Document generation method and device based on multi-agent collaboration and medium

Abstract

The invention discloses a document generation method and device based on multi-agent collaboration and a medium, and relates to the technical field of artificial intelligence. The method comprises the steps of receiving a natural language research theme input by a user; inputting the research theme into a preset planning agent to generate a keyword set for executing network search according to the research theme; inputting the keyword set into a preset search agent, and executing a plurality of times of network search by utilizing a network search tool to obtain a plurality of key point type abstracts; and inputting the research theme and the plurality of key point type abstracts into a writing agent to generate a structured research report, and outputting the structured research report to a user interface.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a document generation method based on multi-agent collaboration, a device and a medium. BACKGROUND

[0002] In the field of computer software technology, especially in the direction of artificial intelligence driven information processing, the existing technology generally adopts a single dimension solution to meet the document generation demand. The traditional manual method relies on researchers to manually perform the whole process of searching, screening, integrating and writing, which has inherent defects of low efficiency and difficulty in guaranteeing information integrity. Although there are current automatic search tools and basic document generation models, the core limitation is the fragmentation of functional modules - the search engine only returns original web links or simple summaries, lacking the ability to strategically decompose complex research topics. And the writing model operating independently lacks precise structured input, resulting in logical gaps or missing key dimensions in the generated report.

[0003] The existing technology attempts to integrate search and writing functions through a single agent architecture, but due to the lack of a specialized division mechanism, it still exposes significant deficiencies when dealing with complex research tasks. On the one hand, a single agent cannot simultaneously consider multiple goals such as search strategy optimization, information density extraction, and report structure planning, resulting in output content often containing redundant information or deviating from the core topic. On the other hand, the lack of collaboration protocols between modules forms data silos between search and writing stages, with search results directly input into the writing module without deep compression, causing information overload and reducing the professionalism of the final report.

[0004] Therefore, how to establish a multi-agent collaboration mechanism to achieve seamless connection from research topic analysis, accurate search to structured report generation has become a technical problem that needs to be solved by those skilled in the art. SUMMARY

[0005] The embodiments of the present application provide a document generation method based on multi-agent collaboration, a device and a medium, to solve the technical problem of how to establish a multi-agent collaboration mechanism to achieve seamless connection from research topic analysis, accurate search to structured report generation.

[0006] In a first aspect, the embodiments of the present application provide a document generation method based on multi-agent collaboration, the method comprising: receiving a natural language research topic input by a user; inputting the research topic into a preset planning agent to generate a set of keywords for performing network search according to the research topic; inputting the set of keywords into a preset search agent to perform a plurality of network searches using a network search tool to obtain a plurality of point summaries; inputting the research topic and the plurality of point summaries into a writing agent to generate a structured research report, and outputting the structured research report to a user interface.

[0007] In an embodiment of the present application, the keyword set for performing the network search is generated according to the research topic, specifically comprising: parsing the syntax structure and semantic relationship of the research topic by using the natural language processing technology, and extracting the core concept entity based on the parsing result; combining the core concept entity to generate a plurality of keywords to obtain the keyword set.

[0008] In an embodiment of the present application, the network search tool is used to perform a plurality of network searches to obtain a plurality of point abstracts, specifically comprising: generating a plurality of keyword combinations based on the keywords in the keyword set; calling the network search tool to perform a search for each keyword combination to obtain a plurality of search results; and compressing and extracting the content of the plurality of search results to obtain a plurality of point abstracts; wherein the point abstract is a text abstract containing a plurality of paragraphs, only containing core point information and not containing subjective comments.

[0009] In an embodiment of the present application, the content of the plurality of search results is compressed and extracted to obtain a plurality of point abstracts, specifically comprising: analyzing the text content in the search result content to identify the core point information related to the keyword combination; deleting the repeated expressions and non-essential description content in the search result content; and reorganizing the remaining core point information into a text unit containing a plurality of paragraphs according to the logical relationship; wherein the text unit only retains objective fact description and does not contain subjective evaluation sentences.

[0010] In an embodiment of the present application, the structured research report is generated, specifically comprising: generating a to-be-applied report outline according to the research topic and the plurality of point abstracts; filling the report outline with the text content based on the plurality of point abstracts to generate the structured research report.

[0011] In an embodiment of the present application, the to-be-applied report outline is generated according to the research topic and the plurality of point abstracts, specifically comprising: analyzing the research topic and the plurality of point abstracts, and planning the overall structural framework and writing logic of the structured research report based on the analysis result; and constructing the to-be-applied report outline based on the overall structural framework and writing logic of the structured research report.

[0012] In an embodiment of the present application, after the to-be-applied report outline is constructed based on the overall structural framework and writing logic of the structured research report, the method further comprises: comparing the matching degree of the plurality of point abstracts and the node content of the to-be-applied report outline; when an abstract missing node that does not cover the core dimension of the research topic is identified, issuing a supplementary search instruction to the search agent; and based on the supplementary search instruction, searching by using the network search tool to obtain the to-be-supplemented point abstract.

[0013] In an embodiment of the present application, the structured research report is output to the user interface, specifically including: monitoring the generation state of each chapter of the report in real time in the process of generating the structured research report by the writing agent; when a new complete chapter is detected, immediately converting the chapter content into a streaming data transmission format; pushing the converted chapter data to the user interface for dynamic rendering through the server push event protocol.

[0014] In a second aspect, the embodiments of the present application also provide a document generation device based on multi-agent collaboration, the device comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any one of the above document generation methods based on multi-agent collaboration.

[0015] In a third aspect, the embodiments of the present application also provide a non-volatile computer storage medium storing computer executable instructions, and the computer executable instructions are executed to implement any one of the above document generation methods based on multi-agent collaboration.

[0016] The document generation method, device and medium based on multi-agent collaboration provided by the embodiments of the present application have the following beneficial effects: through the professional division and collaborative mechanism of the planning agent, the search agent and the writing agent, the technical contradiction that the efficiency and quality are difficult to balance in the field of document generation is fundamentally solved. The semantic analysis of the research theme by the planning agent realizes the accurate formulation of the search strategy, significantly improves the targeting of information acquisition; the key point summary generation executed by the search agent ensures the high information density and objectivity of the input content through forced information compression and redundancy filtering; the two-stage generation mode adopted by the writing agent combined with the dynamic outline adjustment mechanism maintains the rigor of the report structure while realizing the depth coverage of the content, finally achieves the collaborative gain of efficiency improvement in the whole process and strong guarantee of output quality. The multi-agent collaborative architecture not only reduces the need for human intervention to the minimum, but also realizes the end-to-end automation closed loop from the research theme input to the structured report output in the industrial application scenario, providing a revolutionary knowledge production tool for the fields of academic research, business analysis and the like. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings illustrated herein are used to provide further understanding of the present application, constitute a part of the present application, the illustrative embodiments of the present application and the description thereof are used to explain the present application, and do not constitute improper limitation on the present application. In the drawings: Figure 1 A document generation method based on multi-agent collaboration provided by the embodiments of the present application is shown in the flowchart; Figure 2This is a schematic diagram of the internal structure of a document generation device based on multi-agent collaboration, provided as an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] This application provides a document generation method, device, and medium based on multi-agent collaboration to solve the following technical problem: how to establish a multi-agent collaboration mechanism to achieve seamless integration of the entire process from research topic analysis and accurate search to structured report generation.

[0020] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0021] Figure 1 This document generation method based on multi-agent collaboration is illustrated in an embodiment of this application. Figure 1 As shown in the figure, the document generation method based on multi-agent collaboration provided in this application embodiment specifically includes the following steps: Step 101: Receive the natural language research topic input by the user.

[0022] In this embodiment, the user submits an unstructured research topic description, such as "the latest advances of artificial intelligence in medical image diagnosis," through a graphical user interface (such as a web chat box or a mobile application input area).

[0023] Understandably, "natural language" here specifically refers to unformatted, everyday human expression. Its core principle is to allow users to express their needs in free text form, without adhering to specific grammatical templates. It's important to note that the system immediately initiates a preprocessing mechanism upon receiving input: first, it filters out irrelevant characters (such as advertising code and special symbols); then, it detects and blocks inappropriate content (such as sensitive words); and finally, it encapsulates the purified text into a structured data object (such as a JSON-formatted {"query":"The latest advancements in artificial intelligence in medical image diagnosis"}).

[0024] Step 102: Input the research topic into the preset planning agent to generate a set of keywords for performing web searches based on the research topic.

[0025] In an embodiment of the present application, a keyword set for performing a network search is generated according to a research topic, specifically comprising: using a natural language processing technique to parse the grammatical structure and semantic relationship of the research topic, and extracting core concept entities based on the parsing result; combining the core concept entities to generate a plurality of keywords to obtain the keyword set.

[0026] In the embodiment, the planning agent serves as a task decomposition engine and receives the research topic text preprocessed in step 101. It can be understood that the agent uses a hierarchical semantic parsing technique: first, a syntax tree is constructed through dependency syntax analysis (such as identifying "artificial intelligence" as the subject and "medical image diagnosis" as the object); then, based on the domain knowledge graph, the core concept entities are located (such as mapping "artificial intelligence" to the technical entity set ["machine learning", "deep learning"], and associating "medical image diagnosis" to the application entity set ["CT image analysis", "MRI lesion identification"]), which directly implements the feature of "parsing grammatical structure and semantic relationship to extract core concept entities".

[0027] It should be noted that the entity combination strategy uses a dynamic weight mechanism: high-frequency co-occurring entities (such as the strong association of "deep learning" and "medical image" in literature) are automatically merged into composite keywords (generating "deep learning medical image application" instead of independent keywords), while weakly associated entities are retained for independent search value (such as retaining "medical data privacy" independently). Specifically, this combination mechanism is significantly different from the mechanical enumeration mode of traditional keyword generation tools, as each keyword generated by it carries logical association information between entities (such as "artificial intelligence + medical image diagnosis + ethical challenges" forming a triple structure), providing semantic guidance for the subsequent search agent in performing keyword combination search. Exemplarily, when the input topic is "new energy battery fast charging technology bottleneck", the planning agent not only extracts basic entities such as "lithium ion battery" and "fast charging protocol", but also generates professional term combinations such as "solid-state electrolyte interface growth" and "lithium precipitation effect" through semantic association, ensuring that the keyword set covers multi-dimensional perspectives of technical principle layer and problem layer.

[0028] Step 103, inputting the keyword set into a preset search agent to perform a plurality of network searches using a network search tool to obtain a plurality of point abstracts.

[0029] In an embodiment of the present application, the network search tool is used to perform several network searches to obtain several point abstracts, specifically including: generating several keyword combinations based on the keywords in the keyword set; calling the network search tool to perform a search for each keyword combination to obtain several search results; and compressing and extracting the content of the several search results to obtain several point abstracts; wherein the point abstract is a text abstract containing multiple paragraphs, only containing core point information and no subjective comments.

[0030] In an embodiment of the present application, the content of the several search results is compressed and extracted to obtain several point abstracts, specifically including: analyzing the text content in the search result content to identify core point information related to the keyword combination; deleting repetitive expressions and non-essential description content in the search result content; and reorganizing the retained core point information into a text unit containing multiple paragraphs according to the logical relationship; wherein the text unit only retains objective fact descriptions and does not contain subjective evaluation sentences.

[0031] In this embodiment, the search agent starts the adaptive grouping strategy based on the semanticized keyword set output by the planning agent (such as ["solid electrolyte interface growth", "lithium precipitation risk" in the new energy battery case]). It can be understood that this strategy first calculates the semantic coupling degree between entities, automatically merges high coupling keywords (such as "interface growth" and "dendrite formation" have strong causal relationship) into the same search group (generates a composite search instruction "solid electrolyte interface growth AND dendrite formation"), and keeps independent search groups for isolated entities (such as "charging pile compatibility").

[0032] It should be noted that the deep crawling mode of the network search tool is called when each group of keywords performs a search: preferentially obtaining authoritative sources (such as academic journals, patent databases), and stripping non-essential content such as advertisements and navigation bars through DOM tree analysis.

[0033] Specifically, the compression and extraction of the original content adopts a three-stage process: the first stage identifies technical core parameters (such as extracting "SEI membrane thickening rate", "lithium deposition critical current density" and other key data from long papers), the second stage deletes repetitive discussions (such as only keeping the most complete version of the experimental method in multiple papers that are similar), and the third stage reorganizes the information according to the "problem-mechanism-impact" logical chain (such as integrating fragmented data into "1. Problem: fast charging causes interface growth; 2. Mechanism: lithium ion concentration gradient causes phase change; 3. Impact: battery cycle life is reduced"), and the finally generated point abstract strictly meets the "multi-paragraph text unit containing only objective facts" defined in claim 4.

[0034] Exemplarily, when processing the "fast charging safety standard" search group, the summary forces to exclude subjective evaluations (such as deleting "a certain manufacturer's scheme is extremely innovative"), and only retains the regulatory provisions ("International Electrotechnical Commission IEC 62660-3 stipulates that fast charging rate > 3C requires additional overcharge protection") and technical facts ("Graphite negative electrode lithium intercalation potential platform < 0.1V is easy to cause lithium precipitation").

[0035] In step 104, the research topic and the several summary points are input into the writing agent to generate a structured research report, and the structured research report is output to the user interface.

[0036] In an embodiment of the present application, the structured research report is generated, specifically including: generating a to-be-applied report outline according to the research topic and the several summary points; and filling the report outline with text content based on the several summary points to generate the structured research report.

[0037] In an embodiment of the present application, the to-be-applied report outline is generated according to the research topic and the several summary points, specifically including: analyzing the research topic and the several summary points, and planning an overall structural framework and writing logic of the structured research report based on the analysis result; and constructing the to-be-applied report outline based on the overall structural framework and writing logic of the structured research report.

[0038] In an embodiment of the present application, after the to-be-applied report outline is constructed based on the overall structural framework and writing logic of the structured research report, the method further includes: comparing the matching degrees of the several summary points and the node content of the to-be-applied report outline; when an abstract missing node that does not cover the core dimension of the research topic is identified, initiating a supplementary search instruction to the search agent; and based on the supplementary search instruction, searching by using a network search tool to obtain a to-be-supplemented summary point.

[0039] In an embodiment of the present application, the structured research report is output to the user interface, specifically including: monitoring the generation state of each chapter of the report in real time during the process that the writing agent generates the structured research report; when a new complete chapter is detected to be generated, immediately converting the chapter content into a streaming data transmission format; and pushing the converted chapter data to the user interface for dynamic rendering by using a server push event protocol to segment the data.

[0040] In this embodiment, the writing agent starts a two-stage generation protocol: in the first stage of constructing a dynamic outline, a report framework template ("present situation analysis → technical conflict → regulatory comparison → governance suggestion") is automatically matched based on the field attribute of the research topic (such as "blockchain technology regulatory difficulties" being identified as a policy research type).

[0041] It can be understood that the process synchronously performs the abstract-outline matching detection: the key semantic nodes (such as "EU MiCA regulation requirements") of each point abstract are mapped to the corresponding chapter of the outline (the "regulation comparison" node), and when a key dimension is missing (such as the "governance suggestion" node without abstract support), a supplementary instruction is immediately sent to the search agent, and the instruction content fuses the original research topic and the missing dimension identifier (generates a new search word: "blockchain regulatory sandbox implementation case").

[0042] It should be noted that in the text generation stage, the agent does not simply copy the abstract content: for the first occurrence position of technical terms (such as "zero-knowledge proof"), automatically insert bracket annotations to explain the basic principles (such as "a cryptographic protocol in which the verifier does not learn the specific data"); add logical conjunction words at the chapter transition (such as "in view of the above regulatory differences" to connect the regulation comparison and governance suggestion chapters).

[0043] In this embodiment, the output link adopts a real-time segmented streaming mechanism: when the writing agent completes any chapter (such as "technical conflict analysis"), the content generation listening module immediately converts it into an event stream data packet (containing chapter ID, MD format text, and completion timestamp), which is transmitted to the front end in batches through the server push event protocol, and the user interface dynamically renders the chapter card and supports independent editing. For example, when generating the "new energy battery recycling technology" report, the user can make comments and revisions on the "hydrometallurgical process" chapter that has been completed in advance while the writing agent is still writing the "economic analysis" chapter, realizing a human-machine collaborative document production closed loop.

[0044] The above is a method embodiment of the present application. Based on the same inventive concept, the embodiments of the present application also provide a document generation device based on multi-agent collaboration, which has a structure as shown in Figure 2 .

[0045] Figure 2 A document generation device based on multi-agent collaboration provided by an embodiment of the present application has an internal structure as shown in Figure 2 . The device includes: at least one processor 201; and a memory 202 in communication connection with the at least one processor; wherein the memory 202 stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: receive a natural language research topic input by a user; inputting the keyword set into a preset search agent to perform a plurality of network searches by using a network search tool to obtain a plurality of point abstracts; inputting the research topic and the plurality of point abstracts into a writing agent to generate a structured research report, and outputting the structured research report to a user interface.

[0046] Some embodiments of the present application provide a non-volatile computer storage medium corresponding to Figure 1 storing computer executable instructions, the computer executable instructions being arranged to: receive a natural language research topic input by a user; inputting the research topic into a preset planning agent to generate a keyword set for performing network searches according to the research topic; inputting the keyword set into a preset search agent to perform a plurality of network searches by using a network search tool to obtain a plurality of point abstracts; inputting the research topic and the plurality of point abstracts into a writing agent to generate a structured research report, and outputting the structured research report to a user interface.

[0047] Each of the embodiments in the present application is described in a progressive manner, and the same or similar parts of each of the embodiments can be referred to each other. Each of the embodiments mainly describes the differences from other embodiments. In particular, the embodiments of the Internet of Things device and medium are described simply because they are basically similar to the method embodiments, and the relevant parts can be referred to the description of the method embodiments.

[0048] The system and medium provided by the embodiments of the present application are one-to-one corresponding to the method, and therefore, the system and medium have similar beneficial technical effects to the method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be described here.

[0049] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0050] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0051] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0052] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0053] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0054] The memory can include non-persistent memory and / or volatile memory, such as a random access memory (RAM) including a cache area for the temporary storage of data. A

[0055] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0056] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0057] The above description is only an embodiment of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the claims of the present application.

Claims

1. A document generation method based on multi-agent collaboration, characterized in that, The method comprises: receiving a user-input natural language research topic; inputting the research topic into a preset planning agent to generate a keyword set for performing a network search according to the research topic; inputting the keyword set into a preset search agent to perform a plurality of network searches by using a network search tool to obtain a plurality of key point summaries; inputting the research topic and the plurality of key point summaries into a writing agent to generate a structured research report and outputting the structured research report to a user interface. 2.The document generation method based on multi-agent cooperation according to claim 1, characterized in that, Generating a keyword set for performing a network search according to the research topic specifically comprises: using natural language processing technology to analyze the grammatical structure and semantic relationship of the research topic and extracting core concept entities based on the analysis result; combining the core concept entities to generate a plurality of keywords to obtain the keyword set. 3.The document generation method based on multi-agent cooperation according to claim 1, characterized in that, Performing a plurality of network searches by using a network search tool to obtain a plurality of key point summaries specifically comprises: generating a plurality of keyword combinations based on the keywords in the keyword set; for each keyword combination, calling a network search tool to perform a search to obtain a plurality of search results; compressing and extracting the content of the plurality of search results to obtain the plurality of key point summaries; wherein the key point summary is a text summary containing multiple paragraphs, only containing core point information and not containing subjective comments.

4. The document generation method based on multi-agent collaboration according to claim 3, characterized in that, Compressing and extracting the content of the plurality of search results to obtain the plurality of key point summaries specifically comprises: analyzing the text content in the search result content to identify core point information related to the keyword combination; deleting repetitive expressions and non-essential description content in the search result content; reorganizing the remaining core point information into a text unit containing multiple paragraphs according to the logical relationship; wherein the text unit only retains objective fact descriptions and does not contain subjective evaluation sentences.

5. The method of claim 1, wherein, Generating a structured research report specifically comprises: generating a report outline to be applied according to the research topic and the plurality of key point summaries; based on the plurality of key point summaries, filling the report outline with text content to generate a structured research report.

6. The method of claim 4, wherein, Generating a report outline to be applied according to the research topic and the plurality of key point summaries specifically comprises: analyzing the research topic and the plurality of key point summaries and planning the overall structural framework and writing logic of the structured research report based on the analysis result; constructing the report outline to be applied based on the overall structural framework and writing logic of the structured research report.

7. The method of claim 6, wherein, After constructing the report outline to be applied based on the overall structural framework and writing logic of the structured research report, the method further comprises: comparing the matching degree of the plurality of key point summaries and the node content of the report outline to be applied; when identifying that there is a missing summary node that does not cover the core dimension of the research topic, initiating a supplementary search instruction to the search agent; based on the supplementary search instruction, searching by using the network search tool to obtain a key point summary to be supplemented.

8. The document generation method based on multi-agent collaboration according to claim 1, characterized in that, Outputting the structured research report to a user interface specifically comprises: In the process of generating the structured research report by the writing agent, the generation state of each chapter of the report is monitored in real time; When a new complete chapter is detected to be generated, the chapter content is immediately converted into a streaming data transmission format; The converted chapter data is segmented and pushed to the user interface for dynamic rendering through a server push event protocol.

9. A document generation device based on multi-agent cooperation, characterized by, The device comprises: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a document generation method based on multi-agent collaboration as claimed in any one of claims 1-8.

10. A non-transitory computer storage medium storing computer-executable instructions that, when executed, cause a computer to perform: The computer executable instructions, when executed, implement a document generation method based on multi-agent collaboration as claimed in any one of claims 1-8.

Images