based on recorded user feedback.

4. The digital document management method according to claim 1, wherein: Also includes: Get configuration parameters; Generate corresponding sorting rules based on the obtained configuration parameters.

5. The digital document management method according to claim 1, wherein: When retrieving files, it also includes obtaining the file retrieval method; According to the obtained file retrieval method, the retrieval is performed, and a list of retrieved file information is returned.

6. A storage medium, characterized in that The storage medium stores a program of a digital document management method, and when the program of the digital document management method is executed by a processor, the steps of the digital document management method according to any one of claims 1 to 5 are implemented.

7. A digital document management system, characterized in that: include: The acquisition module is used to obtain the files to be organized and the file organization instructions when organizing files, and is also used to obtain the file search conditions when searching files; A large language model decision engine, connected to the acquisition module, configured to identify file organization instructions based on the large language model, integrate the contents of the files to be organized, generate summaries corresponding to the files to be organized, and then generate a file organization plan according to the set organization rules; Data storage module; an execution module connected to the acquisition module, the large language model decision engine, and the data storage module, and configured to, upon receiving a confirmation instruction of the file organization plan, execute the file organization plan to store the files to be organized in the data storage module, thereby completing the file organization; A file retrieval module is connected to the acquisition module and the data storage module. The file retrieval module is used to search the data storage module according to the file retrieval conditions and return a list of file information that meets the file retrieval conditions.

8. The digital document management system according to claim 7, wherein: The large language model decision engine is further configured to extract key parameters from the identified file sorting instructions, the key parameters including classification basis, classification dimension, and target storage location; The large language model decision engine is further used to obtain metadata of the files to be sorted, and the obtained metadata includes the creation time, modification time and access time of the files to be sorted; The large language model decision engine is also used to assign a storage path and a new name to each file based on the metadata and summary of each file and the key parameters obtained, thereby completing the generation of a file organization plan.

9. The digital document management system according to claim 7, wherein: The acquisition module is also used to obtain user feedback; The large language model decision engine is further used to store user feedback in the data storage module, and is further used to understand and identify file organization instructions and file retrieval conditions based on the stored user feedback.

10. The digital document management system according to claim 7, wherein: The acquisition module is also used to acquire configuration parameters; The large language model decision engine is further configured to generate corresponding sorting rules according to the acquired configuration parameters.

Images