Document generation system and method based on large model

Abstract

The application discloses a large model-based official document generation system and method, belongs to the technical field of artificial intelligence, and aims to solve the technical problems of low accuracy of official document format and content, poor material invalidity, and poor content innovation in official document generation. The application comprises the following modules: a model training module, which is used for training an official document generation model based on official document samples in an official document sample library; a model fine-tuning module, which is used for fine-tuning the trained official document generation model in layers; a prompt word optimization module, which is used for matching prompt words based on an internal prompt word set; a summary generation module, which is used for generating an official document summary based on prompt word combination and user input, and calling a final official document generation model; an official document generation module, which is used for generating and outputting official document texts based on user input and the final official document summary, calling the final official document generation model, and calling a privacy data masking module to mask privacy data in the official document texts.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a big model-based official document generation system and method. BACKGROUND

[0002] At present, domestic official document generation technology mainly focuses on academic research based on deep learning methods. Various large models based on Transformer decoder pre-train models through massive unsupervised data, and the generated text outperforms other models in terms of context coherence and emotional expression, making it one of the main technologies for text generation.

[0003] However, the current generation system has problems such as low accuracy of official document format and content, poor timeliness of materials, and poor innovation of content, which is not conducive to official document creation and cannot meet the growing demand for official document writing.

[0004] The low accuracy of official document format and content, the poor timeliness of materials, and the poor innovation of content in official document generation are technical problems that need to be solved. SUMMARY

[0005] The technical task of the present application is to solve the technical problems of low accuracy of official document format and content, poor timeliness of materials, and poor innovation of content in the process of official document generation by providing a big model-based official document generation system and method.

[0006] In a first aspect, the present application provides a big model-based official document generation system, comprising:

[0007] A model training module is configured to periodically update an official document sample library, and train a document generation model based on official document samples in the official document sample library. The document generation model is a pre-trained large language model, which is used to generate an official document outline based on user input, and generate an official document text based on user input and the official document outline. The user input includes an official document title, an official document type, a document date, a document location, and keywords, and the keywords include a main point, a length, and a context.

[0008] A model fine-tuning module is configured to fine-tune the trained official document generation model in layers to obtain a final official document generation model.

[0009] A prompt word optimization module is configured to match user input with prompt words based on an internal prompt word set, and output a prompt word combination.

[0010] An outline generation module is configured to generate and output an official document outline based on the prompt word combination and user input, and call the final official document generation model.

[0011] a user interaction editing module configured to display a document outline through a user interaction interface and support user review and adjustment of the document outline to obtain a final document outline;

[0012] a document generation module configured to generate and output a document body based on user input and the final document outline by calling a final document generation model, and to call a privacy data masking module to mask privacy data in the document body;

[0013] a security detection module configured to supervise and conduct security audit on the model training module, the model fine-tuning module, the prompt word optimization module, the outline generation module, the user interaction editing module, the document generation module, and the privacy data masking module;

[0014] a rewriting expansion module configured to support user rewriting and expansion of the generated document body;

[0015] an evaluation optimization module configured to receive user feedback and optimize the document generation model and the system based on the user feedback.

[0016] Preferably, the model training module is configured to perform the following:

[0017] collecting document samples covering various types of documents;

[0018] performing data preprocessing on the collected document samples, including data cleaning, denoising, text format standardization, text segmentation and part-of-speech tagging, and text augmentation and enhancement, to obtain preprocessed document samples, and periodically updating the preprocessed document samples to the sample library;

[0019] training the document generation model based on the document samples in the sample library to obtain a trained document generation model.

[0020] Preferably, the model fine-tuning module is configured to perform the following hierarchical fine-tuning on the trained document generation model:

[0021] fine-tuning the structure and language style of the document generation model for document types;

[0022] fine-tuning the document model for document formats and application industry characteristics to determine the coherence and logic of the generated document;

[0023] fine-tuning the document generation model in a supervised manner for the coherence and logic requirements of long documents to generate long documents with clear structure and rigorous logic.

[0024] As preferred, for the prompt optimization module, the prompt set includes basic official document elements and covers industry data and professional vocabulary;

[0025] The prompt optimization module is used to adjust the weight combination of the matched prompts based on the context specified in the user input through the preconfigured context awareness strategy, and output the prompt combination that meets the specified context.

[0026] As preferred, the user interaction module is configured with a template library, editing tools and editing services, which support user editing optimization of the official document outline, and adjust the font and paragraph settings of the official document outline through editing optimization.

[0027] As preferred, the official document generation module is used to generate multi-modal content including recommended charts and cited literature according to the official document type and keywords specified in the user input.

[0028] As preferred, the security detection module is used to encrypt and access control the data transmission between the model training module, the model fine-tuning module, the prompt optimization module, the outline generation module, the user interaction editing module, the official document generation module and the privacy data masking module based on the preconfigured encryption method and access control strategy. The data involved in the model training module, the model fine-tuning module, the prompt optimization module, the outline generation module, the user interaction editing module, the official document generation module and the privacy data masking module are stored in a high-security server with firewall, intrusion detection service and data backup service.

[0029] As preferred, the privacy data masking module is used to identify privacy data in the official document text through entity recognition technology and replace the privacy data with preconfigured special symbols.

[0030] As preferred, the rewrite expansion module is configured with a rewrite expansion strategy to support user optimization and expansion of specific paragraphs or the entire text in the official document text, and to adjust the rewrite expansion strategy based on user feedback and user input.

[0031] In the second aspect, the present application is a kind of official document generation method based on large model, for generating official document text through a kind of official document generation system based on large model as any one of the first aspect, including the following steps:

[0032] Model training: update the official document sample library through the model training module, and train the official document generation model based on the official document samples in the official document sample library to obtain the trained official document generation model;

[0033] Model fine-tuning: the trained official document generation model is fine-tuned through the model fine-tuning module, and the final official document generation model is obtained;

[0034] Prompt word optimization: Based on user input, the prompt word optimization module is called to output prompt word combinations;

[0035] Outline generation: Taking the user input and prompt word combination as input, the final document generation model is called to generate and output the document outline;

[0036] Outline adjustment: Users review and adjust the document outline based on the user interactive editing module to obtain the final document outline;

[0037] Official document generation: Based on user input and the final official document outline, the official document generation module calls the final official document generation model to generate and output the official document body, and calls the privacy data masking module to mask the private data in the official document body;

[0038] Security Audit: Supervision and security audit of model training, model fine-tuning, prompt word optimization, outline generation, outline adjustment, official document generation, and privacy data shielding;

[0039] Rewrite and expand: users can rewrite and expand the generated official document text through the rewrite expansion module;

[0040] Evaluation and optimization: Receive user feedback through the evaluation and optimization module, and optimize the document generation model and document generation system based on user feedback.

[0041] The document generation system and method based on the large model of the present invention have the following advantages:

[0042] 1. Improve efficiency: significantly shorten the document writing cycle and reduce the burden on staff;

[0043] 2. Enhance accuracy: Ensure that official documents are standardized in format and accurate in content, reducing human errors;

[0044] 3. Enhanced innovation and personalization: Improve the innovation and personalized expression of official documents through specific optimization and user interaction;

[0045] 4. Real-time and flexibility: Quickly respond to the latest policy changes and information updates to ensure the timeliness of official document content. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] The present invention will be further described below with reference to the accompanying drawings.

[0048] Figure 1 This is a flowchart of a large-model-based document generation system for generating documents in Example 1. DETAILED DESCRIPTION

[0049] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments given are not intended to limit the present invention. Unless there is a conflict, the embodiments of the present invention and the technical features in the embodiments may be combined with each other.

[0050] The embodiments of the present invention provide a document generation system and method based on a large model, which are used to solve the technical problems of low accuracy of document format and content, poor material expiration, and poor content innovation in document generation.

[0051] Example 1:

[0052] The present invention provides a document generation system based on a large model, comprising a model training module, a model fine-tuning module, a prompt word optimization module, an outline generation module, a user interaction editing module, an official document generation module, a privacy data shielding module, a security detection module, a rewriting expansion module, and an evaluation optimization module.

[0053] The model training module is used to regularly update the official document sample library and perform model training on the official document generation model based on the official document samples in the official document sample library. The official document generation model is a pre-trained large language model, which is used to generate an official document outline based on user input, and to generate the official document body based on user input and the official document outline. The user input includes the official document title, official document type, date of issuance, location of issuance and keywords, and the keywords include subject, length and context.

[0054] In this embodiment, the model training module is used to perform the following:

[0055] (1) collecting official document samples, wherein the official document samples cover various types of official documents;

[0056] (2) Preprocessing the collected official document samples, performing data cleaning, denoising, text format standardization, text segmentation and part-of-speech tagging, and text expansion and enhancement on the official document samples to obtain preprocessed official document samples, and regularly updating the preprocessed official document samples to the sample library;

[0057] (3) The official document generation model is trained based on the official document samples in the sample library to obtain the trained official document generation model.

[0058] In this embodiment, the work of the model training module includes three stages: data collection, data preprocessing and basic model training.

[0059] During the data collection phase, a large number of different types of official document samples need to be collected and organized, covering 15 types of official documents such as reports, notifications, requests, and replies, to ensure diversity and representativeness of the samples to cover various industries and fields. In order to improve the model's ability to generate official documents for specific industries, special sample collection is needed for specific industries. For example, when collecting official documents of government departments, special attention should be paid to policy documents and regulatory documents; while collecting enterprise official documents, attention should be paid to contracts and reports. Collect representative and timely official documents. Regularly update the sample library to include the latest policy documents, industry reports, etc. to ensure that the model can keep pace with the times and generate timely official documents.

[0060] During the data preprocessing phase, the collected samples need to be cleaned to remove irrelevant information and noise data, standardize the text format, and ensure the standardization and consistency of the data; the text is processed for word segmentation, part-of-speech tagging, etc. to better extract key information from the text. In addition, in order to ensure the generalization ability of the model, the text is expanded and enhanced through synonym replacement, sentence reorganization, etc. to increase the diversity of the text. Prevent the influence of data problems on the training effect of the model.

[0061] During the basic model training phase, a pre-trained large language model such as BERT or GPT is used as the basic model, and the model is trained through a large number of official document samples to make it better adapt to the language style and structure specific to official documents. This process not only improves the accuracy of the model, but also enhances its flexibility and adaptability in practical applications.

[0062] The model fine-tuning module is used to fine-tune the trained official document generation model in layers to obtain the final official document generation model.

[0063] As a specific implementation of the model fine-tuning module, the module is used to fine-tune the trained official document generation model in layers as follows:

[0064] (1) For official document types, fine-tune the structure and language style of the official document generation model;

[0065] (2) For official document formats and application industry characteristics, fine-tune the official document model to determine the coherence and logic of the generated official document;

[0066] (3) For the coherence and logic requirements of long official documents, fine-tune the official document generation model in a supervised manner to generate long official documents with clear structure and rigorous logic.

[0067] The model fine-tuning module in this embodiment first fine-tunes the macrostructure and language style of official documents, then makes micro-detailed adjustments based on specific document formats and industry characteristics, gradually refining the model's performance in specific scenarios to ensure the coherence and logic of generated documents. To address the coherence and logic requirements of long-form documents, a supervised approach is used to carefully fine-tune the model, ensuring it can generate long, coherent, and logically rigorous articles. This improves the overall quality of official documents and ensures the fluency and readability of the document content.

[0068] The prompt word optimization module is used to match prompt words input by the user based on the built-in prompt word set and output prompt word combinations.

[0069] The prompt word optimization module includes basic document elements within the prompt word set, along with industry data and specialized vocabulary. Based on the context specified in the user input, the prompt word optimization module uses preconfigured context-aware strategies to weight matching prompt words and output a prompt word combination that matches the specified context. When matching prompt words within the prompt word set, matching query methods disclosed in the prior art can be used; this embodiment does not limit the specific method.

[0070] In order to improve the pertinence and professionalism of official documents, the prompt word optimization module of this embodiment has designed a richer and more detailed prompt word set for 15 types of national standard official documents (such as minutes, notices, announcements, etc.). These prompt word sets not only include basic official document elements, but also cover industry terms, professional vocabulary, etc., to ensure that the generated official documents are both in line with standards and professional. The prompt word optimization module also provides a context-aware prompt strategy, which can dynamically adjust the weight and combination of prompt words based on the official document background information entered by the user (such as industry, subject, urgency, etc.), so that the official documents generated by the model are more in line with specific situations. These prompt word sets are designed to optimize the starting point of the model when generating specific types of official documents, thereby enhancing the relevance and innovation of the content. In this way, it can be ensured that every official document can accurately meet the needs of the user.

[0071] The outline generation module is used to generate and output the official document outline based on the prompt word combination and user input, calling the final official document generation model.

[0072] The outline generation module of this embodiment automatically constructs a preliminary document outline framework based on the parameters input by the user (such as length requirements, title, keywords, time, location, etc.) through the final document production model. This framework provides structural guidance for subsequent content generation, ensuring that the document's organizational structure is reasonable and efficient.

[0073] The user interactive editing module displays the official document outline through the user interactive interface, and supports users to review and adjust the official document outline to obtain the final official document outline.

[0074] As a specific implementation of the user interaction module, the module is configured with a template library, editing tools and editing services. The template library, editing tools and editing services support users to edit and optimize the official document outline, and adjust the font and paragraph settings of the official document outline through editing optimization.

[0075] The user interactive editing module of this embodiment provides a user interface that allows users to review and adjust the automatically generated outline. In addition, it also provides a variety of editing tools and functions, such as font adjustment and paragraph setting, so that users can more conveniently edit and optimize official documents. This ensures that the content fully meets the user's specific needs and preferences. In this way, users can participate in every aspect of official document creation, ensuring that the final document is both professional and personalized. Template library technology is also introduced, providing users with a rich variety of official document templates for selection and use.

[0076] The official document generation module is used to generate and output the official document body based on user input and the final official document outline, call the final official document generation model, and call the privacy data masking module to mask the privacy data in the official document body.

[0077] As a specific implementation of the official document generation module, this module is used to generate multimodal content including recommended charts and references based on the official document type and keywords specified in the user input.

[0078] The document generation module of this embodiment uses a streaming output method to generate the content of the document based on the final confirmed outline. This method ensures that the generated document content is smooth and coherent, and can quickly respond to user input and editing operations. The generated document is polished and optimized through natural language generation technology to make it more consistent with the language style and expression habits of the document. In addition to the text content, the document generation module can also automatically generate multimodal content such as recommended charts and references based on the document type and user needs, enhancing the information richness and persuasiveness of the document.

[0079] The privacy data masking module of this embodiment is used to identify private data in the body of official documents through entity recognition technology and replace the private data with pre-configured special symbols. This private data includes sensitive information such as names, phone numbers, and email addresses. Once this private data is identified, it is replaced with a specific symbol (such as "xxx") to prevent the leakage of private data. In addition, the privacy data masking module provides a service for user-defined masking rules. Users can customize the masking rules and mask the identified private data based on the user-defined masking rules to meet their specific needs.

[0080] The security detection module is used to supervise and conduct security audits on the model training module, model fine-tuning module, prompt word optimization module, outline generation module, user interaction editing module, official document generation module, and privacy data masking module.

[0081] As a specific implementation of the security detection module, this module is used to encrypt and control access to data transmission between the model training module, the model fine-tuning module, the prompt word optimization module, the outline generation module, the user interaction editing module, the official document generation module and the privacy data shielding module based on pre-configured encryption methods and access control policies. The data involved in the model training module, the model fine-tuning module, the prompt word optimization module, the outline generation module, the user interaction editing module, the official document generation module and the privacy data shielding module are stored in a high-security server with a firewall, intrusion detection service and data backup service.

[0082] The security detection module of this embodiment adopts strict data protection measures and implements a full-link security audit mechanism from data collection to document output to ensure that data processing at each link meets the most stringent security standards and prevents potential data leakage risks. The security and privacy of user data are ensured by strictly supervising and auditing the system's data collection and processing processes. Encryption technology and access control policies are used to prevent unauthorized access and data leakage. Security audits and vulnerability scans are also performed on the system regularly to ensure the stability and security of the system. The document data is stored on a server with higher security, and it is ensured that the server has security measures such as firewalls, intrusion detection and data backup.

[0083] The rewrite extension module is used to support users to rewrite and expand the generated official document text.

[0084] As a specific implementation of the rewrite extension module, the module is configured with a rewrite extension strategy to support users in optimizing and expanding specific paragraphs or the entire text of the document, and to adjust the rewrite extension strategy based on user feedback and user input.

[0085] The rewriting and expansion module of this embodiment provides users with intelligent rewriting, expansion, and continuation tools. These tools can not only help users optimize and expand specific paragraphs or the entire text, but also automatically adjust the rewriting and expansion strategies and methods based on user input and feedback. In this way, it can be ensured that every official document can meet the highest standards.

[0086] The evaluation and optimization module is used to receive user feedback and optimize the document generation model and the system based on the user feedback.

[0087] The evaluation and optimization module of this embodiment regularly evaluates and optimizes the system. By collecting user feedback and evaluation indicators (such as the quality of generated documents, user satisfaction, system stability, etc.), it can promptly identify problems and make improvements. This continuous improvement approach enables the system to always maintain optimal performance and user experience. It uses machine learning technology to automatically analyze and optimize user behavior and system performance. For example, by analyzing user usage habits and preferences, it can optimize the system's recommendation algorithm and interface design to improve user experience and satisfaction.

[0088] After the system of this embodiment trains and fine-tunes the document generation model, Figure 1 The following execution is shown to generate an official document: based on user input, a prompt word combination is matched through a prompt word set, and based on the user input and the prompt word combination, an official document outline is output through an official document generation model. After the user adjusts the official document outline, the official document body is generated and output through the official document generation model based on the user input and the official document outline. The user adjusts the official document body output by the model to obtain the final official document body.

[0089] Example 2:

[0090] The present invention provides a method for generating official documents based on a large model, which generates the main body of the official document through the system disclosed in Example 1, and includes the following steps:

[0091] S100 model training: The model training module updates the document sample library and trains the document generation model based on the document samples in the library to obtain a trained document generation model.

[0092] S200 model fine-tuning: The model fine-tuning module performs layer-by-layer fine-tuning on the trained document generation model to obtain the final document generation model.

[0093] S300 prompt word optimization: Based on user input, the prompt word optimization module is called to output prompt word combinations;

[0094] S400 Outline Generation: takes user input and prompt word combination as input, calls the final document generation model to generate and output the document outline;

[0095] S500 outline adjustment: Users review and adjust the document outline based on the user interactive editing module to obtain the final document outline;

[0096] S600 Official Document Generation: Based on user input and the final official document outline, the official document generation module calls the final official document generation model to generate and output the official document body, and calls the privacy data masking module to mask the privacy data in the official document body;

[0097] S700 Security Audit: Monitors and conducts security audits on model training, model fine-tuning, prompt word optimization, outline generation, outline adjustment, document generation, and privacy data shielding.

[0098] S800 rewriting and expansion: users can rewrite and expand the generated official document text through the rewriting and expansion module;

[0099] S900 Evaluation and Optimization: Receive user feedback through the evaluation and optimization module, and optimize the document generation model and system based on user feedback.

[0100] The method of this embodiment can automatically generate the body of an official document, and can optimize and update the core sample library and the official document generation model, thereby improving accuracy and real-time performance.

[0101] The present invention has been shown and described in detail above through the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above multiple embodiments, those skilled in the art can know that the means in the above different embodiments can be combined to obtain more embodiments of the present invention, and these embodiments are also within the scope of protection of the present invention.

Claims

1. A document generation system based on a large model, characterized in that: include: A model training module, which is used to regularly update the document sample library and train the document generation model based on the document samples in the document sample library. The document generation model is a pre-trained large language model and is used to generate a document outline based on user input, and to generate the document body based on the user input and the document outline. The user input includes the document title, document type, document issuance date, document issuance location, and keywords. The keywords include subject, length, and context. A model fine-tuning module, which is used to perform layer-by-layer fine-tuning on the trained document generation model to obtain a final document generation model; A prompt word optimization module, which is used to match prompt words input by the user based on a built-in prompt word set and output a prompt word combination; An outline generation module, which is used to generate and output an official document outline based on a prompt word combination and user input and by calling a final official document generation model; A user interactive editing module, which displays the document outline through a user interactive interface and supports users to review and adjust the document outline to obtain the final document outline; An official document generation module, which is used to generate and output the official document body based on user input and the final official document outline, call the final official document generation model, and call the privacy data masking module to mask the privacy data in the official document body; A security monitoring module, which is used to monitor and conduct security audits on the model training module, model fine-tuning module, prompt word optimization module, outline generation module, user interaction editing module, official document generation module, and privacy data shielding module; A rewriting extension module, which is used to support users in rewriting and expanding the generated official document body; The evaluation and optimization module is used to receive user feedback and optimize the document generation model and the system based on the user feedback.

2. The document generation system based on a large model according to claim 1 is characterized in that: The model training module is used to perform the following: Collecting official document samples, which cover various types of official documents; Perform data preprocessing on the collected official document samples, including data cleaning, denoising, text format standardization, text segmentation and part-of-speech tagging, as well as text expansion and enhancement, to obtain preprocessed official document samples, which are then regularly updated to the sample library; The official document generation model is trained based on the official document samples in the sample library to obtain a trained official document generation model.

3. The document generation system based on a large model according to claim 1 is characterized in that: The model fine-tuning module is used to perform the following layered fine-tuning on the trained document generation model: Fine-tune the structure and language style of the document generation model based on the document type; Fine-tune the document model based on the document format and application industry characteristics to ensure the coherence and logic of the generated document; In response to the coherence and logic requirements of long official documents, the official document generation model is fine-tuned in a supervised manner to ensure that long official documents with clear organization and rigorous logic are generated.

4. The document generation system based on a large model according to claim 1 is characterized in that: For the prompt word optimization module, the prompt word set includes basic official document elements and covers data and professional vocabulary of various industries; The prompt word optimization module is used to adjust the weights of the matching prompt words based on the context specified in the user input through a pre-configured context-aware strategy, and output a prompt word combination that meets the specified context.

5. The document generation system based on a large model according to claim 1 is characterized in that: The user interactive editing module is configured with a template library, editing tools and editing services, and supports users to edit and optimize the official document outline through the template library, editing tools and editing services, and adjust the font and paragraph settings of the official document outline through editing optimization.

6. The document generation system based on a large model according to claim 1 is characterized in that: The document generation module is used to generate multimodal content including recommended charts and cited documents according to the document type and keywords specified in the user input.

7. The document generation system based on a large model according to claim 1 is characterized in that: The security detection module is used to encrypt and control access to data transmission between the model training module, the model fine-tuning module, the prompt word optimization module, the outline generation module, the user interaction editing module, the official document generation module and the privacy data shielding module based on a preconfigured encryption method and access control policy. The data involved in the model training module, the model fine-tuning module, the prompt word optimization module, the outline generation module, the user interaction editing module, the official document generation module and the privacy data shielding module are stored in a high-security server with a firewall, intrusion detection service and data backup service.

8. The document generation system based on a large model according to claim 1 is characterized in that: The privacy data masking module is used to identify the privacy data in the body of the document through entity recognition technology and replace the privacy data with pre-configured special symbols.

9. The document generation system based on a large model according to claim 1 is characterized in that: The rewriting extension module is configured with a rewriting extension strategy for supporting users to optimize and expand specific paragraphs or the entire text of the document, and for adjusting the rewriting extension strategy based on user feedback and user input.

10. A method for generating official documents based on a large model, characterized in that: The method for generating the body of an official document by using the official document generation system based on a large model as claimed in any one of claims 1 to 9 comprises the following steps: Model training: The document sample library is updated through the model training module, and the document generation model is trained based on the document samples in the document sample library to obtain the trained document generation model; Model fine-tuning: The model fine-tuning module is used to perform layer-by-layer fine-tuning on the trained document generation model to obtain the final document generation model. Prompt word optimization: Based on user input, the prompt word optimization module is called to output prompt word combinations; Outline generation: Taking the user input and prompt word combination as input, the final document generation model is called to generate and output the document outline; Outline adjustment: Users review and adjust the document outline based on the user interactive editing module to obtain the final document outline; Official document generation: Based on user input and the final official document outline, the official document generation module calls the final official document generation model to generate and output the official document body, and calls the privacy data masking module to mask the private data in the official document body; Security Audit: Supervision and security audit of model training, model fine-tuning, prompt word optimization, outline generation, outline adjustment, official document generation, and privacy data shielding; Rewrite and expand: users can rewrite and expand the generated official document text through the rewrite expansion module; Evaluation and optimization: Receive user feedback through the evaluation and optimization module, and optimize the document generation model and document generation system based on user feedback.

Images