Professional scheme generation method, system and equipment based on large model

Abstract

The invention relates to the technical field of natural language processing, and provides a professional scheme generation method, system and equipment based on a large model, and the method comprises the steps: obtaining a document type and project information, and calling a template corresponding to the document type; generating a directory structure through a large model after constructing a directory prompt based on the directory structure generation instruction and the project information; for each chapter in the directory structure, querying a corresponding generation rule in the template, retrieving related reference data as content elements in the internal knowledge base based on the types and retrieval requirements of the content elements in the generation rule, integrating the chapter prompt template, the content elements and project information into chapter prompts, and generating chapter contents through a large model; and splicing all the chapter contents according to the chapter sequence in the directory structure to obtain the technical scheme. And factual errors are greatly reduced.

Description

Technical Field

[0001] This invention belongs to the field of natural language processing technology, and in particular relates to a method, system and device for generating professional solutions based on a large model. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of document and solution generation, technological development shows an evolutionary trend from manual to automated, and from template-based to intelligent.

[0004] First, using traditional manual writing and template software is the most primitive method. Using office software, relying on manual brainstorming, organizing materials, and writing word by word, is flexible but extremely inefficient, and the quality is highly dependent on individual ability. Template software can provide a format, but it cannot intelligently generate content, making it difficult for non-core experts to quickly write high-quality proposals with rigorous logic and detailed content.

[0005] Secondly, the use of general artificial intelligence (AI) writing assistants has emerged. With the rise of large language models such as GPT-4 and Wenxin Yiyan, a large number of AI writing assistants have appeared. These tools can generate fluent text paragraphs based on simple prompts. However, their shortcomings are as follows: Lack of structured planning ability, namely the disconnect between macro-structure planning and micro-content generation: difficulty in handling long documents that require complex directory structures and logical coherence, such as tender documents; Domain knowledge limitations, namely the disconnect between general model capabilities and specific domain knowledge: knowledge comes from general corpora, which cannot cover internal terminology, norms, confidential data, etc. of specific industries or enterprises. The generated content often appears vague or contains factual errors (i.e., illusions) due to the lack of specific domain knowledge. The interaction mode is simplistic, namely the disconnect between AI batch generation and human fine-tuning: most interactions remain at the level of "question and answer" or "generating full text given a topic," lacking an effective mechanism for in-depth, iterative, and localized optimization of the generated content. For AI-generated content, users often need to perform a lot of secondary modifications and proofreading. The existing human-computer interaction mode is usually one-click generation followed by manual modification. Users need to copy the content into the editor and adjust it outside the AI ​​environment. This process interrupts the creative process, is inefficient, and cannot utilize AI for more refined iterative optimization. Summary of the Invention

[0006] To address the technical problems mentioned above, this invention provides a method, system, and device for generating professional solutions based on a large model. The template library ensures the structure and standardization of the solutions, and the link to an internal knowledge base guarantees the accuracy and professionalism of the content source, significantly reducing factual errors.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for generating specialized solutions based on a large model, comprising: Obtain document type and project information, and retrieve the template corresponding to the document type; the template contains directory structure generation instructions and generation rules for each chapter, and the generation rules include the type of content elements, chapter prompt templates, and search requirements; Based on the directory structure generation instructions and project information, after constructing the directory hints, the directory structure is generated through the large model; For each chapter in the directory structure, the corresponding generation rules are queried in the template. Based on the type of content elements and retrieval requirements in the generation rules, relevant reference materials are retrieved from the internal knowledge base as content elements. After integrating the chapter prompt template, content elements and project information into chapter prompts, the chapter content is generated through the large model. By piecing together the contents of all chapters according to the chapter order in the directory structure, a technical solution can be obtained.

[0008] Furthermore, it also includes: for a certain chapter, in response to the chapter debugging command, launching an interactive dialogue interface bound to the content of that chapter, and supporting the modification and regeneration of the content of that chapter through multiple rounds of dialogue within the interface.

[0009] Furthermore, the initial context of the multi-turn dialogue includes: a unique session identifier, a chapter ID, all generated text of the chapter, content elements, and dialogue history.

[0010] Furthermore, the types of content elements include text, summaries, images, and flowcharts.

[0011] Furthermore, the generation rules also include reference content; based on the type of content element in the generation rules and the reference content, text or images are generated as content elements.

[0012] Furthermore, the step of retrieving relevant reference materials from the internal knowledge base includes: Extract several keywords from document types, project information, and generation rules; For each keyword, calculate the number of times it appears in each reference and the total number of times it appears in the entire internal knowledge base, and calculate the weight based on its extraction source; For each reference, calculate the cosine similarity with all keywords, and then perform a weighted sum based on the weights to obtain a comprehensive score; Select the reference materials with the highest overall scores as relevant reference materials.

[0013] Furthermore, the weights are: ; Among them, B(k) i ) represents the keyword k i The base weights are determined based on their extraction source; TF(c j ,k i ) represents the keyword k i References in section j j The number of times it appears in the document; Inverse Document Frequency (IDF(k)) i )=log(N / (DF(k i )+1)),DF(k i This indicates that the internal knowledge base contains the keyword k. i The total number of references; N represents the number of keywords; K represents the keyword set.

[0014] Furthermore, the overall score is: SC(c j )=∑(w(k i )×sim(c j ,k i )); where w(k i ) represents the keyword k i The weights, sim(c j ,k i ) indicates document c j With keyword k i The cosine similarity.

[0015] A second aspect of the present invention provides a professional solution generation system based on a large model, comprising: The data acquisition module is configured to: acquire document type and project information, and retrieve the template corresponding to the document type; the template contains directory structure generation instructions and generation rules for each chapter, the generation rules including the type of content elements, chapter prompt templates and retrieval requirements; The directory generation module is configured to: generate directory hints based on directory structure generation instructions and project information, and then generate the directory structure through a large model; The chapter generation module is configured to: for each chapter in the directory structure, query the corresponding generation rules in the template, and based on the type of content elements and retrieval requirements in the generation rules, retrieve relevant reference materials from the internal knowledge base as content elements, integrate the chapter prompt template, content elements and project information into chapter prompts, and then generate chapter content through the large model; The solution generation module is configured to concatenate all chapter contents according to the chapter order in the directory structure to obtain the technical solution.

[0016] A third aspect of the present invention provides a computer device including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method for generating specialized solutions based on a large model.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention proposes a professional solution generation method based on a large model. The template library ensures the structure and standardization of the solution, and the link with the internal knowledge base guarantees the accuracy and professionalism of the content source, significantly reducing factual errors.

[0018] This invention proposes a professional solution generation method based on a large model. Through chapter-level dialog-based debugging, it achieves WYSIWYG fine-grained modification, allowing users to deeply refine the content without losing context. The modification efficiency and accuracy far exceed traditional editing methods.

[0019] This invention proposes a professional solution generation method based on a large model. The comprehensive score is obtained by accumulating the weighted similarity between the document and each keyword, which comprehensively considers the association between the document and all search keywords. It can accurately select documents that meet the association requirements of multiple keywords at the same time, greatly reducing the possibility of irrelevant documents being mistakenly selected, and making the search results more in line with expectations. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a flowchart of a professional solution generation method based on a large model according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the interactive dialogue interface of Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the structure of a computer device according to Embodiment 3 of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0023] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] Example 1 This embodiment provides a method for generating professional solutions based on a large model.

[0025] To address the knowledge limitations of large models, RAG (Retrieval Enhanced Generation) technology was explored. RAG is often used as a general knowledge question-and-answer or information retrieval portal. However, it has not yet been deeply integrated with a complete and structured document generation-assembly-debugging process. In particular, there are still gaps in how to accurately inject the retrieved knowledge into the generation process of specific chapters of the document, and how to interactively modify the knowledge around it.

[0026] This embodiment provides a professional solution generation method based on a large model, which realizes a closed-loop process from solution framework construction and intelligent content filling to interactive and precise optimization. It empowers users to quickly, efficiently, and with low barriers to generate high-quality solution documents that are structured, professional, and conform to the enterprise's internal knowledge standards, by only providing core project information.

[0027] This embodiment provides a method for generating professional solutions based on a large model, such as... Figure 1 As shown, it includes the following steps: Step 1: Receive the document type and project information input by the user.

[0028] After logging in, users can select a document type (doc_type) on the project creation interface, for example, "XX Project Technical Proposal".

[0029] Users input the core information of the project, project_info, such as: project name is Smart Park Brain Construction Project, client name is Future City Group, project type is park, construction type is box culvert, construction technology is box culvert, etc.

[0030] The user interaction layer sends document type and project information to the task orchestration controller via API (Application Programming Interface) requests.

[0031] Step 2: Based on the input information (document type and project information) and combined with preset rules, call the large model to generate the document's directory structure.

[0032] After receiving a request, the task orchestration controller first loads the template corresponding to the document type (e.g., XX Project Technical Proposal) and obtains the directory structure generation instructions from the template.

[0033] Based on the directory structure generation instructions and combined with project information, the controller constructs a directory Prompt, as follows: You are a top-level solutions architect. Please design a professional and complete directory structure for a technical proposal named '{{Smart Park Brain Construction Project}}'. The following is a reference template for construction project documents: {{Template Information Queryed}}.

[0034] The directory Prompt is sent to the large model for inference services. The large model generates a JSON-formatted directory structure and returns it to the controller.

[0035] After parsing the JSON-formatted directory structure, the controller sends the directory structure data to the user interaction layer, rendering a clickable directory tree on the interface.

[0036] Step 3: Based on the directory structure and predefined templates, determine the content elements (text, summary, images, etc.) of each chapter; for each chapter, perform a document search to retrieve relevant references; combine the references, project information, and templates, and call the large model to generate the specific content of each chapter; concatenate all the chapter content according to the chapter order in the directory structure to obtain the technical solution.

[0037] The task orchestration controller processes each chapter in the order of the directory tree. The following example uses the chapter "3.1 Overall Technical Architecture".

[0038] The controller retrieves the generation rules for each chapter from the template, for example: {"type":"text","prompt_template":"Please elaborate on the overall technical architecture of '{{project name}}', including..., please refer to the following resources: {{retrieved_docs}}","use_rag":true}. Here, 'type' indicates the type of content element, 'text' represents text, 'prompt_template' represents the chapter prompt template, 'retrieved_docs' represents the list of relevant documents retrieved from the database by the search engine, and 'use_rag":true' indicates the use of RAG technology, signifying the search requirement. The template contains generation rules for several chapters (e.g., rules for generating abstracts, overall technical architecture, summaries, conclusions, etc.). Based on the generated directory structure, the controller matches the most similar chapter in the template and retrieves the generation rules for that most similar chapter.

[0039] Based on the type of content element in the generation rules and the retrieval requirements, relevant reference materials are retrieved from the internal knowledge base and used as content elements. Alternatively, based on the type of content element in the generation rules and the reference content, text (e.g., abstracts) or images (e.g., flowcharts) are generated as content elements. For example: Triggering document retrieval: Because "type":"text" and "use_rag":true, the controller will extract keywords (e.g., technical architecture, smart park, etc.) from the document type, project information, and chapter generation rules, and call the API of the document retrieval subsystem; The document retrieval subsystem performs a similarity search in its vector database based on several keywords, finds the 3-5 most relevant historical solution fragments or technical documents, and returns these text contents to the controller.

[0040] Constructing the final chapter Prompt: The controller integrates project information, the chapter Prompt template (prompt_template) obtained from the template, and relevant reference materials (such as the content of retrieved documents) into a rich and context-complete chapter Prompt. For example, the variable {{project name}} is replaced with "Smart Park Brain Construction Project", and {{retrieved_docs}} is replaced with the actual retrieved knowledge text.

[0041] Content generation from the large model: The final chapter's Prompt is sent to the large model for inference services. Based on this rich input, the large model generates the content for each chapter, such as a detailed description of the overall technical architecture of the "Smart Park Brain Construction Project".

[0042] Once the generated content is returned, it is sent by the controller to the user interaction layer, where the corresponding progress is displayed in the appropriate location on the interface.

[0043] In this embodiment, the content element types can also be summary, image, and process. This part supports custom expansion, which can form rich content unit generation logic. If {"type":"summary"} is encountered, the previously generated content (reference content) will be used as input, and the large model will be called to summarize. If {"type":"image"} is encountered, a placeholder with a title will be inserted into the document, and a suitable image will be found and inserted into the placeholder based on the information required for the image. If {"type":"process"} is encountered, a placeholder with a process title will be inserted into the document, the process will be generated according to the flowchart format, and updated into the document as an image.

[0044] In this embodiment, based on several keywords, a similarity search is performed in the internal knowledge base to find the 3-5 most relevant historical solution fragments or technical documents. The specific steps are as follows: (1) Suppose there are M keywords K={k1,k2,...,k M The internal knowledge base contains N references C={c1,c2,...,c N}; (2) For each keyword, calculate the number of times it appears in each reference (document) and the total number of times it appears in the entire internal knowledge base, and calculate the weight based on its extraction source: ; Among them, B(k) i ) represents the keyword k i The base weight is determined based on its source: if it originates from the chapter generation rules, it takes the value 'a'; if it originates from the document type, it takes the value 'b'; and if it originates from project information, it takes the value 'c'. TF(c j ,k i ) represents the keyword k i In document c j (i.e., the number of times it appears in the j-th reference), TF(c j ,k i The frequency of a word increases linearly. When the word frequency is too high (e.g., a keyword appears dozens of times in the references), the weight will expand rapidly, causing the results to be dominated by a few high-frequency words. This embodiment introduces a logarithmic function to slow down the growth of word frequency, avoiding excessively abrupt weighting of high-frequency words and making the influence of word frequency on weighting smoother. Inverse Document Frequency (IDF) i )=log(N / (DF(k i )+1)),DF(k i This indicates that the entire internal knowledge base contains the keyword k. i The total number of documents; (3) Calculate document c j With keyword k i The cosine similarity, combined with weights, is used to calculate document c. j With keyword k i Similarity: S(c j ,k i )=w(k i )×sim(c j ,k i ), sim(c j ,k i ) indicates document c j With keyword k i Cosine similarity; (4) For each document, calculate the overall score SC(cj )=∑S(c j ,k i Select the 3-5 documents with the highest overall scores.

[0045] This embodiment allows keyword weights to reflect both their density within a single document and their rarity within the entire knowledge base. By combining their source with multi-dimensional comprehensive calculation of keyword weights, each keyword's weight accurately reflects its importance to the retrieval process. Furthermore, keywords that appear frequently in a document (such as "smart park" appearing repeatedly in a document) improve the document's relevance, ensuring that core content is prioritized for retrieval. Rare keywords (such as the technical term "overall technical architecture") have higher IDF values, effectively filtering out generalized information (avoiding dilution of results by generic terms like "smart park"), allowing the retrieval to focus on more targeted documents.

[0046] The overall score comprehensively considers the relevance of a document to all search keywords by accumulating the weighted similarity between the document and each keyword. This can accurately filter out documents that meet the relevance requirements of multiple keywords, greatly reducing the possibility of irrelevant documents being mistakenly selected, and making the search results more in line with expectations.

[0047] Step 4: Respond to the user's chapter debugging command and launch an interactive dialog interface bound to the content of that chapter, such as... Figure 2 As shown, this interface allows users to modify and regenerate content in real time through multiple rounds of dialogue.

[0048] When users preview the generated solution, they may find that the description of a certain section (e.g., 3.1 Overall Technical Architecture) is correct but does not highlight the company's technical advantages. This can be done by changing the model or through guided dialogue.

[0049] When a user clicks the "Debug" button next to the chapter, the user interaction layer responds to the click event, pops up a chapter debugging dialog box, and sends a request to the backend to start a debugging session, which includes the chapter ID.

[0050] The task orchestration controller calls the session management module to create a new debug session, whose initial context includes: session_id: A unique session identifier; chapter_id: Chapter ID; current_content: All text generated for this chapter; retrieved_docs_ids: A list of reference document IDs used when this section was generated; chat_history: Conversation history, initially empty.

[0051] Dialogic editing: The user entered the following in the dialog box: This description is good, but please help me emphasize the advantages of our self-developed "spatiotemporal data fusion engine" and compare it with industry-standard solutions. The text is sent to the backend via API. After receiving it, the task orchestration controller constructs a new Prompt for modification along with the session context (especially current_content and retrieved_docs) and sends it to the large model inference service. After the large model understands the instructions, it regenerates the content of this chapter. The new content will highlight the "spatiotemporal data fusion engine" and include comparative analysis. Real-time rendering: New content is returned to the front-end debug dialog box in real time via streaming API, and users can immediately see the effect of the changes.

[0052] Multiple iterations: Users can continue to issue instructions, such as making the comparison in tabular form clearer or refining the language, to conduct multiple rounds of optimization until they are satisfied; Confirm changes: The user clicks the "Confirm and Replace" button in the dialog box; the front end sends the final version of the content to the back end, the back end updates the content of that chapter in the database, and refreshes the display of the main editing area; the debugging session ends.

[0053] Step 5: Finalize and export the plan.

[0054] Once users are satisfied with all chapters, they can click the "Download Document" button to format all content according to the preset template and generate a complete, deliverable professional solution document.

[0055] The professional solutions, through customization, have already been tested online in the field of construction solutions, reducing the preparation time of professional solutions by 70%.

[0056] This embodiment provides a professional solution generation method based on a large model. It defines the skeleton and content paradigm of different types of solutions through a template library. The template is not only a static format, but also an "intelligent blueprint" that includes content types (text, summary, image, flowchart), logical relationships and variable placeholders. At runtime, it can automatically perform global replacement of key information such as project name, and call different functional modules (such as text generation, summary generation, and chart placeholders) according to template instructions, realizing the automated and standardized assembly of content. This method is more efficient and can better ensure consistency than the traditional manual modification of templates.

[0057] This embodiment provides a professional solution generation method based on a large model. By using a built-in internal knowledge base, when generating specific chapter content, it can retrieve relevant information (such as historical project data, technical specifications, success cases, etc.) from the enterprise's internal knowledge base or a designated document library in real time and accurately. This solves the problems of the "illusion" of large model content and lack of domain depth, so that the generated solution content has both the fluency of the model and the accuracy and professionalism of enterprise knowledge.

[0058] This embodiment provides a professional solution generation method based on a large model. The template library ensures the structure and standardization of the solution, and the link with the internal knowledge base guarantees the accuracy and professionalism of the content source, greatly reducing factual errors.

[0059] This embodiment provides a professional solution generation method based on a large model, which reduces the traditional solution writing time, which takes several days or even weeks, to the hour level. It frees experts from repetitive writing work, allowing them to focus on higher-value creative thinking. It is estimated that it can save more than 80% of manual workload. Even non-senior personnel can quickly generate professional-level solution drafts based on this embodiment, realizing the popularization of expert knowledge.

[0060] This embodiment provides a professional solution generation method based on a large model. Through chapter-level dialogic debugging, it achieves WYSIWYG fine-grained modification. Users can deeply refine the content through natural language and AI collaboration without leaving the context. The modification efficiency and accuracy far exceed traditional editing methods.

[0061] This embodiment provides a professional solution generation method based on a large model. When the debugging dialog box is launched, it automatically loads the context of the current chapter, including the generated content, the documents referenced during generation, and the basic information of the entire solution. This makes the communication between the user and the large model extremely focused and efficient.

[0062] This embodiment provides a professional solution generation method based on a large model. Users can engage in multi-round dialogues with the large model through natural language commands, preview the modification effects in real time, and continue until they are satisfied.

[0063] This embodiment provides a professional solution generation method based on a large model, which forms a rapid closed loop within a single interface and a specific chapter by integrating the three stages of generation, evaluation, and modification, greatly improving the efficiency and accuracy of content optimization.

[0064] Example 2 This embodiment provides a professional solution generation system based on a large model, including: The data acquisition module is configured to: acquire document type and project information, and retrieve the template corresponding to the document type; the template contains directory structure generation instructions and generation rules for each chapter, the generation rules including the type of content elements, chapter prompt templates and retrieval requirements; The directory generation module is configured to: generate directory hints based on directory structure generation instructions and project information, and then generate the directory structure through a large model; The chapter generation module is configured to: for each chapter in the directory structure, query the corresponding generation rules in the template, and based on the type of content elements and retrieval requirements in the generation rules, retrieve relevant reference materials from the internal knowledge base as content elements, integrate the chapter prompt template, content elements and project information into chapter prompts, and then generate chapter content through the large model; The solution generation module is configured to: concatenate all chapter contents according to the chapter order in the directory structure to obtain the technical solution; The interaction module is configured to, in response to a chapter debugging command, launch an interactive dialogue interface bound to the content of that chapter, and support modification and regeneration of the chapter content through multiple rounds of dialogue within the interface.

[0065] Furthermore, the initial context of a multi-turn dialogue includes: a unique session identifier, a chapter ID, all generated text of the chapter, content elements, and dialogue history.

[0066] Furthermore, the types of content elements include text, summaries, images, and flowcharts.

[0067] Furthermore, the generation rules also include reference content; based on the type of content element in the generation rules and the reference content, text or images are generated as content elements.

[0068] Furthermore, the steps for retrieving relevant reference materials from the internal knowledge base include: Extract several keywords from document types, project information, and generation rules; For each keyword, calculate the number of times it appears in each reference and the total number of times it appears in the entire internal knowledge base, and calculate the weight based on its extraction source; For each reference, calculate the cosine similarity with all keywords, and then perform a weighted sum based on the weights to obtain a comprehensive score; Select the reference materials with the highest overall scores as relevant reference materials.

[0069] Furthermore, the weights are: ; Among them, B(k) i ) represents the keyword ki The base weight is determined based on its source: if it originates from the chapter generation rules, it takes the value 'a'; if it originates from the document type, it takes the value 'b'; and if it originates from project information, it takes the value 'c'. TF(c j ,k i ) represents the keyword k i In document c j (i.e., the number of times it appears in the j-th reference), TF(c j ,k i The frequency of a word increases linearly. When the word frequency is too high (e.g., a keyword appears dozens of times in the references), the weight will expand rapidly, causing the results to be dominated by a few high-frequency words. This embodiment introduces a logarithmic function to slow down the growth of word frequency, avoiding excessively abrupt weighting of high-frequency words and making the influence of word frequency on weighting smoother. Inverse Document Frequency (IDF) i )=log(N / (DF(k i )+1)),DF(k i This indicates that the entire internal knowledge base contains the keyword k. i The total number of documents.

[0070] Furthermore, the overall score is: SC(c j )=∑(w(k i )×sim(c j ,k i )); where w(k i ) represents the keyword k i The weights, sim(c j ,k i ) indicates document c j With keyword k i The cosine similarity.

[0071] It should be noted that each module in this embodiment corresponds one-to-one with each step in Embodiment 1, and their specific implementation processes are the same, so they will not be repeated here.

[0072] Example 3 This embodiment provides a computer device, such as... Figure 3 As shown, the system includes a computer-readable storage medium 1003, a processor 1001, a communication interface 1002, and a computer program stored on the computer-readable storage medium 1003 and executable on the processor 1001. The processor 1001, communication interface 1002, and computer-readable storage medium 1003 can be connected via a bus or other means. The communication interface 1002 is used to receive and send data. When the processor 1001 executes the program, it implements the steps in the professional solution generation method based on a large model as described in Embodiment 1 above.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating professional solutions based on a large model, characterized in that, include: Retrieve document type and project information, and retrieve the template corresponding to the document type; The template contains directory structure generation instructions and generation rules for each chapter. The generation rules include the type of content elements, chapter prompt templates, and search requirements. Based on the directory structure generation instructions and project information, after constructing the directory hints, the directory structure is generated through the large model; For each chapter in the directory structure, the corresponding generation rules are queried in the template. Based on the type of content elements and retrieval requirements in the generation rules, relevant reference materials are retrieved from the internal knowledge base as content elements. After integrating the chapter prompt template, content elements and project information into chapter prompts, the chapter content is generated through the large model. By piecing together the contents of all chapters according to the chapter order in the directory structure, a technical solution can be obtained.

2. The method for generating professional solutions based on a large model as described in claim 1, characterized in that, It also includes: for a certain chapter, in response to the chapter debugging command, launching an interactive dialogue interface bound to the content of that chapter, and supporting modification and regeneration of the content of that chapter through multiple rounds of dialogue in the interface.

3. The method for generating professional solutions based on a large model as described in claim 2, characterized in that, The initial context of the multi-turn dialogue includes: a unique session identifier, a chapter ID, all generated text of the chapter, content elements, and dialogue history.

4. The method for generating professional solutions based on a large model as described in claim 1, characterized in that, The types of content elements include text, summaries, images, and flowcharts.

5. The method for generating professional solutions based on a large model as described in claim 1, characterized in that, The generation rules also include reference content; based on the type of content element in the generation rules and the reference content, text or images are generated as content elements.

6. The method for generating professional solutions based on a large model as described in claim 1, characterized in that, The steps for retrieving relevant reference materials from the internal knowledge base include: Extract several keywords from document types, project information, and generation rules; For each keyword, calculate the number of times it appears in each reference and the total number of times it appears in the entire internal knowledge base, and calculate the weight based on its extraction source; For each reference, calculate the cosine similarity with all keywords, and then perform a weighted sum based on the weights to obtain a comprehensive score; Select the reference materials with the highest overall scores as relevant reference materials.

7. The method for generating professional solutions based on a large model as described in claim 6, characterized in that, The weights are: ； Among them, B(k) i ) represents the keyword k i The base weights are determined based on their extraction source; TF(c j ,k i ) represents the keyword k i References in document j j The number of times it appears in the document; Inverse Document Frequency (IDF(k)) i )=log(N / (DF(k i )+1)),DF(k i This indicates that the internal knowledge base contains the keyword k. i The total number of references; N represents the number of keywords; K represents the keyword set.

8. The method for generating professional solutions based on a large model as described in claim 6, characterized in that, The overall score is: SC(c) j )=∑(w(k i )×sim(c j ,k i )); where w(k i ) represents the keyword k i The weights, sim(c j ,k i ) indicates document c j With keyword k i The cosine similarity.

9. A professional solution generation system based on a large model, characterized in that, include: The data acquisition module is configured to: acquire document type and project information, and retrieve the template corresponding to the document type; The template contains directory structure generation instructions and generation rules for each chapter. The generation rules include the type of content elements, chapter prompt templates, and search requirements. The directory generation module is configured to: generate directory hints based on directory structure generation instructions and project information, and then generate the directory structure through a large model; The chapter generation module is configured to: for each chapter in the directory structure, query the corresponding generation rules in the template, and based on the type of content elements and retrieval requirements in the generation rules, retrieve relevant reference materials from the internal knowledge base as content elements, integrate the chapter prompt template, content elements and project information into chapter prompts, and then generate chapter content through the large model; The solution generation module is configured to concatenate all chapter contents according to the chapter order in the directory structure to obtain the technical solution.

10. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method for generating professional solutions based on a large model as described in any one of claims 1-8.

Images