Standard declaration-oriented multi-Agent dynamic collaborative material editing system

Abstract

The invention provides a standard declaration-oriented multi-Agent dynamic collaborative material editing system, which solves the pain point of the existing mode and realizes'efficient, compliant, collaborative and dynamic 'standard declaration material editing through multi-Agent collaboration (Agents such as demand analysis, content generation and compliance inspection), a standard knowledge base (structured standard data) and an intelligent editing tool (automatic generation and real-time collaboration).

Description

Technical Field

[0001] This invention relates to the field of information editing technology, and in particular to a multi-agent dynamic collaborative material editing system for standard application, a multi-agent dynamic collaborative material editing method, electronic device, and computer-readable storage medium for standard application. Background Technology

[0002] Currently, the editing of materials submitted by enterprises for industry standards and national standards is primarily done manually, supplemented by simple document tools (such as Word and Excel). The core process is: manually collecting enterprise information (such as product parameters and test reports) → organizing the content according to the standard text → repeatedly checking the format and compliance → sending revisions via email to multiple departments → final submission. This model has the following fatal flaws:

[0003] 1. The editing efficiency is extremely low and the cycle is lengthy.

[0004] Manual information collection requires coordination among multiple departments, including technology, quality, and administration, with each department taking 1-2 days to respond, and the collection process itself taking 3-5 days. Organizing the content requires manual structuring and text writing, taking 2-3 days. Compliance checks require line-by-line comparison with the standard text, taking 2-3 days. The editing cycle for a complex standard application typically takes 7-11 days, which cannot meet the needs of companies for timely submissions (such as submitting before the standard application deadline).

[0005] 2. Poor standard consistency and high compliance risks.

[0006] Different editors have varying interpretations of standard clauses (e.g., the definition of "energy efficiency ratio" is sometimes interpreted as "cooling capacity / input power" and sometimes as "heating capacity / input power"), leading to conflicts between material content and standard requirements. According to statistics from an industry association, the pass rate for standard compliance checks under the current method is only 70%, requiring repeated revisions and adding 1-2 days of extra cost.

[0007] 3. Difficulties in multi-department collaboration and version confusion.

[0008] When multiple departments collaborate, documents are often sent via email, which can easily lead to "version overlap" (e.g., the technical department modifies product parameters, but the editors fail to update in time, resulting in the final materials still using the old parameters) or "version loss" (e.g., email attachments are accidentally deleted). According to research, 30% of application materials need to be re-edited due to version issues, and collaboration time accounts for 40% of the total timeframe.

[0009] 4. Serious omissions in information, resulting in a high rejection rate for applications.

[0010] Manual editing can easily lead to the omission of key information in the application requirements (such as the required attachment "copy of the company's business license" and the format requirement "the main text should be in Song font, size 12"). In 2023, one company submitted 10 applications, 3 of which were rejected due to missing information, requiring the company to prepare new materials, which took 3-5 days and severely impacted the application progress.

[0011] 5. Poor dynamic adjustment capability and slow response speed.

[0012] When standards are updated (such as adding a "standby power" requirement) or application requirements change (such as adjusting the format template), all relevant materials need to be manually revised. For example, after a certain standard was updated in 2024, companies needed to revise 10 historical application materials, which took 5 days, making it impossible to respond quickly to standard changes. Summary of the Invention

[0013] To address the technical problems existing in the prior art, the present invention provides the following technical solution:

[0014] On the one hand, a multi-agent dynamic collaborative material editing system for standard declaration is provided, the system comprising:

[0015] The Agent management module is used to manage the registration, allocation, scheduling and monitoring of multiple agents, including demand analysis agent, content generation agent, standard compliance agent, collaboration coordination agent and review agent;

[0016] The user interaction module is used to enable interaction between users and the system, receive user requests, display the progress and results of material editing, and convey user feedback.

[0017] The standard knowledge base module is used to store standard-related structured data (including standard texts, application guidelines, and historical application cases). It is built using knowledge graph technology and supports semantic retrieval and related recommendations.

[0018] The materials editing module is used to generate and edit application materials based on multi-agent collaboration, including content generation, automatic formatting, insertion of attachments, and modification suggestions.

[0019] The collaboration and coordination module is used to handle collaboration among multiple users and agents, including permission management, version control, real-time synchronization, and communication tools.

[0020] The intelligent review module is used to conduct a comprehensive review of the application materials, including standard compliance check, format check, information completeness check, and logical consistency check, and generate a review report.

[0021] The version management module is used to record every revision of the application materials and supports version comparison and rollback.

[0022] The data storage module is used to store user information, standard knowledge base data, application material data, version records, and review records.

[0023] Preferably, the Agent management module uses the least connection method to allocate Agents and supports dynamic scheduling of multiple types of Agents: the demand analysis Agent is based on the BERT+CRF model, the content generation Agent is based on the GPT-4 model with fine-tuned industry data, and the standard compliance Agent is based on the knowledge graph inference engine; Prometheus+Grafana is used to monitor Agent status, and resource expansion is automatically triggered when an anomaly occurs.

[0024] Preferably, the user interaction module supports access from Web, desktop (Electron) and / or mobile (Flutter) terminals, and the interface includes a requirement input field, a material preview area, a modification feedback area, and a collaboration invitation button.

[0025] Preferably, the standard knowledge base module uses Neo4j to construct a knowledge graph, with nodes including standards, clauses, examples, application guidelines, and cases, and relationships including "contains", "references", and "complies", and semantic retrieval is implemented using Elasticsearch.

[0026] Preferably, the content generation agent in the material editing module adopts the GPT-4 model and is fine-tuned using LoRA low-rank adaptation technology; automatic typesetting uses the Pandoc tool (which supports Markdown to Word / PDF conversion), and when inserting attachments, it associates the list of required attachments in the standard knowledge base through a rule engine.

[0027] Preferably, the coordination module adopts the RBAC model for permission management: permission allocation is defined through a role-permission mapping matrix; version control adopts Git technology: recording the commit log of each modification, including the modifier, timestamp, and changed content; real-time synchronization adopts the WebSocket protocol combined with the operation transformation (OT) algorithm to represent user operations as a commutative sequence, and conflict merging is achieved through the transform function library.

[0028] Preferably, the standard compliance check in the intelligent review module adopts the BERT-base model, pre-trained on ≥8 million Chinese standard text corpora, and the fine-tuning data contains ≥5000 pairs of "material fragments-standard clauses"; the format check adopts a regular expression engine + PDFBox parsing; the information integrity check adopts template matching, that is, based on the application guide XML template; the logical consistency check adopts knowledge graph path reasoning, based on the TransE model to map entity relationships to a low-dimensional vector space, and optimizes the objective function d(h+r,t)→0.

[0029] On the other hand, a multi-agent dynamic collaborative material editing method for standard-oriented declarations is provided, implemented based on the aforementioned multi-agent dynamic collaborative material editing system for standard-oriented declarations. The method includes the following steps:

[0030] (1) User demand initiation and parsing: Users input their demand through the user interaction module, and the demand analysis agent uses the BERT+CRF (bidirectional encoder representation from transformer + conditional random field) model to parse and generate a structured demand document;

[0031] (2) Agent allocation and collaboration preparation: The Agent management module allocates multiple Agents and monitors their status to ensure that resources meet the requirements;

[0032] (3) Content generation and preliminary layout: The content generation agent extracts standard terms and product parameters, and uses the fine-tuned GPT-4 (generative pre-trained converter 4) model to generate and layout the first draft;

[0033] (4) Standard compliance check and dynamic adjustment: The standard compliance agent checks compliance through knowledge graph and BERT-base (basic version bidirectional encoder representation from converter) model, and makes dynamic adjustments based on user feedback;

[0034] (5) Multi-department collaborative editing: The collaboration and coordination module realizes real-time collaboration among multiple users based on the RBAC (role-based access control) model and OT (operation conversion algorithm);

[0035] (6) Final review and submission: After reviewing the Agent's multimodal check, the system automatically submits and records the status.

[0036] On the other hand, an electronic device is provided, comprising: a processor; and a memory storing computer-readable instructions, which, when executed by the processor, implement any of the methods in the multi-agent dynamic collaborative material editing system for standard-oriented applications described above.

[0037] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement any of the methods in the above-described multi-Agent dynamic collaborative material editing system for standards declaration.

[0038] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0039] This invention addresses the pain points of existing methods by employing multi-agent collaboration (agents for requirements analysis, content generation, compliance checks, etc.), a standard knowledge base (structured standard data), and intelligent editing tools (automatic generation and real-time collaboration), achieving "efficient, compliant, collaborative, and dynamic" editing of standard application materials. It completely resolves the problems of low efficiency, poor consistency, and difficulty in collaboration inherent in existing standard application material editing methods, significantly improving the efficiency and success rate of enterprise standard applications, and possessing extremely high commercial value and promotional significance.

[0040] This system addresses the pain points of existing methods, and its technical effects are as follows:

[0041] 1. Improved editing efficiency

[0042] The existing editing process takes 7-11 days, but this system shortens the cycle to 2 days through automatic requirements analysis, content generation, and compliance checks (requirements analysis 0.5 days → content generation 0.5 days → compliance check 0.5 days → collaborative modification 0.5 days).

[0043] 2. Improved standard consistency

[0044] This system improves the pass rate to 95% through a standard knowledge base and intelligent auditing (with a 100% compliance rate for key clauses such as energy efficiency ratio and standby power).

[0045] 3. Improved collaboration efficiency

[0046] This system reduces collaboration time to 20% through real-time synchronized editing and access control (e.g., the technical department only needs 0.5 days to modify product parameters, without the need for email transmission).

[0047] 4. Reduced information omission rate

[0048] This system uses intelligent auditing for comprehensive checks, increasing the pass rate to 98% (e.g., the omission rate of attachments has been reduced from 20% to 2%).

[0049] 5. Dynamically adjust speed boost

[0050] The existing standard update takes 5 days to modify. This system shortens the modification time to 1 day by automatically updating the standard knowledge base and dynamically adjusting multiple agents (for example, when a new "standby power" requirement is added, the content generation agent automatically adds this parameter). Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the structure of a hardware system provided in an embodiment of the present invention;

[0053] Figure 2 This is a block diagram of a multi-Agent dynamic collaborative material editing system for standard declaration provided by an embodiment of the present invention;

[0054] Figure 3 This is a flowchart of a multi-Agent dynamic collaborative material editing method for standard declaration provided by an embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0056] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0057] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0058] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0059] In this embodiment of the invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0060] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0061] This invention provides a multi-agent dynamic collaborative material editing system and method for standard declarations. The method can be implemented by an electronic device, which can be a terminal or a server.

[0062] This system consists of a hardware system and a software system. The hardware provides the foundation for the operation of the software, while the software is the carrier of core functions.

[0063] I. Hardware System

[0064] like Figure 1 As shown, the hardware system communication connection adopts a layered architecture design: the cloud server and the local server cluster establish an encrypted channel (tunneling protocol ESP (Encapsulating Security Payload), encryption algorithm AES-256 (Advanced Encryption Standard 256-bit)) via IPSec VPN (Internet Protocol Secure Virtual Private Network) to achieve cross-regional data transmission; the local server and network devices are connected via Gigabit Ethernet (TCP / IP protocol), and the switch divides VLANs (Virtual Local Area Networks) to isolate different functional areas (such as management VLAN and service VLAN); the storage device connects to the server cluster via iSCSI (Internet Small Computer System Interface) protocol, supporting block-level data access. The firewall is configured with NAT (Network Address Translation) rules and access control lists (ACLs) to allow only traffic on specified ports (such as port 8080 for Web services and port 3306 for MySQL access) to pass through, and prohibits unauthorized access.

[0065] The relationship between hardware and software systems is realized through service interfaces and data buses: cloud servers run core services such as Agent management modules and standard knowledge base modules, and provide Agent scheduling and knowledge retrieval capabilities to software systems through RESTful APIs (Representation State Transition Application Programming Interfaces); local server clusters deploy containerized software modules (such as material editing modules and intelligent review modules) using Docker (containerization platform), and perform container orchestration through Kubernetes (container orchestration system); the MySQL (relational database management system) cluster in the storage device interacts with the data storage module through the JDBC (Java Database Connectivity) interface, and Alibaba Cloud OSS (Object Storage Service) connects with the attachment management service through the SDK (Software Development Kit) to achieve unified storage and access of structured and unstructured data.

[0066] In this embodiment, the following hardware system configuration scheme can be referenced, adopting a "cloud + local" hybrid architecture to balance elastic scalability and data security. The specific configuration is as follows:

[0067] Components Selection and Configuration Function Description cloud server Alibaba Cloud ECS (Elastic Compute Service, 4 cores, 8GB memory, 100GB SSD storage, 5Mbps bandwidth) Run core services (Agent management, standard knowledge base, collaboration and coordination) and support elastic scaling (such as temporarily adding instances during peak application periods). Local server cluster Master-slave architecture (Master server: 2 Intel Xeon E5-2680 v4 CPUs, 64GB RAM, 1TB SSD; Slave server: same configuration as master server) The main server stores sensitive data (such as core technical parameters of an enterprise), runs services, and the slave servers are responsible for backup and load balancing. Network equipment Gigabit router (Huawei AR650), enterprise-grade switch (Huawei S5735), firewall (Sangfor NGAF) Ensure network stability (gigabit bandwidth supports real-time collaboration among multiple users), and configure firewall access control policies (allowing only internal enterprise IPs (Internet Protocol) to access). storage devices MySQL (Relational Database Management System) cluster (3 nodes, each node with 1TB SSD), Alibaba Cloud OSS (Object Storage Service, 10TB storage space, redundant storage) MySQL (Relational Database Management System) stores structured data (such as user information and standard terms); OSS (Object Storage Service) stores unstructured data (such as test report PDFs (Portable Document Format) and product images), and supports automatic backup (daily full backup + hourly incremental backup).

[0068] II. Software System

[0069] like Figure 2The diagram shown illustrates the components of a multi-agent dynamic collaborative material editing system for standards-oriented applications. The system includes:

[0070] The Agent management module is used to manage the registration, allocation, scheduling and monitoring of multiple agents, including demand analysis agent, content generation agent, standard compliance agent, collaboration coordination agent and review agent;

[0071] The user interaction module is used to enable interaction between users and the system, receive user requests, display the progress and results of material editing, and convey user feedback.

[0072] The standard knowledge base module is used to store standard-related structured data (including standard texts, application guidelines, and historical application cases). It is built using knowledge graph technology and supports semantic retrieval and related recommendations.

[0073] The materials editing module is used to generate and edit application materials based on multi-agent collaboration, including content generation, automatic formatting, insertion of attachments, and modification suggestions.

[0074] The collaboration and coordination module is used to handle collaboration among multiple users and agents, including permission management, version control, real-time synchronization, and communication tools.

[0075] The intelligent review module is used to conduct a comprehensive review of the application materials, including standard compliance check, format check, information completeness check, and logical consistency check, and generate a review report.

[0076] The version management module is used to record every revision of the application materials and supports version comparison and rollback.

[0077] The data storage module is used to store user information, standard knowledge base data, application material data, version records, and review records.

[0078] The software system adopts a microservice architecture (Spring Cloud (microservice framework)), which breaks down core functionalities into independent modules, supporting flexible expansion and maintenance. The specific modules are as follows:

[0079] 1. Agent Management Module --- Agent registration, allocation, scheduling, and monitoring; maintenance of the agent pool (agents for demand analysis, content generation, etc.).

[0080] The system employs a load balancing algorithm (least connections) to allocate agents and supports dynamic scheduling of multiple agent types: demand analysis agent (based on an intent recognition model using BERT+CRF (bidirectional encoder representation from transformer + conditional random field), content generation agent (using a GPT-4 (generative pre-trained transformer 4) model fine-tuned from industry standard data), standard compliance agent (knowledge graph association reasoning engine), and collaboration coordination agent (based on a real-time message bus using WebSocket). Agent status is monitored via Prometheus (monitoring system) + Grafana (visualization platform) (CPU (central processing unit) utilization < 80%, memory utilization < 70%, response time < 500ms), and agent restart or resource expansion is automatically triggered in case of anomalies.

[0081] 2. User Interaction Module --- Supports access via Web (World Wide Web), desktop (Electron (a cross-platform desktop application development framework)), and mobile (Flutter (a cross-platform mobile application development framework)); provides functions for requirement input, material preview, modification feedback, and collaboration invitation.

[0082] The web (World Wide Web) uses the React (front-end JavaScript library) framework, the desktop uses Electron (a cross-platform desktop application development framework), and the mobile uses Flutter (a cross-platform mobile application development framework); the interface includes a "requirement input field", "material preview area", "modification feedback area", and "collaboration invitation button".

[0083] 3. Standard Knowledge Base Module --- Stores structured standard data (standard text, application guidelines, historical cases); supports semantic retrieval and related recommendations.

[0084] Use Neo4j (graph database) to build a knowledge graph (nodes: standard, clause, example, case; relations: "contain", "reference", "comply"); use Elasticsearch (search engine) to implement semantic retrieval (such as "What are the application requirements for XX standard?").

[0085] 4. Materials Editing Module --- Automatically generates initial draft materials (combining standards and user information), automatically formats the materials (conforming to the application format), and inserts attachments (linked to the required list).

[0086] The content generation agent uses the GPT-4 (Generative Pre-trained Transformer 4) model (training data includes 5000+ industry standard texts and 3000+ historical application cases, with fine-tuned parameters: learning rate 2e-5 and training epochs 10); automatic formatting uses Pandoc (a document conversion tool) (converting Markdown (lightweight markup language) to Word (word processing software) / PDF (portable document format), supporting custom style templates); attachment insertion uses a rule engine (based on the "required attachment list" in the Drools (rule engine) rule matching standard knowledge base, triggering user reminders when attachments are missing).

[0087] 5. Collaboration and Coordination Module --- Permission Management (RBAC (Role-Based Access Control) model), Version Control (Git (Distributed Version Control System)), Real-time Synchronization (WebSocket (Network Socket Protocol)), Built-in Communication Tools (Chat + Comments).

[0088] Access control: Roles are divided into administrators, editors, reviewers, and observers (e.g., editors can modify product parameters); Version control: Records each commit of changes (e.g., "2024-05-01 10:00, Zhang San modified the product parameters"); Real-time synchronization: Uses WebSocket (network socket protocol) to merge content when multiple users edit simultaneously (e.g., A modifies the main text, B modifies the attachment, and the system automatically merges them).

[0089] 6. Intelligent Audit Module --- Standard Compliance Check (compare with standard clauses), Format Check (regular expression), Information Integrity Check (template matching), Logical Consistency Check (knowledge graph association).

[0090] The standard compliance agent uses a BERT-base model (pre-trained on a Chinese standard text corpus) for clause matching, achieving an F1-score of 92%. Format checks are performed using regular expressions (e.g., matching "body text in Songti, size 12" and "page numbers centered"). Information integrity checks are performed using template matching (based on XML (Extensible Markup Language) templates generated from the declaration guidelines). Logical consistency checks are performed using knowledge graph-based reasoning (e.g., verifying the formula relationship "cooling capacity / input power = energy efficiency ratio").

[0091] 7. Version Management Module --- Records every modification to the application materials and supports version comparison and rollback.

[0092] The version management module uses Git (a distributed version control system) to store version history; the web interface integrates Diff (difference comparison tool) tools (such as CodeMirror (code editor component)) to display the differences between two versions (such as "The 'standby power' parameter has been added to the 3rd paragraph of the text"); the rollback function allows users to restore to a previous version with one click (such as "Restore to the version at 10:00 on 2024-05-01").

[0093] 8. Data storage module --- stores user information (MySQL), standard knowledge base data (Neo4j + Elasticsearch), application materials (MySQL + OSS), version records (Git), and review records (MySQL).

[0094] User information is stored in encrypted form (password is hashed using BCrypt); standard knowledge base data is updated regularly (e.g., by connecting to the API of the National Standardization Committee to automatically synchronize standard revision information); when storing application materials, the main text is stored in MySQL (Markdown format), and attachments are stored in OSS (attachment ID associated with the main text).

[0095] III. Material Editing Method Based on the System

[0096] Based on the above system, the editing method includes the following steps:

[0097] 1. User Request Initiation and Analysis

[0098] Users input their application requirements through the user interaction module, including application type, standard number, deadline, and special requirements. The requirement analysis agent uses a BERT+CRF (bidirectional encoder representation from transformer + conditional random field) model (training set contains 50,000 industry application corpora, entity recognition accuracy of 91%) to analyze the requirements. Combined with the TextCNN (text convolutional neural network) intent classification model (F1-score of 89%), a structured requirement document is generated, which clarifies the information collection list, content structure, and format requirements.

[0099] 2. Agent allocation and collaboration preparation

[0100] The Agent management module allocates multiple Agents based on demand type (demand analysis, content generation, standards compliance, etc.), uses the least connection method to schedule resources, and monitors Agent status (CPU (central processing unit) utilization < 80%, memory utilization < 70%, response time < 500ms) through Prometheus (monitoring system) + Grafana (visualization platform), and automatically triggers expansion when anomalies occur.

[0101] 3. Content generation and initial layout

[0102] The content generation agent extracts structured terms (such as "energy efficiency ratio ≥ 3.8") from the standard knowledge base, calls the enterprise API (Application Programming Interface) to obtain real-time parameters, and uses a fine-tuned GPT-4 (Generative Pre-trained Transformer 4) model (LoRA (Low Rank Adaptation) technology r=16, α=32, content accuracy 92%) to generate a draft. The draft is then automatically formatted using the Pandoc (document conversion tool) tool (SimSun font, size 12, line spacing 1.5x), and the rule engine associates it with a list of necessary attachments.

[0103] 4. Standards compliance checks and dynamic adjustments

[0104] The standard compliance agent uses a knowledge graph reasoning engine (entity extraction F1-score 88%) and a BERT-base model (basic bidirectional encoder representation from the converter) (accuracy 90% when cosine similarity ≥ 0.9) for collaborative checks: extracting key parameters and comparing them with clause thresholds, and the Drools (rule engine) verifying the completeness of attachments; after generating the check report, users provide feedback and suggestions for modification, and the content generation agent automatically updates and re-triggers the check until compliance is achieved.

[0105] 5. Multi-department collaborative editing

[0106] The collaboration and coordination module allocates permissions based on the RBAC (role-based access control) model (e.g., the technical department only modifies parameters), and achieves real-time synchronization through the WebSocket (network socket protocol) + OT (operation conversion algorithm) algorithm (99.2% success rate in conflict resolution, latency <100ms); multiple user modifications are detected for conflicts through an optimistic locking mechanism, and the OT (operation conversion algorithm) merges operations to ensure version consistency.

[0107] 6. Final review and submission

[0108] The audit agent employs multimodal checks: BERT-base (basic bidirectional encoder representing the converter) verifies clause matching (90% accuracy), regular expressions + PDFBox (PDF document parsing tool) verify format, TransE (transformation embedded model) model inference logic consistency (89% accuracy), and Drools (rule engine) verifies attachments; an audit report is generated, and after user confirmation, the system automatically submits it to the application platform, records the status, and sends a reminder.

[0109] Example: The core logic of this system is "user demand driven → multi-agent collaboration → dynamic adjustment → final review". The specific process is as follows (taking "applying for 2024 industry standard xxx" as an example):

[0110] Step 1: User initiates application request

[0111] Users log in to the system via the web interface and enter the following in the "Requirements Input Field":

[0112] Application type: Industry standard

[0113] Standard Number: XX-2024

[0114] Deadline: October 31, 2024

[0115] Special requirements: The test report must include the "standby power" parameter and must be stamped with an official seal.

[0116] The user interaction module passes the requirements to the Agent management module.

[0117] Step 2: Agent Allocation and Requirements Analysis

[0118] The Agent management module allocates the following Agents from the Agent pool based on the demand type (industry standard declaration):

[0119] Demand Analysis Agent: Based on the BERT+CRF named entity recognition model, its algorithm principle is as follows: BERT (Bidirectional Encoder Representations from Transformers) captures contextual semantic features through a 12-layer bidirectional Transformer encoder. In the pre-training stage, it adopts the masked language model (MLM) and next sentence prediction (NSP) tasks; CRF (Conditional Random Field) optimizes the sequence labeling results by learning the label transition probability matrix (such as the probability of "parameter name" followed by "value" is 0.92), thus solving the problem of ambiguous entity boundaries. The training process is as follows: ① Training data: Collect the corpus of industry standard application from 2019 to 2023 (50,000 user requirement texts + manually annotated entities, with entity types including "standard number", "parameter name", "deadline" and 8 other categories); ② Pre-training: Use the Chinese BERT-base pre-trained model (bert-base-chinese), freeze the parameters of the bottom 8 layers, fine-tune the top 4 layers and the CRF layer, with a learning rate of 5e-5, a batch size of 32, and 20 training epochs. Use early stopping (stop if the F1-score on the validation set does not improve for 3 consecutive epochs); ③ Validation: Use 10% of the corpus as the test set, and the entity recognition accuracy reaches 91% (precision 90.5%, recall 91.3%, F1-score 90.9%). The application steps are as follows: 1. Text preprocessing (Jieba word segmentation + custom industry dictionary, removing special characters such as "#" and "@"); 2. Entity extraction (inputting into the BERT model to generate word vectors, the CRF layer outputs the optimal label sequence, extracting entities such as "standby power" and "October 31, 2024"); 3. Intent classification (the TextCNN (text convolutional neural network) model extracts local features through 128 / 256 / 512-dimensional convolutional kernels, softmax (normalized exponential function) outputs the probability distribution of the declaration type, and the F1 score is 89%); 4. Entity linking (calculating the cosine similarity between the entity and the standard clause based on ELMo (embedded language model) word vectors, with a threshold of 0.85 to achieve the mapping from "standby power" to clause 4.2 of XX-2024), finally generating a structured requirements document.

[0120] Step 3: Content Generation and Initial Layout

[0121] The content generation agent receives the requirement document, which is extracted from the standard knowledge base:

[0122] Standard clauses: "XX-2024 Clause 4.1: Energy efficiency ratio ≥ 3.8; Clause 4.2: Standby power ≤ 0.5W";

[0123] Example: A successful case study of a company's application in 2023 (content structure, format template);

[0124] Information provided by the user: Company name "XX Air Conditioning Co., Ltd.", product model "KFR-35GW / XX", cooling capacity 3500W, input power 1000W (energy efficiency ratio 3.5), test report number "XX-2024-001".

[0125] The content generation agent uses the GPT-4 model. Its algorithm works as follows: GPT-4 (Generative Pre-trained Transformer 4) is based on a Transformer decoder architecture with 1.8 trillion parameters. It models long text dependencies through a self-attention mechanism (using FlashAttention to optimize computational efficiency). The pre-training stage includes multimodal data such as books and websites. LoRA (Low-Rank Adaptation) technology decomposes the weight update amount into W=W0+BA (B∈Rd×r, A∈Rr×d) through low-rank matrix factorization (rank parameter r=16, α=32), training only the adapter matrices B and A (reducing the parameter size by 99.9%), thus avoiding catastrophic forgetting. The training and validation process is as follows: ① Training data: Construct an industry standard corpus (5000+ industry / national standard texts, 3000+ historical successful application cases, manually cleaned and deduplicated, and labeled with key structural fields); ② Fine-tuning: Based on the GPT-4 base model, freeze the main parameters of the model, train only the adapter layer, with a learning rate of 2e-5, a batch size of 16, 10 training epochs, and the AdamW optimizer (Adam optimizer with weight decay, weight decay of 1e-2); ③ Validation: Randomly select 300 application cases that did not participate in the training as the test set, and manually evaluate the "structural integrity" (whether all required sections are included), "parameter accuracy" (the degree of matching between numerical values ​​and standard clauses), and "format compliance" (font / line spacing, etc.) of the generated content. The overall accuracy rate reached 92% (structural integrity 95%, parameter accuracy 91%, format compliance 93%). The application steps are as follows: 1. Clause extraction (using Cypher (a cryptographic query language) to retrieve structured requirements such as "energy efficiency ratio ≥ 3.8" from the standard knowledge base); 2. Parameter acquisition (retrieving dynamic data such as cooling capacity and input power in real time via the enterprise API (Application Programming Interface); 3. Structure organization (using the Jinja2 template engine to nest standard chapter frameworks and enterprise data variables); 4. Multi-round generation (using autoregressive sampling with a temperature coefficient of 0.7, first generating an outline and then filling in details, with parameter logic validated by a rule engine after each round of generation), generating at a speed of 300 words / second, with a content accuracy of 92%, including:

[0126] Main body: Introduction (introducing the company's qualifications), Company overview (establishment time, registered capital), Product description (model, function), Energy efficiency parameters (energy efficiency ratio 3.5, standby power 0.5W), Test report (number and stamp instructions);

[0127] Attachments: "Copy of Business License" and "Test Report XX-2024-001";

[0128] Formatting: Automatically apply standard requirements (SimSun font, size 12, line spacing 1.5).

[0129] Step 4: Standards Compliance Check

[0130] The standard compliance agent works in collaboration with the BERT-base model (a basic bidirectional encoder representing a transformer) through a knowledge graph-based inference engine. The algorithm works as follows: the knowledge graph uses an entity-relation-entity triplet structure; entity extraction uses BERT + BiLSTM (Bidirectional Long Short-Term Memory Network) + CRF model (BERT generates contextual word vectors, BiLSTM captures sequence features, and CRF learns label transfer rules); relation extraction uses remote supervision combined with Few-Shot learning (automatically labeling training data based on entity pair co-occurrence relations). The BERT-base model calculates cosine similarity (cosθ=A·B / (||A||·||B||)) using the 768-dimensional vector representation of the CLS token (classification label) to measure the semantic matching degree of the text. Technical details are as follows: ① Knowledge graph construction: The training data consists of 20,000 labeled standard clause texts. Entity extraction achieves an F1-score of 88%, and relation extraction achieves an F1-score of 85%, constructing a knowledge graph containing 100,000+ nodes and 200,000+ relations; ② BERT-base model training: Pre-trained on a Chinese standard text corpus (8 million standard clauses and application guide texts). Fine-tuning uses 5,000 pairs of "material fragment - standard clause" matching data. The optimization objective is cosine similarity loss, with a learning rate of 3e-5 and 15 training epochs. Epochs (training rounds), the cosine similarity threshold of 0.85 is determined on the validation set (at this point, the accuracy is 90% and the recall is 88%); ③ Application process: 1. Text preprocessing (Jieba word segmentation tool loads a custom industry dictionary containing terms such as "energy efficiency ratio" and "standby power"); 2. Entity recognition (calling the BERT+CRF model to extract key parameters such as "energy efficiency ratio 3.5"); 3. Clause matching (BERT-base model calculates the similarity between the material fragment and the standard clause, ≥0.85 indicates a successful match); 4. Threshold comparison (knowledge graph Cypher query language is used to obtain the clause threshold, such as MATCH (n: clause {number: 'XX-2024 Article 4.1'}) RETURN n. threshold); 5. Attachment verification (Drools rule engine matches the "required attachment list" in the knowledge base).

[0131] Energy efficiency ratio: 3.5 < 3.8 (does not meet clause 4.1);

[0132] Standby power: 0.5W (compliant with clause 4.2);

[0133] Attachment: Does not include a "copy of the business license" (does not comply with the application guidelines).

[0134] Generate a compliance inspection report, mark "Energy efficiency ratio not up to standard" and "Business license attachment missing", and propose modification suggestions ("Increase energy efficiency ratio to ≥3.8" and "Supplement with a copy of business license").

[0135] Step 5: User Modification and Dynamic Adjustment

[0136] Users reviewed the initial draft and compliance check report, and provided feedback for revisions.

[0137] The energy efficiency ratio will be adjusted to 3.8 (new product parameters need to be provided by the technical department).

[0138] Please provide a copy of the company's business license (provided by the administrative department).

[0139] Modification suggestions are passed to the content generation agent through the user interaction module. The content generation agent then...

[0140] Adjust the energy efficiency parameters (energy efficiency ratio 3.8);

[0141] Add "Copy of Business License" to the attachment list;

[0142] The main text is updated synchronously ("Energy efficiency ratio of 3.8 meets the requirements of Clause 4.1 of XX-2024").

[0143] The standard compliant agent was re-checked and confirmed:

[0144] Energy efficiency ratio 3.8 (compliant with clause 4.1);

[0145] Standby power is 0.5W (compliant with clause 4.2);

[0146] The attachments include a "copy of the company's business license" and a "test report" (which complies with the application guidelines).

[0147] Step 6: Multi-department collaborative editing

[0148] Users need to have the technical department confirm the product parameters and call the collaboration coordination agent through the collaboration coordination module. The technical implementation includes: ① Real-time message bus: Based on the WebSocket protocol (Network Socket Protocol) full-duplex communication mechanism, the server and client achieve bidirectional real-time data transmission through TCP (Transmission Control Protocol) long connection. It adopts the publish-subscribe model (Pub / Sub), and each editing operation is used as an event topic (such as "parameter modification event"). Subscribers (participating users) receive event notifications in real time. Operation Transformation (OT Algorithm) Principle: By representing user operations (such as inserting / deleting text) as a commutative sequence of operations, when multiple users operate concurrently, the OT algorithm converts local operations into equivalent operations based on remote versions (e.g., if user A inserts "3600W" at position 5 and user B deletes "3500W" at position 5, the algorithm automatically converts it into user B's deletion operation being executed after user A's insertion). The training data simulates 20 types of operation conflict scenarios with 100 concurrent users, and the validation set has a conflict resolution success rate of 99.2% (synchronization latency < 100ms); ② Access Control Model: RBAC (Role-Based Access Control) defines a role-permission mapping matrix (e.g., the "Editor" role is associated with the "Modify Product Parameters" permission). During access, the intersection of user roles and resource permissions is verified. The training set contains 500 permission configuration cases, and the verification accuracy is 100%. The application steps are as follows: 1. Conflict detection (optimistic locking mechanism compares version number + timestamp; if versions are inconsistent, conflict resolution is triggered); 2. Operation transformation (calling the transform function library to perform vector space transformation on concurrent operations); 3. Data merging (executes operations according to the transformed sequence, updating document version); 4. Real-time synchronization (WebSocket pushes the merged results to all collaborating users). After logging into the system, technical department users can then:

[0149] Locate the "Product Parameters" section in the "Materials Preview Area";

[0150] Change "Cooling capacity" to 3600W (input power 1000W, energy efficiency ratio 3.6 → adjust to input power 947W, energy efficiency ratio 3.8).

[0151] Click "Save," and the changes will be synchronized to the user in real time via WebSocket (Network Socket Protocol).

[0152] After the user reviews and confirms the changes, the content generation agent automatically updates the main text and energy efficiency parameters.

[0153] Step 7: Final Review and Submission

[0154] The review agent employs a multimodal fusion inspection mechanism. The algorithm principles of each sub-model are as follows: ① BERT-base (basic version bidirectional encoder representation from converter) clause matching model: It uses a 768-dimensional vector of CLS token (classification tag) to represent material fragments and standard clauses, calculates cosine similarity (cosθ=A·B / (||A||·||B||)) to determine semantic matching degree, and is pre-trained on 8 million Chinese standard text corpora; ② TransE (transformation embedding model) knowledge graph reasoning model: It maps entities and relations to a low-dimensional vector space, and learns reasoning rules by optimizing the objective function d(h+r,t)→0 (h: head entity vector, r: relation vector, t: tail entity vector), such as "cooling capacity / input power = energy efficiency ratio" is represented as h(cooling capacity)+r(divided by)=t(energy efficiency ratio); ③ Drools (rule engine) rule engine: It constructs attachment inspection logic based on production rules (IF-THEN) (such as IF declaration type=industry standard THEN required attachments include business license). Training and validation were conducted as follows: ① BERT model: Fine-tuning data included 5000 pairs of "material fragments - standard clauses", learning rate 3e-5, training for 15 epochs, accuracy 90% and recall 88% when cosine similarity ≥ 0.9; ② TransE model: Training was conducted with 20,000 parameter relation triples, vector dimension 100, batch size 128, training for 50 epochs, inference accuracy 89%; ③ Rule engine: 200 attachment rules were constructed, training set included 500 cases with missing attachments, matching accuracy 98%. The application steps are as follows: 1. Distributed scheduling (the Spark framework (a distributed computing framework) decomposes the inspection task into parallel subtasks such as clause matching and format validation); 2. Multimodal inspection (BERT matches clauses, TransE validates parameter logic, and the rule engine validates attachments); 3. Result fusion (a weighted voting mechanism integrates the results of each sub-model, with weight allocation: clause matching 0.4, logic validation 0.3, and attachment inspection 0.3); 4. Report generation (summarizing anomalies and modification suggestions). The inspection time for a single document is ≤30 seconds.

[0155] Standard compliance: Energy efficiency ratio 3.8 (compliant with clause 4.1), standby power 0.5W (compliant with clause 4.2);

[0156] Format: SimSun font, size 12, line spacing 1.5 (meets requirements);

[0157] Information completeness: Includes business license and test report (meets requirements);

[0158] Logical consistency: The product parameters are consistent with the data in the test report (cooling capacity 3600W).

[0159] The system generates an audit report ("Complies with XX-2024 standard application requirements"). After user confirmation, the user clicks "Submit". The system then...

[0160] Upload the materials to the standard application platform (such as the official website of the National Standardization Administration).

[0161] Record the submission date (October 25, 2024) and the application status ("Submitted");

[0162] Send a reminder ("Application materials have been submitted and the review is expected to be completed within 15 working days").

[0163] Figure 3 This is a flowchart illustrating a multi-agent dynamic collaborative material editing method for standard declarations, according to an exemplary embodiment. The method includes the following steps:

[0164] (1) User demand initiation and parsing: Users input their demand through the user interaction module, and the demand analysis agent uses the BERT+CRF (bidirectional encoder representation from transformer + conditional random field) model to parse and generate a structured demand document;

[0165] (2) Agent allocation and collaboration preparation: The Agent management module allocates multiple Agents and monitors their status to ensure that resources meet the requirements;

[0166] (3) Content generation and preliminary layout: The content generation agent extracts standard terms and product parameters, and uses the fine-tuned GPT-4 (generative pre-trained converter 4) model to generate and layout the first draft;

[0167] (4) Standard compliance check and dynamic adjustment: The standard compliance agent checks compliance through knowledge graph and BERT-base (basic version bidirectional encoder representation from converter) model, and makes dynamic adjustments based on user feedback;

[0168] (5) Multi-department collaborative editing: The collaboration and coordination module realizes real-time collaboration among multiple users based on the RBAC (role-based access control) model and OT (operation conversion algorithm);

[0169] (6) Final review and submission: After reviewing the Agent's multimodal check, the system automatically submits and records the status.

[0170] Please refer to the previous examples for a detailed understanding of the specific implementation steps.

[0171] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 4 As shown, the electronic device may include the above-mentioned Figure 3The illustrated method is a multi-agent dynamic collaborative material editing approach for standard-compliant applications. Optionally, electronic device 410 may include a first processor 2001.

[0172] Optionally, the electronic device 410 may also include a memory 2002 and a transceiver 2003.

[0173] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0174] The following is combined with Figure 4 A detailed description of each component of electronic device 410 is provided below:

[0175] The first processor 2001 is the control center of the electronic device 410. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0176] Optionally, the first processor 2001 can perform various functions of the electronic device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0177] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 are shown in the diagram.

[0178] In a specific implementation, as one example, the electronic device 410 may also include multiple processors, for example... Figure 4 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0179] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0180] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0181] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0182] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0183] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0184] It should be noted that, Figure 4 The structure of the electronic device 410 shown does not constitute a limitation on the router. Actual knowledge structure identification devices may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0185] Furthermore, the technical effects of the electronic device 410 can be referenced from the technical effects of the multi-Agent dynamic collaborative material editing system for standard declaration described in the above method embodiments, and will not be repeated here.

[0186] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0187] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0188] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0189] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0190] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0191] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0192] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0193] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0194] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0195] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0197] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0198] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-agent dynamic collaborative material editing system for standard declaration, characterized in that, The system includes: The Agent management module is used to manage the registration, allocation, scheduling and monitoring of multiple agents, including demand analysis agent, content generation agent, standard compliance agent, collaboration coordination agent and review agent; The user interaction module is used to enable interaction between users and the system, receive user requests, display the progress and results of material editing, and convey user feedback. The standard knowledge base module is used to store standard-related structured data (including standard texts, application guidelines, and historical application cases). It is built using knowledge graph technology and supports semantic retrieval and related recommendations. The materials editing module is used to generate and edit application materials based on multi-agent collaboration, including content generation, automatic formatting, insertion of attachments, and modification suggestions. The collaboration and coordination module is used to handle collaboration among multiple users and agents, including permission management, version control, real-time synchronization, and communication tools. The intelligent review module is used to conduct a comprehensive review of the application materials, including standard compliance check, format check, information completeness check, and logical consistency check, and generate a review report. The version management module is used to record every revision of the application materials and supports version comparison and rollback. The data storage module is used to store user information, standard knowledge base data, application material data, version records, and review records.

2. The multi-agent dynamic collaborative material editing system for standard declaration as described in claim 1, characterized in that, The Agent management module uses the least connections method to allocate Agents and supports dynamic scheduling of multiple Agent types: demand analysis Agents are based on the BERT+CRF model, content generation Agents are based on the GPT-4 model with fine-tuned industry data, and standard compliance Agents are based on the knowledge graph inference engine; Prometheus+Grafana is used to monitor Agent status, and resource expansion is automatically triggered when anomalies occur.

3. The multi-agent dynamic collaborative material editing system for standard declaration as described in claim 1, characterized in that, The user interaction module supports access via Web, desktop (Electron), and / or mobile (Flutter) devices. The interface includes a requirement input field, a material preview area, a modification feedback area, and a collaboration invitation button.

4. The multi-agent dynamic collaborative material editing system for standard declaration as described in claim 1, characterized in that, The standard knowledge base module uses Neo4j to build a knowledge graph. The nodes include standards, clauses, examples, application guidelines and cases. The relationships include "contains", "references" and "complies". Semantic retrieval is implemented using Elasticsearch.

5. The multi-agent dynamic collaborative material editing system for standard declaration as described in claim 1, characterized in that, The content generation agent in the material editing module adopts the GPT-4 model and is fine-tuned using LoRA low-rank adaptation technology; automatic typesetting uses the Pandoc tool (which supports Markdown to Word / PDF conversion), and when inserting attachments, it associates the list of required attachments in the standard knowledge base through a rule engine.

6. The multi-agent dynamic collaborative material editing system for standard declaration as described in claim 1, characterized in that, The coordination module uses the RBAC model for permission management: permission allocation is defined through a role-permission mapping matrix; version control uses Git technology: it records the commit log of each modification, including the modifier, timestamp, and changes. Real-time synchronization uses the WebSocket protocol combined with the Operation Transform (OT) algorithm to represent user operations as a commutative sequence and uses the transform function library to merge conflicts.

7. The multi-agent dynamic collaborative material editing system for standard declaration as described in claim 1, characterized in that, The standard compliance check in the intelligent review module adopts the BERT-base model, which is pre-trained on ≥8 million Chinese standard text corpora, and the fine-tuning data contains ≥5000 pairs of "material fragments-standard clauses"; the format check adopts a regular expression engine + PDFBox parsing; the information integrity check adopts template matching, that is, based on the application guide XML template. Logical consistency checks employ knowledge graph path reasoning, which maps entity relationships to a low-dimensional vector space based on the TransE model, optimizing the objective function d(h+r,t)→0.

8. A multi-agent dynamic collaborative material editing method for standard-oriented applications, implemented based on the multi-agent dynamic collaborative material editing system for standard-oriented applications as described in any one of claims 1-7, characterized in that... The method includes the following steps: (1) User demand initiation and parsing: Users input their demand through the user interaction module, and the demand analysis agent uses the BERT+CRF (bidirectional encoder representation from transformer + conditional random field) model to parse and generate a structured demand document; (2) Agent allocation and collaboration preparation: The Agent management module allocates multiple Agents and monitors their status to ensure that resources meet the requirements; (3) Content generation and preliminary layout: The content generation agent extracts standard terms and product parameters, and uses the fine-tuned GPT-4 (generative pre-trained converter 4) model to generate and layout the first draft; (4) Standard compliance check and dynamic adjustment: The standard compliance agent checks compliance through knowledge graph and BERT-base (basic version bidirectional encoder representation from converter) model, and makes dynamic adjustments based on user feedback; (5) Multi-department collaborative editing: The collaboration and coordination module realizes real-time collaboration among multiple users based on the RBAC (role-based access control) model and OT (operation conversion algorithm); (6) Final review and submission: After reviewing the Agent's multimodal check, the system automatically submits and records the status.

9. An electronic device, characterized in that, The electronic device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in claim 8.

Images