Intelligent writing method based on multi-agent cooperation

Abstract

The invention discloses an intelligent writing method based on multi-agent cooperation, and the method comprises the following steps: receiving documents of various formats uploaded by a user, extracting text and image resources, and understanding an image through a multi-modal large language model to generate a description text; constructing a cue word template by using a large language model, processing three types of input files, extracting key information, and constructing a knowledge model; receiving user input construction enhanced cue words, and calling a large language model to generate an article outline; chapter contents are generated one by one according to a title segmentation outline, and a writing execution agent generates a text according to requirements and automatically illustrates; and the MarkDown format output is converted into Word with a basic format. The method has the beneficial effects that intelligent writing is completed through multi-step cooperation, information is extracted from multiple sources, a model is constructed, an outline and content are efficiently generated, a diagram is automatically illustrated, finally, the format is converted, and convenient, efficient and functional intelligent writing services are provided for users.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and natural language processing technology, and more particularly to an intelligent writing method based on multi-agent collaboration. Background Art

[0002] With the rapid development of artificial intelligence and natural language processing technologies, intelligent writing systems have been widely used in fields such as news reporting, academic research, and commercial copywriting. Current intelligent writing systems are primarily based on large language models (LLMs), capable of generating high-quality text based on user input prompts or instructions. As shown in patent application number CN118917294B, entitled "Intelligent Writing Method and System Based on Multi-Agent Agents," this system and method generates text naturally based on text content and writing parameters. However, these existing systems still have many limitations when processing complex documents: Limited file parsing capabilities: Existing systems typically only process plain text and lack support for documents containing multimodal information such as images and tables. While some tools can extract text from PDFs or Word documents, they lack effective mechanisms for understanding and describing embedded charts and graphs.

[0003] Insufficient image semantic understanding: Currently, most intelligent writing systems are unable to automatically analyze images embedded in documents and use their semantic information for subsequent writing tasks. For example, when writing scientific papers, if the visual information of charts and graphs can be combined for interpretation, it will significantly improve the professionalism and readability of the article.

[0004] Rigid control of writing style: Many systems offer only fixed writing style options (e.g., formal, informal, etc.), making it difficult to meet users' needs for personalized styles. Furthermore, existing methods often rely on predefined rules or templates and lack dynamic adaptation capabilities, resulting in unnatural generated text styles.

[0005] Efficient document utilization: In academic writing, users often need to reference a large amount of literature. However, existing systems lack efficient mechanisms for extracting and matching document content, making it difficult to accurately identify key passages and integrate them into newly generated articles, resulting in insufficient citations or excessive duplication. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide an intelligent writing system and method based on multi-agent collaboration, aiming to achieve full-process automated generation from original documents to structured Markdown and then to mixed text and image Word documents through technologies such as multimodal understanding, dynamic style adaptation and intelligent document matching.

[0007] To achieve the above-mentioned object, the present invention provides the following technical solution: an intelligent writing method based on multi-agent collaboration, characterized in that it includes the following steps: Step 1: Receive documents in various formats uploaded by users, use extraction tools to extract text content and image resources from the documents, then call a multimodal large language model to perform semantic understanding on the extracted image resources and generate image description text for image-text matching recommendations in subsequent chapter generation; Step 2: Using a large language model, we construct a prompt word template and uniformly process the three types of input files provided by users: style reference files, writing requirements files, and content reference files. We extract key information from each file and construct a knowledge model that can be used for subsequent writing control. Step 3: Accept the user's input of the title, writing requirements, style requirements, and literature abstract, construct enhanced prompt words, and call the large language model to generate the article outline; Step 4: Segment the outline generated in step 3 by title, generate chapter content one by one, and use the writing execution agent to generate text based on the literature content, style, and writing requirements required for the current chapter. During the generation process, relevant images are automatically inserted by analyzing the current writing content and image descriptions. Step 5: Convert the MarkDown model output from step 4 into Word with basic document format.

[0008] As a further improvement of the present invention, the specific steps of extracting text content and image resources in step 1 and then performing semantic understanding are as follows: Step 1: Use python-magic or filetype library to identify the file type; if it is a Word file, call python-docx to read it; if it is a PDF file, call MinerU or PyMuPDF to extract text and images; the reading results are stored in MarkDown format; In steps one and two, for images embedded in Word / PDF files, extract them and save them to a local fixed path. Then, a customized prompt word template is constructed. The large language model is used to perform semantic understanding of the image and generate image description text.

[0009] As a further improvement of the present invention, the specific steps of performing semantic understanding in steps one and two are as follows: Step 121: Encode the input image through VisionTransformer: in, is the input image, are height, width, and number of channels respectively; Step 1, 2, 2, then use the large model as a decoder to generate image description text: ; Steps 1, 2, and 3: Store the image description text to the original image location in the following format:<image_illustrate> Image Description< / image_illustrate><image_path> Image Path< / image_path> .

[0010] As a further improvement of the present invention, the prompt word model module constructed in step 2 performs the following specific steps for uniformly processing the three types of input files provided by the user: style reference file, writing requirement file, and content reference file: Step 21: Construct a customized prompt word template to extract writing style and guide the large language model to extract writing style features; Step 22: Construct a customized prompt word template to extract writing requirements and guide the large language model to identify key elements; In steps 2 and 3, a semantic boundary recognition algorithm is used to intelligently segment the reference document uploaded by the user, dividing the full text into several paragraph units with clear semantic themes while strictly preserving the paragraph order and structure of the original document.

[0011] As a further improvement of the present invention, the semantic boundary recognition algorithm is used in steps 2 and 3 to intelligently segment the reference document uploaded by the user, dividing the full text into several paragraph units with clear semantic themes while strictly preserving the paragraph order and structure of the original document. The specific steps are as follows: Step 231, for the current paragraph , calculate the average sentence vector: ; Step 232, determine the segmentation point: If the starting sentence of the next paragraph Vector The cosine similarity with the average sentence vector of the paragraph is lower than the set threshold , it is considered that the semantics have changed significantly and a new paragraph should be opened: Among them, θ is automatically adjusted according to the document type, and is set to 0.65 for scientific papers and 0.7 for policy documents; In step 2, 3, and 3, use the big model to summarize the segmented paragraph units and store them in structured JSON.

[0012] As a further improvement of the present invention, the specific steps of generating the article outline in step 3 are as follows: In step 31, the user enters the article title and additional writing requirements, and selects the required style reference file, writing requirements file, and reference file. All files are optional. Step 32: Integrate the user input content and the parsing results of the three types of files to build a large model prompt word; In step 33, the large model prompt word constructed in step 32 is submitted to the user for confirmation, and the user modifies or regenerates it.

[0013] As a further improvement of the present invention, the specific steps of dividing the outline generated in step 3 and generating chapter contents one by one in step 4 are as follows: Step 41: Use regular expressions or AST parsers to segment the outline into the smallest level of chapters, and convert the titles and their attributes into structured JSON data; Step 42: Based on the file name, obtain the text content of the corresponding paragraph unit of the file, integrate the information of chapter title, references, writing requirements and style, construct writing prompts, and write by chapter; Among them, due to the uniqueness of the reference, the original information of the reference is used as input here. At the same time, for the image information in the reference, the prompt words guide the large model to use a special format for image citation during the writing process.

[0014] As a further improvement of the present invention, the specific steps of constructing writing prompt words and writing chapters in step 42 are as follows: Step 421: Maintain a global variable or cache structure to store information about generated chapters. After each chapter is completed, save its title, text summary, and other information in a structured manner. Also, record the information about inserted images and the number of generated tables to ensure the rationality of the figure numbering. Step 422: When writing the next chapter, automatically integrate the summary and image information of the previous chapter into the prompt to guide the model to generate content with logical continuity; Step 423: Generate text according to the enhanced prompt. After the current chapter is generated, cache its summary information again and record the newly used pictures, which will be used as the new "previous chapter" context information for the next writing.

[0015] As a further improvement of the present invention, the specific steps of converting the model output in the MarkDown format of step 4 into Word with a basic document format in step 5 are as follows: Step 5.1: Convert the title hierarchy, text content, tables, and image paths in Markdown to Word document objects, preserving the original text and image layout structure. Step 52: Read the image path information in Markdown and insert the locally stored image resources into the corresponding position of the Word document to ensure the consistency of the image and text; Step 53: Save Word to a local path.

[0016] Beneficial effects of the present invention: Dynamic adaptation of writing style. The present invention introduces customized prompt word templates, which can extract detailed writing style features from the user's style reference file, including language style, tone tendency, sentence complexity, vocabulary density, etc., to achieve document imitation.

[0017] Intelligent document segmentation and summary generation: Using a semantic boundary detection algorithm and the Sentence-BERT model, the system can segment long documents into multiple paragraph units with clear themes and generate summaries for each unit. This process not only preserves the logical structure of the original document but also improves the utilization of the document content, helping to generate more accurate and rich article content.

[0018] Intelligent outline generation and chapter writing: The system integrates information such as titles, writing requirements, style guidelines, and literature abstracts to construct enhanced prompts, guiding the large language model to generate a clearly structured and logically coherent article outline. Based on this outline, the system then generates each chapter on a chapter-by-chapter basis, incorporating content summaries and image usage from generated chapters to ensure that each paragraph has a good contextual connection and avoid repetition or fragmentation.

[0019] During the content generation process, the system automatically analyzes the current writing content and image descriptions and inserts relevant images at appropriate locations, enhancing the article's visual quality. Finally, the Markdown output is converted into a Word document with basic document formatting, ensuring consistency between text and images and meeting the user's requirements for the final document format. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flowchart of the file preprocessing and multimodal data analysis of the present invention; Figure 2 The writing knowledge extraction process for the present invention; Figure 3 outline generation and chapter writing process for this invention; Figure 4 This is a flow chart of an intelligent writing system and method based on multi-agent collaboration of the present invention. DETAILED DESCRIPTION

[0021] The present invention will be further described below with reference to the embodiments shown in the accompanying drawings.

[0022] Reference Figures 1 to 4 As shown, an intelligent writing system and method based on multi-agent collaboration in this embodiment includes the following steps: S1. File preprocessing and multimodal data analysis S1.1. Use python-magic or the filetype library to identify the file type. If it is a Word file, call python-docx to read it. If it is a PDF file, call MinerU or PyMuPDF to extract text and images. The reading results are stored in MarkDown format.

[0023] S1.2. For images embedded in Word / PDF files, extract them and save them to a local fixed path. Construct a customized prompt word template, call the large language model to perform semantic understanding of the image, and generate image description text.

[0024] S1.2.1. Image encoding via VisionTransformer (ViT): in, is the input image, are height, width, and number of channels respectively.

[0025] S1.2.2. Then use the large model as a decoder to generate image description text: S1.3. Store the image description text to the original image location in the following format:<image_illustrate> Image Description< / image_illustrate><image_path> Image Path< / image_path> .

[0026] S2. Document Knowledge Extraction S2.1 Writing style extraction S2.1.1. Construct a customized prompt word template to extract writing style. An example prompt word is as follows: Please analyze the writing style of the following documents and return the results in JSON format: { "language_style":"Formal / Informal / Semi-formal", "tone":"objective / subjective / critical / suggestive", "sentence_complexity":"Simple / Medium / Complex", "vocabulary_density":"low density / medium density / high density", "typical_phrases":["common phrase 1","common phrase 2"], "paragraph_structure":"General-specific / parallel / progressive", "outline":"Article outline", } The document content is as follows: [Insert document text] S2.1.2 Guide the large language model to extract writing style features, including language style (formal / informal), sentence complexity, vocabulary density, common phrase list, paragraph structure pattern, article directory structure, etc., and store them in standard JSON format.

[0027] S2.2. Writing content requirements extraction S2.2.1. Construct a customized prompt word template to extract writing requirements. Taking the tender document as an example, an example prompt word is as follows: Please extract the core elements of the writing requirements from the following document and return them in JSON format: { "word_count":"Word limit (integer)", "format_requirements":["Format requirements list"], "focus_areas": ["Key coverage areas"], "prohibited_words":["prohibited words"], "evaluation_criteria":["Evaluation Criteria"] } The document content is as follows: [Insert document text] S2.2.2. Guide the large language model to identify key elements, including word count limits, format requirements, content focus, banned words, and review criteria, and store them in standard JSON format.

[0028] S2.3. Use the semantic boundary recognition algorithm to intelligently segment the reference files uploaded by users, dividing the full text into several paragraph units with clear semantic themes, while strictly preserving the paragraph order and structure of the original document.

[0029] S2.3.1. Segment the original text at the sentence level to obtain a sentence sequence: in Indicates the A sentence.

[0030] S2.3.2. Use the Sentence-BERT model to map each sentence into a vector representation: S2.3.3. Without disrupting the original order, the system traverses the paragraphs from front to back and determines whether a new semantic paragraph should be divided here.

[0031] S2.3.3.1. For the current paragraph , calculate the average sentence vector: S2.3.3.2, determine the segmentation point: if the starting sentence of the next paragraph Vector The cosine similarity with the average sentence vector of the paragraph is lower than the set threshold , it is considered that the semantics have changed significantly and a new paragraph should be opened: Among them, θ can be automatically adjusted according to the document type, for example, it is set to 0.65 for scientific papers and 0.7 for policy documents.

[0032] S2.3.3. Use the large model to summarize the segmented paragraph units and store them in structured JSON. An example storage structure is as follows: [ { "content":"#Introduction\nIn recent years, artificial intelligence technology has made significant progress.\nIn particular, the development of convolutional neural networks has driven innovation in the field of computer vision.", "abstract": "This paper introduces the current development of artificial intelligence and convolutional neural networks in computer vision." }, { However, the application of these technologies in medical diagnosis still faces many challenges, such as data privacy issues and insufficient model interpretability. "abstract":"This paper points out the main problems in the application of AI technology in the medical field." } ] S3. Generate an article writing outline S3.1. The user enters the article title and additional writing requirements, and selects the required style reference file, writing requirements file, and reference file. All files are optional.

[0033] S3.2. Integrate user input and the parsing results of the three types of files to build a large model prompt word. An example prompt word is as follows: Please generate an article outline that meets the writing standards based on the following information: Title: "{title}"

Writing requirements

[0034] The example format is as follows: #The example outline is as follows: ##1[Chapter Title] ###1.1[Subtitle] ####1.1.1[Subtitle] #####Chapter Overview: A more detailed chapter overview #####Reference documents: Article 1-1, Article 2-1 In the example, article 1-1 indicates the first paragraph unit of article 1.

[0035] S3.3. Submit it to the user for confirmation, and the user can modify or regenerate it.

[0036] S4: Based on the article outline generated by S3, execute the content writing agent to generate content and mix text and images.

[0037] S4.1. Use regular expressions or an AST parser to segment the outline into the smallest level of chapters, and convert the titles and their attributes (such as chapter overview, references, etc.) into structured JSON data. A simple example is as follows: [ { "section_heading":"1 Introduction\n1.1 Research Background", "content_summary":"Describes the current development status of AI technology.", "references":["Reference 1-1","Reference 2-1"] }, ] S4.2. Based on the file name, obtain the text content of the corresponding paragraph unit. Integrate information such as chapter titles, references, writing requirements, and style to construct writing prompts and conduct chapter-by-chapter writing. Due to the unique nature of references, the original text of the reference (in paragraph units) is used as input. Furthermore, for images in the references, prompts are included to guide the large model to use a special format for image citations during the writing process.

[0038] S4.2.1. Maintain a global variable or cache structure to store information about generated chapters. After each chapter is completed, save its title, summary, and other information in a structured manner. Also, record the information about inserted images and the number of generated tables to ensure the rationality of the figure numbering. A simple example structure is as follows: { "section_heading":"2.1 Current Status of Artificial Intelligence Development", "summary":"This section summarizes the development trends of artificial intelligence technology in recent years, focusing on breakthroughs in deep learning and natural language processing.", } S4.2.2. When writing the next chapter, automatically integrate the summary and image information of the previous chapter into the prompt to guide the model to generate content with logical continuity. A simple prompt word example is as follows: Please write your section "{current_section_heading}" in the following style: -Writing language style: {language_style} - Tone tendency: {tone} -Sentence complexity: {sentence_complexity} - Vocabulary density: {vocabulary_density} Writing requirements include: -Word limit: {word_count} words -Format requirements: {format_requirements} -Content focus: {focus_areas} [Note] You are currently writing Chapter {n}. Please refer to the following chapters to ensure logical coherence and avoid repetition: <<<Previous Chapter Summary>>> Section title: {last_section_heading} Summary: "{last_summary}" Images used: {images_used} Please write the main text based on the following references: {reference_content_1} {reference_content_2} Please insert the image reference in Markdown format at the appropriate location, such as: Figure 4: AI Development Trends (images / ai_trend.png) ... Main text begins: S4.2.3. After the text is generated according to the enhanced prompt, its summary and other information are cached again and the newly used pictures are recorded, which will be used as the new "previous chapter" context information when writing next time.

[0039] S5. Convert to formatted Word output.

[0040] S5.1. Convert the heading hierarchy, text content, tables, and image paths in Markdown to Word document objects, preserving the original text and image layout structure. S5.2. Read the image path information in Markdown and insert the locally stored image resources into the corresponding position of the Word document to ensure the consistency of the image and text.

[0041] S5.3. Save the Word file to a local path.

[0042] In summary, the intelligent writing system and method based on multi-agent collaboration of this embodiment has the following characteristics: Multi-agent collaboration: This invention utilizes a multi-agent collaborative architecture, where each agent is responsible for a specific task (e.g., document preprocessing, knowledge extraction, outline generation, content writing, etc.). Agents collaborate by sharing global state and cached information, thereby improving the efficiency and accuracy of the overall system.

[0043] Multimodal Understanding and Image-Text Matching: The system not only extracts the text content within documents but also semantically understands embedded images and generates corresponding descriptions. This image description information is then used to make image-text matching recommendations during the subsequent chapter generation process, making the generated articles more expressive and professional.

[0044] Dynamic style adaptation and personalized writing: By extracting detailed writing style features from the user's style reference file and dynamically adjusting them during the generation process, the system can generate text that meets the user's personalized needs. This flexible style control mechanism makes the system suitable for writing tasks in different scenarios, such as academic papers, business reports, and marketing copywriting.

[0045] Intelligent Document Processing: This paper introduces a semantic boundary detection algorithm and the Sentence-BERT model to intelligently segment long documents and generate summaries for each paragraph. This processing method not only improves the utilization of document content but also provides rich reference materials for subsequent content generation.

[0046] Enhanced prompt word construction and context-aware writing: The system constructs enhanced prompt words, integrating information such as the title, writing requirements, style characteristics, and literature abstracts to ensure that the generated article conforms to formatting standards and has depth of content. Furthermore, the system records the content summary and image usage of the generated chapter and automatically integrates this information when writing the next chapter to ensure logical coherence of the generated content.

[0047] Full-process automation and format conversion: From parsing the original document to generating the final Word document, the entire process is fully automated. The system can convert Markdown-formatted output into a Word document with basic document formatting, ensuring consistency between images and text, greatly simplifying the user's operation process.

[0048] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent writing method based on multi-agent collaboration, characterized by: The steps include: Step 1: Receive documents in various formats uploaded by users, use extraction tools to extract text content and image resources from the documents, then call a multimodal large language model to perform semantic understanding on the extracted image resources and generate image description text for image-text matching recommendations in subsequent chapter generation; Step 2: Using a large language model, we construct a prompt word template and uniformly process the three types of input files provided by users: style reference files, writing requirements files, and content reference files. We extract key information from each file and construct a knowledge model that can be used for subsequent writing control. Step 3: Accept the user's input of the title, writing requirements, style requirements, and literature abstract, construct enhanced prompt words, and call the large language model to generate the article outline; Step 4: Segment the outline generated in step 3 by title, generate chapter content one by one, and use the writing execution agent to generate text based on the literature content, style, and writing requirements required for the current chapter. During the generation process, relevant images are automatically inserted by analyzing the current writing content and image descriptions. Step 5: Convert the MarkDown model output from step 4 into Word with basic document format.

2. The intelligent writing method based on multi-agent collaboration according to claim 1, characterized in that: The specific steps for extracting text content and image resources in step 1 and then performing semantic understanding are as follows: Step 1: Use python-magic or filetype library to identify the file type; if it is a Word file, call python-docx to read it; if it is a PDF file, call MinerU or PyMuPDF to extract text and images; the reading results are stored in MarkDown format; In steps one and two, for images embedded in Word / PDF files, extract them and save them to a local fixed path. Then, a customized prompt word template is constructed. The large language model is used to perform semantic understanding of the image and generate image description text.

3. The intelligent writing method based on multi-agent collaboration according to claim 2, characterized in that: The specific steps for performing semantic understanding in steps 1 and 2 are as follows: Step 121: Encode the input image through VisionTransformer: in, is the input image, are height, width, and number of channels respectively; Step 1, 2, 2, then use the large model as a decoder to generate image description text: ； Steps 1, 2, and 3: Store the image description text to the original image location in the following format:<image_illustrate> Image Description< / image_illustrate><image_path> Image Path< / image_path> .

4. The intelligent writing method based on multi-agent collaboration according to any one of claims 1 to 3, characterized in that: The specific steps for the prompt word model module constructed in step 2 to uniformly process the three types of input files provided by the user, namely, the style reference file, the writing requirement file, and the content reference file, are as follows: Step 21: Construct a customized prompt word template to extract writing style and guide the large language model to extract writing style features; Step 22: Construct a customized prompt word template to extract writing requirements and guide the large language model to identify key elements; In steps 2 and 3, a semantic boundary recognition algorithm is used to intelligently segment the reference document uploaded by the user, dividing the full text into several paragraph units with clear semantic themes while strictly preserving the paragraph order and structure of the original document.

5. The intelligent writing method based on multi-agent collaboration according to claim 4 is characterized in that: In steps 2 and 3, a semantic boundary recognition algorithm is used to intelligently segment the reference document uploaded by the user, dividing the full text into several paragraph units with clear semantic themes while strictly preserving the paragraph order and structure of the original document. The specific steps are as follows: Step 231, for the current paragraph , calculate the average sentence vector: ； Step 232, determine the segmentation point: If the starting sentence of the next paragraph Vector The cosine similarity with the average sentence vector of the paragraph is lower than the set threshold , it is considered that the semantics have changed significantly and a new paragraph should be opened: Among them, θ is automatically adjusted according to the document type, and is set to 0.65 for scientific papers and 0.7 for policy documents; In step 2, 3, and 3, use the big model to summarize the segmented paragraph units and store them in structured JSON.

6. The intelligent writing method based on multi-agent collaboration according to any one of claims 1 to 3, characterized in that: The specific steps for generating the article outline in step 3 are as follows: In step 31, the user enters the article title and additional writing requirements, and selects the required style reference file, writing requirements file, and reference file. All files are optional. Step 32: Integrate the user input content and the parsing results of the three types of files to build a large model prompt word; In step 33, the large model prompt word constructed in step 32 is submitted to the user for confirmation, and the user modifies or regenerates it.

7. The intelligent writing method based on multi-agent collaboration according to any one of claims 1 to 3, characterized in that: The specific steps for dividing the outline generated in step 3 into sections and generating chapter contents one by one in step 4 are as follows: Step 41: Use regular expressions or AST parsers to segment the outline into the smallest level of chapters, and convert the titles and their attributes into structured JSON data; Step 42: Based on the file name, obtain the text content of the corresponding paragraph unit of the file, integrate the information of chapter title, references, writing requirements and style, construct writing prompts, and write by chapter; Among them, due to the uniqueness of the reference, the original text information of the reference is used as input here. At the same time, for the image information in the reference, the prompt words guide the large model to use a special format for image citation during the writing process.

8. The intelligent writing method based on multi-agent collaboration according to claim 7, characterized in that: The specific steps for constructing writing prompts in step 42 and writing chapters are as follows: Step 421: Maintain a global variable or cache structure to store information about generated chapters. After each chapter is completed, save its title, text summary, and other information in a structured manner. Also, record the information about inserted images and the number of generated tables to ensure the rationality of the figure numbering. Step 422: When writing the next chapter, automatically integrate the summary and image information of the previous chapter into the prompt to guide the model to generate content with logical continuity; Step 423: Generate text based on the enhanced prompt. After the current chapter is generated, cache its summary information again and record the newly used pictures, which will be used as the new "previous chapter" context information for the next writing.

9. The intelligent writing method based on multi-agent collaboration according to any one of claims 1 to 3, characterized in that: The specific steps for converting the MarkDown model output in step 4 into Word with basic document format in step 5 are as follows: Step 5.1: Convert the title hierarchy, text content, tables, and image paths in Markdown to Word document objects, preserving the original text and image layout structure. Step 52: Read the image path information in Markdown and insert the locally stored image resources into the corresponding position of the Word document to ensure the consistency of the image and text; Step 53: Save Word to a local path.

Images