Two-stage long text report generation method and system

Abstract

The invention discloses a two-stage long text report generation method and system, and relates to the technical field of artificial intelligence. The invention provides a two-stage long text report generation system. The two-stage long text report generation system comprises an article outline generation module, a report generation module and an abstract and summary report generation module, the long text report generation method based on the system comprises the following steps: according to a report generation demand, performing theme investigation and view discovery, simulating a domain expert dialogue through an intelligent agent to guide a report theme, establishing a rich knowledge base, and generating a detailed article outline based on the knowledge base; according to the word number and style requirements of the user, the knowledge base and the article outline are combined, long text writing is carried out, reference literature numbers are embedded, generated report content and related literatures are exported as PDF, and the user can store and share the report content and the related literatures conveniently. Therefore, by the adoption of the method, the outline with strict logic can be generated, diversified personalized requirements are met, and the method is suitable for report requirements in different fields.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a two-stage long text report generation method and system. Background Art

[0002] With the rapid development of the information society, the demand for high-quality text content in various fields is increasing day by day. From the business plans of enterprises to the academic reports of research institutions and then to special reports, long text content plays a crucial role in communication, decision-making, and knowledge transfer.

[0003] Currently, the existing technical solutions related to long text report generation mainly focus on template-based content generation, information retrieval and data integration tools, language models based on short text generation, etc. However, the existing technologies have the following disadvantages: (1) The content depth and basis are insufficient. Existing technologies usually have difficulty integrating external retrieval data and knowledge base content when generating reports, resulting in the generated content lacking factual basis or in-depth analysis; (2) Lack of diversification and customization support. Existing tools are usually highly fixed in terms of style and format, and it is difficult to meet the specific needs of users for report types (such as public opinion analysis, academic reports); (3) Lack of structuring and logic. Many existing solutions generate text mainly in short paragraphs, lacking overall structure planning and poor logical coherence.

[0004] Therefore, it is necessary to provide a two-stage long text report generation system that supports online retrieval and obtaining the materials and information required for the report from the knowledge base, automatically generates an article outline through an agent simulating multi-expert conversations; and, users can specify the style and format of the report according to their needs, while supporting the generation of reports with pictures and texts, and at the same time supporting the summary of large-scale information data and generating analysis reports according to fixed dimensions. Summary of the Invention

[0005] The purpose of the present invention is to provide a two-stage long text report generation method and system that can generate a logically rigorous outline, meet diverse personalized needs, and be applicable to report requirements in different fields.

[0006] To achieve the above object, the present invention provides a two-stage long text report generation method, including the following steps:

[0007] S1. According to the report generation requirements, conduct theme research and perspective discovery, guide in-depth exploration of the report theme through an agent simulating domain expert conversations, establish a rich knowledge base, thereby ensuring the quality of the article, and based on this knowledge base, generate a detailed article outline;

[0008] S2. According to the user's requirements for the number of words and style, combined with the knowledge base and the article outline, write long texts, embed reference numbers to enhance the credibility of the report, and export the generated report content and relevant literature as PDFs for easy storage and sharing by the user.

[0009] Preferably, the theme research and perspective discovery include retrieving and analyzing articles related to the topic, exploring the topic from different angles to ensure the depth and breadth of the content, discovering and generating diverse research perspectives, and screening out perspectives that meet the goals.

[0010] Preferably, step S1 includes:

[0011] S11. Generate report editors with different perspectives and experts based on web resources;

[0012] S12. The report editors and experts conduct simulated Q&A while building a knowledge base;

[0013] S13. Generate an article outline based on the simulated Q&A, the knowledge base, and the report generation requirements.

[0014] Preferably, step S2 includes, when the user needs a graphic report, generating a graphic-rich report as follows:

[0015] First, use the language model to summarize the paragraphs and generate illustration descriptions;

[0016] Secondly, according to the paragraph summaries, illustration descriptions, and report style, use the image generation model to generate corresponding pictures;

[0017] Then, extract the content paragraph by paragraph until the end of the report, and polish it according to the report style.

[0018] A two-stage long text report generation system includes:

[0019] An article outline generation module that generates a draft outline through a large model and simulates multiple rounds of conversations. According to the conversation records, themes, writing purposes, and style requirements, it further optimizes the draft outline to generate a refined article outline;

[0020] A report generation module that selects a predefined format according to the user's needs, or uploads a custom format description and sample file to generate a report that meets specific writing norms, forms traceable and displayable literature, and exports it in PDF format for storage and sharing;

[0021] A summary report generation module that preprocesses the information according to the large-scale information and analysis dimensions provided by the user to generate high-quality summaries and comprehensive reports as the knowledge base for discussion conversations.

[0022] Therefore, the present invention adopts the above-mentioned two-stage long text report generation method and system, and has the following technical effects:

[0023] (1) Through the retrieval-enhanced generation technology, the online retrieval information is combined with the content of the knowledge base to achieve the real-time and accuracy of the content, and it can dynamically obtain and integrate relevant information, ensuring that the report content is well-founded and has higher depth and authority. It not only reduces the workload of manual screening and integration, but also significantly improves the efficiency and quality of report generation.

[0024] (2) Through the agent technology to simulate the cooperation of multi-field experts, and at the same time ensure that the generated report has a clear hierarchical structure and strict logic, solving the situation of loose and incoherent output text structure in the prior art to meet the user's needs for high-quality long text.

[0025] (3) Support users to specify the style, format and application scenarios of the report (such as public opinion analysis, academic reports, etc.), realizing the report generation function of diversification and high flexibility, solving the problem that the prior art is mostly based on fixed templates and is difficult to adapt to complex and personalized needs, making it have stronger applicability and wide application value in practical applications.

[0026] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flowchart of a two-stage long text report generation method;

[0028] Figure 2 is a schematic diagram of the process of generating a graphic report in an embodiment of a two-stage long text report generation method and system;

[0029] Figure 3 is a schematic diagram of the process of generating a large-scale information summary report in an embodiment of a two-stage long text report generation method and system. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The present invention can be more specifically explained through the following embodiments. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following embodiments.

[0031] As Figure 1 shown, the present invention provides a two-stage long text report generation method, including a writing preparation stage and a writing stage. The specific steps are as follows:

[0032] S1. Since in traditional knowledge retrieval - enhanced writing, much information is difficult to obtain through simple topic retrieval. Therefore, in the writing preparation stage, by combining information retrieval and generative artificial intelligence models through retrieval - enhancement technology, before generating an answer, the model retrieves relevant information from an external knowledge base (such as a database, document collection, or the web), inputs this information as context into the generative model, realizes seamless integration of online - retrieved information and knowledge - base content, and makes the generated content more accurate, rich, and in line with actual needs. The specific steps are as follows:

[0033] First, according to the topic and writing purpose provided by the user, conduct topic research and perspective discovery, retrieve and analyze articles related to the topic, explore the topic from multiple perspectives through dialogue, ensure the depth and breadth of the content, discover diverse research perspectives, and finally generate multiple perspectives and screen out the most target - compliant perspective.

[0034] Then, based on retrieval - enhancement technology, through an agent, simulate report editors from different perspectives and experts based on web resources to conduct multiple rounds of simulated Q&A dialogues, generate a draft outline with timeliness and depth, and establish a knowledge base at the same time; according to the simulated Q&A, the knowledge base, and the report - generation requirements, further optimize the draft outline, and finally generate a refined and logically - rigorous article outline to ensure the integrity and clarity of the report structure.

[0035] S2. In the writing stage, the user can select a predefined format (such as a military report, intelligence report, etc.) according to the needs, or upload a custom format description and sample file, and combine it with the refined article outline to generate a report that conforms to specific writing norms, support the generation of articles in multiple formats and styles, such as academic, commercial, public opinion analysis, etc., meet the needs of diverse scenarios, flexibly adapt to the specific requirements of different application fields, and improve the practicality and customization ability.

[0036] As Figure 2 shown, if the user needs a graphic report, they can also use the graph - generation function to generate relevant pictures according to the paragraph summary, and then generate a graphic - rich report to improve the information - conveyance efficiency and expressiveness. The specific process includes:

[0037] First, use a language model to summarize the paragraph and generate an illustration description.

[0038] Second, according to the paragraph summary, illustration description, and report style, use an image - generation model to generate the corresponding pictures.

[0039] Then, extract the content paragraph by paragraph until the end of the report, and then polish it according to the report style.

[0040] Through the technology of generating reports by combining text and images, the system can support the generation of content that combines text with data visualization (such as charts and statistical analysis), enhancing the intuitiveness and expressiveness of the report. It automatically coordinates the logical relationship between the text and image content, improving the reading experience of the report and the efficiency of information transmission.

[0041] In addition, when generating a report, the system will retrieve and construct a relevant knowledge base, embed reference numbers, ensure that the source of the article content is clearly traceable, and enhance the credibility of the report. Each generated article will indicate the source of the reference, including the literature address, title, and part of the content. Users can also independently select references to improve the credibility and professionalism of the article. After generating the report, it supports exporting the generated content and related literature in PDF format for easy storage and sharing of the generated report.

[0042] The present invention also provides a two-stage long text report generation system, including:

[0043] An article outline generation module that generates a draft outline through a large model and simulates multiple rounds of conversations. According to the conversation records, themes, writing purposes, and style requirements, it further optimizes the draft outline to generate a refined article outline.

[0044] A report generation module that selects a predefined format according to user needs or uploads a custom format description and sample file to generate a report that conforms to specific writing norms, forms a traceable and displayable literature, and exports it in PDF format for storage and sharing.

[0045] A summary report generation module. When the user provides large-scale information and analysis dimensions, first, the information is preprocessed through filtering, scoring, and sorting for the analysis dimensions to generate a high-quality summary and establish a vector knowledge base. Using this knowledge base as the retrieval source to enter the system workflow, a comprehensive report containing key information and analysis results is finally generated to meet the needs of large-scale data processing and information extraction, as Figure 3 shown. In this module, the user can provide a custom file as a retrieval library for long text generation and use it as a knowledge base during the simulated conversation.

[0046] Therefore, the present invention adopts the above-mentioned two-stage long text report generation method and system. Through the retrieval-enhanced generation technology, it seamlessly integrates online retrieval information with the content of the knowledge base to efficiently generate a long text report with well-founded content; uses the intelligent agent technology to simulate expert collaboration to automatically generate a logically rigorous outline to ensure the integrity and clarity of the report structure; supports users to specify the report style, format, and required scenarios to meet diverse personalized needs; and through the ability to generate reports with pictures and texts, it improves the intuitiveness and expressiveness of the content, and is widely applicable to the report needs of different fields.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions of the present invention or make equivalent replacements, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A two-stage long text report generation method, characterized in that, It includes the following steps: S1. Generate requirements based on the report, conduct topic research and perspective discovery. Through the intelligent agent to simulate the conversation of domain experts, guide in-depth exploration of the report topic, establish a rich knowledge base, so as to ensure the quality of the article. And based on this knowledge base, generate a detailed article outline; S2. According to the user's word count and style requirements, combine the knowledge base and the article outline to write long texts, embed reference numbers to enhance the credibility of the report, and export the generated report content and relevant literature as PDF for the user to save and share.

2. The two-stage long text report generation method according to claim 1, wherein The topic research and perspective discovery include retrieving and analyzing articles related to the topic, exploring the topic from different angles, ensuring the depth and breadth of the content, discovering and generating diverse research perspectives, and screening out perspectives that meet the goals.

3. A two-stage long text report generation method according to claim 1, characterized in that Step S1 includes: S11. Generate report editors and experts based on web resources with different perspectives; S12. The report editor conducts simulated Q&A with the experts while building the knowledge base; S13. Generate the article outline according to the simulated Q&A, the knowledge base and the report generation requirements.

4. A two-stage long text report generation method according to claim 1, characterized in that Step S2 includes that when the user needs a graphic report, generate a graphic-rich report as follows: First, use the language model to summarize the paragraphs and generate illustration descriptions; Second, according to the paragraph summaries, illustration descriptions and report style, use the image generation model to generate corresponding pictures; Then, extract the content paragraph by paragraph until the end of the report, and polish it according to the report style.

5. A two-stage long text report generation system, characterized in that, It includes: The article outline generation module generates a draft outline through a large model and simulates multiple rounds of conversations. According to the conversation records, topics, writing purposes and style requirements, further optimize the draft outline to generate a refined article outline; The report generation module selects a predefined format according to the user's needs, or uploads a custom format description and sample file to generate a report that meets specific writing specifications, forms a traceable and displayable literature, and exports it in PDF format for saving and sharing; The abstract summary report generation module preprocesses the information according to the large-scale information and analysis dimensions provided by the user to generate high-quality abstracts and comprehensive reports.

6. A two-stage long text report generation system according to claim 5, characterized in that, Support diverse report customization generation, including style, format and scenario-based requirements.

Images