CN-121981082-A - Automatic report generation method and system based on knowledge enhancement and multi-agent cooperation

CN121981082ACN 121981082 ACN121981082 ACN 121981082ACN-121981082-A

Abstract

The application belongs to the technical field of automatic report generation, and discloses an automatic report generation method and system based on knowledge enhancement and multi-agent cooperation, wherein the method analyzes natural language requirements of users, constructs task elements, performs initial semantic retrieval, acquires basic related document sets, analyzes seed documents, combines report types and audience logics, generates and refines a structured preliminary outline, converts each chapter node of the preliminary outline into an enhanced outline with specific knowledge points and document sources, circularly executes a set cooperation process by taking chapters as units to complete content writing, synthesizes each chapter into a complete report, performs automatic post-processing and format optimization, and presents and delivers users; the system comprises a user interaction module, a multi-agent collaboration engine and a knowledge base module. The application realizes full-automatic and pipelined generation from the user input of the subject to the complete professional report, and greatly improves the efficiency.

Inventors

MIN YANLI
Ke Chunxiao
WEI SHUANGSHUANG
LIN YING
Shen Qiangbin
WU FENGMIN
WANG LI
MA LUMING
YIN DONGMIN
DUAN FEIHU

Assignees

同方知网数字科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251226

Claims (10)

1. An automatic report generation method based on knowledge enhancement and multi-agent cooperation, which is characterized by comprising the following steps: Analyzing natural language requirements of users, structuring task elements, and executing initial semantic retrieval to obtain a basic related document set; step 2, analyzing seed literature, combining report types and audience logic, and generating and refining a structured preliminary outline; Step 3, converting each chapter node of the preliminary outline into an enhanced outline attached with specific knowledge points and literature sources; Step 4, circularly executing the set collaborative process by taking the chapters as units to finish content writing; and 5, synthesizing each chapter into a complete report, performing automatic post-processing and format optimization, and presenting and delivering to a user.
2. The automated knowledge-based enhanced multi-agent collaborative report generating method according to claim 1, wherein step1 comprises: Step 1.1, receiving natural language task description submitted by a user; Analyzing task description, identifying and extracting core elements, and generating a structured task work order; step 1.3, constructing a search query according to a task work order, and initiating initial semantic search to a knowledge base vector database; and 1.4, receiving an initial document set which is returned by the knowledge base and is ranked according to semantic relevance and is used as a follow-up working basis.
3. The automatic report generating method based on knowledge enhancement and multi-agent collaboration according to claim 1, wherein step 2 comprises: Step 2.1, deeply analyzing a seed literature set, and summarizing core issues, disputed points, technical paths and key conclusions; Step 2.2, constructing a preliminary tree-like report outline according to the general structure and audience narrative logic; Step 2.3, adding a hierarchy and a brief content description for the outline chapter to clarify the core problem and the content range; and 2.4, submitting the final structured outline to a central controller as the overall planning of the project.
4. The automatic report generating method based on knowledge enhancement and multi-agent collaboration according to claim 1, wherein step 3 comprises: Step 3.1, distributing each chapter node of the preliminary outline to the technical key point extraction and deepening agent treatment; Step 3.2, the agent performs focusing search and depth analysis on each chapter node, and screens out the most relevant document subset; Step 3.3, identifying and extracting the structural core technical key points from the related documents; And 3.4, marking a literature source for each extracted key point, and outputting an enhanced outline with a knowledge point list attached to the original structure.
5. The automated knowledge-based enhanced multi-agent collaborative report generating method according to claim 1, wherein step 4 comprises: Step 4.1, the writing agent initiates accurate secondary retrieval to a self-built document library based on chapter association points to obtain core writing materials; Step 4.2, searching and controlling the intelligent agent to monitor the writing context, and judging whether a knowledge gap exists according to a preset rule base; step 4.3, if the gap is judged to exist, generating a specific query instruction for internet retrieval, and performing credibility filtering and content extraction on the result; and 4.4, integrating all data by the writing agent, and generating texts according to a set paradigm to form microcirculation.
6. The automated knowledge-based enhanced multi-agent collaborative report generating method according to claim 1, wherein step 5 comprises: step 5.1, the central controller combines all written chapter texts according to the outline sequence to generate a complete report manuscript; Step 5.2, post-processing, including unified quotation and reference list generation, format typesetting and chapter integrity and quotation condition checking; and 5.3, outputting a final report, supporting preview and downloading, and receiving user feedback to start a correction flow.
7. An automated report generation system based on knowledge enhancement in conjunction with multiple agents, comprising: the user interaction module is used for converting the fuzzy unstructured requirements of the user into standardized task instructions; the multi-agent coordination engine is used for coordinating a plurality of special agents through the central controller and converting the complex task demands into a structured document based on evidence and logic; The knowledge base module is used for storing structured professional documents, supporting efficient semantic retrieval, acquiring, filtering and archiving external information through the dynamic internet interface, and guaranteeing information quality.
8. The automated knowledge-based enhanced multi-agent collaborative report generating system according to claim 7, wherein the user interaction module includes: the demand acquisition and analysis unit is used for receiving natural language description input by a user; the task state visualization and feedback unit is used for displaying task progress and key information to a user in real time, providing an interactive interface to confirm or adjust key nodes, and optimizing a subsequent flow by the system according to the task progress and key information; And the output presentation and post-processing unit is used for rendering the finally generated structured report data into a format appointed by a user and outputting the structured report data.
9. The knowledge-based enhanced multi-agent collaboration based automated report generation system of claim 7, wherein the multi-agent collaboration engine comprises: The central controller is used for arranging a collaborative process, scheduling tasks, transmitting contexts and branching the decision process according to rules or feedback; The understanding and planning agent is used for decomposing tasks, analyzing data logic and generating a preliminary outline and a blueprint with writing intention; the technical key point is to extract and deepen the intelligent agent, which is used for deeply analyzing and extracting the structured knowledge points from the literature and attaching core knowledge for outline nodes; The retrieval control intelligent agent is used for monitoring information gaps in writing, automatically deciding and triggering supplementary retrieval according to policy rules; and the writing agent is used for generating a coherent narrative text which can be searched according to the enhancement outline and the attached knowledge points and the search evidence.
10. The automated knowledge-based reinforcement and multi-agent collaborative report generating system according to claim 7, wherein the knowledge base module comprises: The self-built professional literature vector database is used for converting unstructured text in the field into a high-dimensional vector through preprocessing and embedding models, and storing the high-dimensional vector in the vector database; the dynamic internet retrieval and adaptation interface is used for packaging search engine call, performing credibility filtering and content extraction, and enabling results to be used by the writing agent and to support archiving to a knowledge base.

Description

Automatic report generation method and system based on knowledge enhancement and multi-agent cooperation Technical Field The application belongs to the technical field of automatic report generation, and particularly relates to an automatic report generation method and system based on knowledge enhancement and multi-agent cooperation. Background With the explosive growth of information, in the fields of academic research, market analysis, policy consultation, project declaration and the like, writing a high-quality professional report needs to undergo complex processes such as massive document retrieval, content carding, main point induction, logic organization, text writing and the like. The traditional mode mainly relies on manual work, and has the following defects: (1) Low efficiency, time and labor consumption. Manual review, understanding and integration of large amounts of data requires extremely long periods. (2) Quality depends on personal experience. The comprehensiveness, accuracy and rationality of the structure of the report content are highly dependent on the expertise and experience of the writer, and it is difficult to ensure consistency. (3) Knowledge coverage is incomplete. Individuals cannot exhaust all relevant documents and easily miss critical information or make up-to-date progress. (4) The structure is difficult to optimize. It is difficult for a human to quickly try multiple report organization schemes to find the optimal logical expression. At present, although a technology for generating text by using a large model exists, the technology generally has the problems of "illusion" (generating false information), content hollowness, lack of professional depth, single logic structure and the like. Reports generated based solely on large model general knowledge are far from satisfactory in terms of specificity and credibility. Disclosure of Invention The application aims to overcome the defects of the prior art, thereby providing an automatic report generation method and an automatic report generation system based on knowledge enhancement and multi-agent cooperation, which can automatically generate professional reports with high quality, high efficiency and high credibility. In order to achieve the above object, the present application provides the following technical solutions: in a first aspect, the present application provides an automatic report generating method based on knowledge enhancement and multi-agent collaboration, including: Analyzing natural language requirements of users, structuring task elements, and executing initial semantic retrieval to obtain a basic related document set; step 2, analyzing seed literature, combining report types and audience logic, and generating and refining a structured preliminary outline; Step 3, converting each chapter node of the preliminary outline into an enhanced outline attached with specific knowledge points and literature sources; Step 4, circularly executing the set collaborative process by taking the chapters as units to finish content writing; and 5, synthesizing each chapter into a complete report, performing automatic post-processing and format optimization, and presenting and delivering to a user. As an embodiment, step 1 includes: Step 1.1, receiving natural language task description submitted by a user; Analyzing task description, identifying and extracting core elements, and generating a structured task work order; step 1.3, constructing a search query according to a task work order, and initiating initial semantic search to a knowledge base vector database; and 1.4, receiving an initial document set which is returned by the knowledge base and is ranked according to semantic relevance and is used as a follow-up working basis. As an embodiment, step 2 includes: Step 2.1, deeply analyzing a seed literature set, and summarizing core issues, disputed points, technical paths and key conclusions; Step 2.2, constructing a preliminary tree-like report outline according to the general structure and audience narrative logic; Step 2.3, adding a hierarchy and a brief content description for the outline chapter to clarify the core problem and the content range; and 2.4, submitting the final structured outline to a central controller as the overall planning of the project. As an embodiment, step 3 includes: Step 3.1, distributing each chapter node of the preliminary outline to the technical key point extraction and deepening agent treatment; Step 3.2, the agent performs focusing search and depth analysis on each chapter node, and screens out the most relevant document subset; Step 3.3, identifying and extracting the structural core technical key points from the related documents; And 3.4, marking a literature source for each extracted key point, and outputting an enhanced outline with a knowledge point list attached to the original structure. As an embodiment, step 4 includes: Step 4.1, the writing agent initiates accurate secondary retriev