Document generation method and system based on autonomous agent

Abstract

The invention discloses a document generation method and system based on an autonomous agent, the autonomous agent realizes efficient decision by integrating memory information and deep demand analysis, and the core is that historical interaction data and domain knowledge are utilized to form a memory module; multi-level disassembly (such as intention recognition, implicit condition mining and associated scene analysis) is performed in combination with a context background of a user demand, so that the problem of demand understanding fragmentation is effectively solved, and the accuracy is remarkably improved; besides, a structured plan generated based on memory information defines a tool calling sequence and operation steps, resources such as a database and an API are dynamically called through a protocol layer tool, historical experience is reused, repeated analysis is avoided, and the decision process is accelerated; besides, an evaluation-adjustment closed-loop mechanism is adopted in plan execution, information integrity is verified in real time, and if a gap exists, the plan is intelligently adjusted or tool calling is supplemented, so that output comprehensiveness is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of document generation, in particular to a document generation method and system based on autonomous agents. BACKGROUND

[0002] With the rapid development of artificial intelligence technology, document generation automation has become an important demand to improve efficiency. Traditional document generation methods rely on pre-set templates or rule engines, which have the problems of insufficient flexibility, poor adaptability, etc., and are difficult to cope with complex or dynamic demand scenarios. In recent years, although natural language processing (NLP) technology based on deep learning has made progress in the field of text generation, most systems still have the following limitations: (1) Insufficient understanding of demand and context association: existing methods usually directly generate text based on user input, lack long-term memory and association analysis of user historical demand or domain knowledge, resulting in generated content that may deviate from the user's actual intent or lack coherence; (2) Low efficiency of knowledge utilization: in the document generation process, the system often cannot effectively extract and reuse historical information (such as user preferences, domain terminology, or past document fragments) related to the current task, resulting in repetitive work or information redundancy; (3) Limited tool collaboration and information acquisition capability: in complex document generation tasks that require the combination of external tools (such as database queries, API calls), hard-coded logic is required to call tools, making it difficult to dynamically adapt to the context interaction needs of different tools, resulting in low information acquisition efficiency. SUMMARY

[0003] Therefore, the present application proposes a document generation method and system based on autonomous agents, which can effectively solve the defects of insufficient understanding of demand and context association, low efficiency of knowledge utilization, and limited tool collaboration and information acquisition capability in the prior art.

[0004] The technical solution of the present application is as follows:

[0005] A document generation method based on autonomous agents, specifically comprising:

[0006] Obtaining a demand text input by a user and sending it to an autonomous agent;

[0007] The autonomous agent extracts memory information related to the demand text as a reference for knowledge based on the demand text;

[0008] The autonomous agent performs a deep demand analysis on the demand text with the memory information as a reference, and generates detailed plan content including model context protocol layer tools and operation steps;

[0009] According to the operation steps of the plan content, the autonomous agent determines the model context protocol layer tools to be called, and obtains information directly related to the execution of the operation steps through the model context protocol layer tools;

[0010] The autonomous intelligent agent assesses whether the currently acquired information meets the planned content execution requirements, and if so, generates a final document based on the information integration, otherwise adjusts the planned content or supplements the information.

[0011] As a further optional solution of the autonomous intelligent agent-based document generation method, the autonomous intelligent agent extracts memory information related to the requirement text as knowledge reference based on the requirement text, specifically including:

[0012] Receiving the requirement text input by the user, and parsing the domain, theme and task target in the requirement text according to natural language processing technology;

[0013] According to the analysis result, retrieving historical document templates, case data or general knowledge matching the domain, theme and task target of the requirement text from the memory module as memory information;

[0014] Transferring the extracted memory information to the autonomous intelligent agent as a knowledge reference basis for requirement analysis and plan generation;

[0015] If no directly matching memory information is retrieved, a new template is dynamically generated through a memory updating mechanism or white-box knowledge is called to supplement the knowledge base.

[0016] As a further optional solution of the autonomous intelligent agent-based document generation method, the new template is dynamically generated through a memory updating mechanism, specifically including:

[0017] Based on the short-term memory module, real-time storage of the context information of the current session is performed, and the context is summarized and compressed to extract key demand features;

[0018] According to the key demand features, it is determined whether there is an existing template of the same domain and theme in the long-term memory module; if there is no existing template, a template framework is generated based on the key demand features, and specific content is filled in combination with user modification feedback in the short-term memory module, and finally stored in the long-term memory module; if there is an existing template, the content structure, data dimension and conclusion logic of the new and old templates are compared, and the advantages and disadvantages of the new and old templates are analyzed in combination with user modification feedback in the short-term memory module to generate a difference evaluation report;

[0019] According to the difference evaluation report, an optimized new template is generated according to a preset fusion rule, and is updated to the long-term memory module.

[0020] As a further optional solution of the autonomous intelligent agent-based document generation method, the white-box knowledge is called to supplement the knowledge base, specifically including:

[0021] Based on the long-term memory module, the commonalities of high-quality templates and the commonalities of low-quality templates are extracted;

[0022] According to the commonalities of the high-quality templates and the commonalities of the low-quality templates, a structured knowledge item is matched from the white-box knowledge base;

[0023] The structured knowledge item is embedded into a general logic framework to generate a new template framework.

[0024] As a further optional solution of the document generation method based on the autonomous agent, the autonomous agent takes the memory information as a reference to perform a deep demand analysis on the demand text, and generates detailed plan content including a model context protocol layer tool and operation steps, specifically including:

[0025] The autonomous agent performs multi-level semantic analysis on the input demand text by using the memory information and natural language processing technology, and generates a demand feature vector;

[0026] According to the demand feature vector, a model context protocol layer tool is selected from a preset tool library;

[0027] Based on the model context protocol layer tool and the demand feature vector, a structured operation step is generated.

[0028] As a further optional solution of the document generation method based on the autonomous agent, the model context protocol layer tool includes:

[0029] A similar document recall tool for recalling relevant documents from a local database;

[0030] An Internet search tool for performing Internet search on ambiguous text in the recalled relevant documents and obtaining results;

[0031] A draft generation tool for generating a document draft by using natural language generation technology in combination with the recalled relevant documents and the information obtained through Internet search;

[0032] A document writing tool for converting the document draft into a preset format and performing document writing;

[0033] A document format conversion tool for converting the preset format document into a document general format.

[0034] A document generation system based on an autonomous agent, comprising:

[0035] A user interaction module for obtaining a demand text input by a user and sending the demand text to the autonomous agent;

[0036] A memory information processing module for extracting memory information related to the demand text as knowledge reference by the autonomous agent based on the demand text;

[0037] The demand analysis and plan generation module is configured to perform deep demand analysis on the demand text by referring to the memory information, and generate detailed plan content including model context protocol layer tools and operation steps.

[0038] The tool calling and information obtaining module is configured to determine the model context protocol layer tools required to be called by the autonomous agent according to the operation steps of the plan content, and obtain information directly related to the execution of the operation steps through the model context protocol layer tools.

[0039] The evaluation and document generation module is configured to evaluate whether the information currently obtained meets the execution requirements of the plan content by the autonomous agent, and if so, generate a final document based on information integration, otherwise adjust the plan content or supplement information.

[0040] As a further optional solution of the autonomous agent-based document generation system, the memory information processing module comprises:

[0041] The demand analysis module is configured to receive the demand text input by the user, and analyze the domain, theme and task target in the demand text according to natural language processing technology;

[0042] The memory information retrieval module is configured to retrieve historical document templates, case data or general knowledge matching the domain, theme and task target of the demand text from the memory module as memory information according to the analysis result;

[0043] The knowledge reference module is configured to deliver the extracted memory information to the autonomous agent as a knowledge reference basis for demand analysis and plan generation;

[0044] The memory updating and white box knowledge supplement module is configured to dynamically generate a new template or call white box knowledge supplement knowledge base through a memory updating mechanism if no directly matching memory information is retrieved.

[0045] A computing device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above autonomous agent-based document generation methods when executing the computer program.

[0046] A computer readable storage medium, wherein the storage medium stores a computer program, and the computer program is executable on a processor to implement the steps of any one of the above autonomous agent-based document generation methods.

[0047] The beneficial effects of the present application are: the autonomous intelligent agent extracts memory information related to the demand text as knowledge for reference, and generates a plan combined with deep demand analysis, the memory information contains historical interaction data or domain knowledge, so that the intelligent agent can understand in combination with the historical background or associated context of user demand, avoid isolated processing of the text, the deep demand analysis performs multi-level disassembly (such as intention, implicit condition, associated scene) on the demand text based on the memory information, significantly improves the accuracy of demand understanding, and solves the problem of fragmented demand understanding; in addition, the autonomous intelligent agent generates a detailed plan based on the memory information as a reference, and dynamically calls tools and information based on the plan, the memory information as a pre-stored knowledge base avoids repeated analysis of known content, directly reuses historical experience to speed up decision-making, integrates fragmented knowledge into a logically coherent generation process through a structured plan (protocol layer tool + operation steps), reduces redundant calculation, and realizes efficient calling and combination of knowledge; in addition, the autonomous intelligent agent calls the model context protocol layer tool according to the plan, and supplements information through an evaluation-adjustment closed loop, the plan content clearly indicates the tool calling sequence and target, the protocol layer tool obtains information (such as database query, API calling) as needed, avoids disordered interaction between tools, the evaluation mechanism checks the integrity of the information, adjusts the plan or calls the tool if it is insufficient, ensures the comprehensiveness of the final document, and realizes the efficiency of tool collaboration and the robustness of information acquisition. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0049] Fig. 1 The flowchart of the document generation method based on the autonomous intelligent agent of the present application;

[0050] Fig. 2 The composition diagram of the document generation system based on the autonomous intelligent agent of the present application;

[0051] Fig. 3 The composition diagram of the computing device of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0053] Reference Figs. 1 to 3 A document generation method based on autonomous agents, specifically comprising:

[0054] Obtain the requirement text input by the user and send it to the autonomous agent.

[0055] The autonomous agent queries the memory module based on the requirement text, extracts memory information related to the domain, theme or scenario of the requirement text as knowledge reference, specifically including:

[0056] Receive the requirement text input by the user, and parse the domain, theme and task target in the requirement text according to natural language processing technology;

[0057] According to the analysis result, retrieve the historical document template, case data or general knowledge matching the domain, theme and task target of the requirement text from the memory module as memory information;

[0058] Pass the extracted memory information to the autonomous agent as the knowledge reference for requirement analysis and plan generation;

[0059] If no directly matching memory information is retrieved, a new template is dynamically generated or white-box knowledge is called to supplement the knowledge base through the memory update mechanism.

[0060] Specifically, the natural language processing technology is used to parse the domain, theme and task target in the user requirement text, accurately identify the user intent, reduce requirement ambiguity, provide structured input for subsequent knowledge retrieval, support complex requirement analysis in multiple domains and multiple themes, and enhance the generalization ability of the system;

[0061] Through the memory module, historical document templates, case data or general knowledge are quickly retrieved to avoid repeated generation of similar content, directly reuse existing high-quality resources, significantly improve response speed, reduce computing resource consumption, and especially have obvious efficiency advantages when handling common requirements;

[0062] Dynamic generation of new templates or calling of white-box knowledge to supplement unmatched memory information can provide basic support for new domain requirements through white-box knowledge (such as rule base, basic theory), and continuously accumulate new templates through the memory update mechanism, gradually covering more scenarios and improving the long-term applicability of the system;

[0063] The retrieval and update mechanism of the memory module forms a closed-loop feedback, continuously enriches the knowledge base during operation, forms an "experience learning" effect, and makes subsequent similar requirement processing more efficient. The collaborative use of historical cases and newly generated templates improves the comprehensiveness and reliability of the solution;

[0064] The end-to-end automated process from requirement analysis to knowledge retrieval and dynamic generation reduces the degree of human participation, especially in standardized tasks, achieving unattended operation. The dynamic generation template function reduces the maintenance cost of the knowledge base.

[0065] In some embodiments, the new template is dynamically generated by a memory updating mechanism, specifically including:

[0066] The short-term memory module stores the context information of the current session in real time, and abstractly compresses the context to extract key demand features;

[0067] According to the key demand features, it is determined whether there is an existing template of the same field and the same theme in the long-term memory module; if there is no existing template, a template framework is generated based on the key demand features, and specific content is filled in combination with the user modification feedback in the short-term memory module, and finally stored in the long-term memory module; if there is an existing template, the content structure, data dimension and conclusion logic of the new and old templates are compared, and the advantages and disadvantages of the new and old templates are analyzed in combination with the user modification feedback in the short-term memory module to generate a difference evaluation report;

[0068] According to the difference evaluation report, an optimized new template is generated according to a preset fusion rule (such as retaining the new template framework and supplementing the old template data), and is updated to the long-term memory module.

[0069] Specifically, the short-term memory module stores the session context in real time, and abstractly compresses the context to extract key demand features. The short-term memory module quickly captures dynamic information of the dialogue, avoids context loss caused by long-term dependence, reduces redundant information through abstract compression, focuses on core demand features, improves subsequent retrieval efficiency, provides structured input for long-term memory retrieval, and reduces the probability of false matching;

[0070] The existing template in the long-term memory is preferentially searched, and if there is no match, a new template is dynamically generated and filled with user feedback. The existing template is directly called to reduce repeated calculation and accelerate response speed. For the fields not covered, the framework is generated and filled with user feedback to realize "cold start", expand the boundary of the knowledge base, and the user modification data in the short-term memory is used to fill the template, so that the generated content is more in line with the actual demand, and the individual service ability is improved;

[0071] The difference evaluation report is generated by comparing the content of the new and old templates in combination with the user feedback, and is fused and optimized according to the rules. The improvement points of the template (such as missing data dimension and logical vulnerability) are identified through comparison and analysis, version degradation is avoided, the preset fusion rule reduces manual intervention, the template is self-evolved, the user modification feedback is directly involved in the evaluation, and the optimization direction of the template is ensured to be consistent with the user demand.

[0072] In some embodiments, the white box knowledge is called to supplement the knowledge base, specifically including:

[0073] Based on the long-term memory module, the commonalities of high-quality templates and low-quality templates are extracted. The commonalities of high-quality templates include structural commonalities (document chapter order (e.g., "background-analysis-conclusion"), data presentation format (e.g., table / chart ratio)) and content commonalities (high-frequency data dimensions (e.g., "user age distribution" in market analysis), standard argumentation logic (e.g., "problem → cause → solution")). The commonalities of low-quality templates include structural defects (missing chapters (e.g., legal documents without a "legal basis" chapter)) and content defects (contradictions between data and conclusions (e.g., "user satisfaction declined but the conclusion is 'service optimization was successful'")).

[0074] Based on the common structural characteristics of high-quality templates, the types of white-box knowledge to be invoked and their embedding positions are determined. Based on the counterexample rules of low-quality templates, taboo conditions for knowledge matching are defined, thereby enabling the matching of structured knowledge items from the white-box knowledge base.

[0075] By embedding structured knowledge items into a general logical framework, a new template framework is generated.

[0076] Specifically, extract the structural commonalities of high-quality templates (such as the chapter order "background-analysis-conclusion") and standard argumentation logic ("problem → cause → solution"), and embed them into a structured framework using white-box knowledge to ensure that new templates follow logical rules. New templates automatically inherit the mature framework of high-quality templates to avoid logical confusion or missing chapters. For example, legal documents must include a "legal basis" chapter, which is forcibly embedded using white-box knowledge to prevent structural defects.

[0077] High-frequency data dimensions (such as user age distribution) of high-quality templates are extracted as mandatory content, while counterexample rules (such as data and conclusions) of low-quality templates define taboo conditions. New templates automatically include key data dimensions to avoid information omissions. Taboo conditions detect contradictions between data and conclusions (such as claiming "service success" despite declining user satisfaction) and force the correction of logical loopholes.

[0078] The long-term memory module continuously accumulates common rules of high-quality and low-quality templates. The white-box knowledge base is dynamically updated through taboo conditions and structured entries. The common rules of high-quality templates become standardized knowledge, and the counterexamples of low-quality templates are transformed into the basis for improvement. In new scenarios, white-box knowledge can be called to quickly generate compliant templates. For example, when a new "User Profile Analysis" chapter is added, data dimension rules are automatically matched.

[0079] Explicit rules of white-box knowledge (such as chapter order and data proportion) guide template generation, while taboo conditions filter low-quality content through explicit rules. Structured rules reduce trial and error costs and accelerate template generation.

[0080] The autonomous intelligent agent, referencing its memory information, performs in-depth requirements analysis on the requirement text, generating a detailed plan that includes model context protocol layer tools and operational steps, specifically:

[0081] The autonomous intelligent agent uses memory information and natural language processing technology to perform multi-level semantic analysis on the input requirement text and generate requirement feature vectors;

[0082] Based on the required feature vectors, select the model context protocol layer tools from the preset tool library;

[0083] Based on the model context protocol layer tools and requirement feature vectors, structured operation steps are generated.

[0084] Specifically, by using natural language processing technology to perform multi-level semantic analysis (such as syntactic, semantic, and contextual analysis) on the demand text, and combining it with historical experience in memory information to aid understanding, it can handle ambiguous, implicit, or cross-domain demand text, identify deep intentions (such as "optimizing user experience" can be associated with interface improvement, performance enhancement, etc.), and memory information provides historical context reference to avoid repetitive analysis or contradictory interpretations. For example, it can prioritize matching high-frequency tools based on users' historical behavior.

[0085] The semantic analysis results are transformed into requirement feature vectors, and tools are selected from the tool library by vector similarity. The vector form simplifies the expression of requirements and facilitates unified matching of cross-domain tools. The most suitable tool is automatically selected based on the feature vector (such as matching Python script tools for data analysis requirements), reducing the cost of manual selection. New tools only need to register their feature vectors for quick adaptation.

[0086] Structured operation steps are generated based on tool capabilities and requirements (e.g., "use API X to extract data → call model Y to analyze → generate report Z"). The steps are automatically arranged according to tool dependencies and requirements, avoiding human planning oversights. If a tool is unavailable, alternative tools can be re-matched based on feature vectors and the steps can be adjusted. The structured steps facilitate subsequent execution monitoring and problem troubleshooting.

[0087] Based on the planned operational steps, the autonomous agent determines the required Model Context Protocol (MTP) layer tools and obtains information directly related to the execution of the operational steps through these tools. In some embodiments, the MTP layer tools include:

[0088] Similar document retrieval tool, used to retrieve related documents from local database;

[0089] Internet search tools are used to perform internet searches and obtain results for ambiguous text in the relevant documents being recalled.

[0090] A draft generation tool that uses natural language generation technology to generate document drafts by combining relevant recalled documents and information obtained from internet searches.

[0091] Document writing tools are used to convert document drafts into Markdown format and to write documents.

[0092] A document format conversion tool used to convert Markdown documents into formats such as doc and ppt.

[0093] Specifically, similar document retrieval tools quickly locate relevant documents through local databases and use vector retrieval or semantic matching technology to ensure retrieval efficiency. Internet search tools obtain the latest external data in real time for ambiguous information to make up for the lack of timeliness of local knowledge bases. Combining local knowledge accumulation with dynamic Internet information avoids information blind spots from a single source. By cross-validating local documents and Internet results, the credibility of the content is improved (e.g., technical parameters need to be compared with official documents).

[0094] The draft generation tool integrates recall information based on NLP technology to generate a logically coherent first draft of text. The document writing tool converts the draft into a structured Markdown format to ensure clear content hierarchy. The NLP model generates targeted content based on context to reduce redundant information. The Markdown format unifies the content structure, making it easier for subsequent editing or conversion. Automatic segmentation, heading levels, and other structured processing optimize the reading experience.

[0095] The document format conversion tool supports conversion from Markdown to formats such as doc / ppt. The doc format is suitable for formal reports, while the ppt format is suitable for presentation scenarios, meeting different usage needs. The conversion tool must ensure that elements such as layout, formulas, and charts are transferred without loss of quality, avoiding manual secondary adjustments.

[0096] The autonomous intelligent agent assesses whether the currently acquired information meets the requirements for executing the plan. If it does, it generates the final document based on the information integration; otherwise, it adjusts the plan or supplements the information.

[0097] Specifically, the intelligent agent ensures that the final document is generated only when the execution requirements are met by evaluating the extent to which information covers the planned objectives in real time. If key data is not acquired (such as the lack of user profiles in market analysis), it actively supplements the data instead of directly outputting incomplete results. Through dynamic verification (such as the requirement that "comparative analysis" in the report requires at least two data sources), it enforces the consistency between the output and the planned objectives.

[0098] When the information does not meet the requirements, the agent can adjust the plan or trigger the information supplementation mechanism. For example, if there is no relevant data in the local database, it can automatically switch to Internet search tools and dynamically adjust the plan according to the difficulty of information acquisition (such as prioritizing easily accessible data and delaying the collection of complex information).

[0099] The system addresses information uncertainty through an "evaluation-adjustment" closed-loop mechanism. When the recalled document is irrelevant to the requirement, the system can expand the search scope or modify the query keywords. It can also periodically evaluate and intercept early errors (such as data format errors) to avoid wasting resources in subsequent processes.

[0100] Based on the evaluation results, a decision will be made on whether to continue investing resources in information integration. Documents will be generated immediately when the information meets the standards to reduce unnecessary processing steps. Tasks with missing key information will be prioritized for supplementation, while secondary tasks can be postponed or simplified.

[0101] A document generation system based on autonomous intelligent agents, comprising:

[0102] The user interaction module is used to obtain the user's input request text and send it to the autonomous intelligent agent;

[0103] The memory information processing module is used by autonomous intelligent agents to extract memory information related to the requirement text as knowledge reference based on the requirement text.

[0104] The requirements analysis and plan generation module is used by autonomous intelligent agents to perform in-depth requirements analysis on the requirements text with reference to the memory information, and generate detailed plan content including model context protocol layer tools and operation steps.

[0105] The tool invocation and information acquisition module is used to determine the model context protocol layer tools to be invoked based on the operation steps of the plan, and to obtain information directly related to the execution of the operation steps through the model context protocol layer tools.

[0106] The evaluation and document generation module is used by the autonomous intelligent agent to evaluate whether the information currently acquired meets the requirements for the execution of the plan. If it does, the final document is generated based on the information integration; otherwise, the plan content is adjusted or supplementary information is added.

[0107] In some embodiments, the memory information processing module includes:

[0108] The requirement parsing module is used to receive the requirement text input by the user and parse the domain, topic and task objectives in the requirement text using natural language processing technology.

[0109] The memory information retrieval module is used to retrieve historical document templates, case data, or general knowledge that match the domain, topic, and task objective of the required text from the memory module as memory information based on the parsing results.

[0110] The knowledge reference module is used to transmit the extracted memory information to the autonomous intelligent agent as a reference for demand analysis and plan generation.

[0111] The memory update and white-box knowledge supplementation module is used to dynamically generate new templates or call white-box knowledge supplementation modules if no directly matching memory information is found.

[0112] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of any of the above-described document generation methods based on autonomous intelligent agents.

[0113] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described document generation methods based on autonomous intelligent agents.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for document generation based on autonomous agents, characterized in that, Specifically comprising: Obtain the demand text input by the user and send it to the autonomous agent; The autonomous agent extracts memory information related to the demand text as knowledge reference based on the demand text; The autonomous agent takes the memory information as reference to conduct deep demand analysis on the demand text, and generates detailed plan content including model context protocol layer tools and operation steps; According to the operation steps of the plan content, the autonomous agent judges the model context protocol layer tools needed to be called, and obtains the information directly related to the operation step execution through the model context protocol layer tools; The autonomous agent evaluates whether the currently obtained information meets the execution requirements of the plan content, and if so, generates a final document based on the information integration, otherwise adjusts the plan content or supplements the information.

2. The autonomous agent-based document generation method of claim 1, wherein, The autonomous agent extracts memory information related to the demand text as knowledge reference based on the demand text, specifically comprising: Receive the demand text input by the user, and parse the domain, theme and task target in the demand text according to natural language processing technology; According to the analysis result, retrieve the historical document template, case data or general knowledge matching the domain, theme and task target of the demand text from the memory module as memory information; Pass the extracted memory information to the autonomous agent as the basis for knowledge reference of demand analysis and plan generation; If no directly matching memory information is retrieved, dynamically generate a new template or call white box knowledge to supplement the knowledge base through the memory update mechanism.

3. The autonomous agent-based document generation method of claim 2, wherein, The new template is dynamically generated through the memory update mechanism, specifically comprising: Based on the short-term memory module, real-time store the context information of the current session, and conduct abstract compression on the context to extract key demand features; According to the key demand features, search the long-term memory module for existing templates of the same domain and theme; if there is no existing template, generate a template framework based on the key demand features, and fill in the specific content combined with the user modification feedback in the short-term memory module, and finally store it in the long-term memory module; if there is an existing template, compare the content structure, data dimension and conclusion logic of the new and old templates, and analyze the advantages and disadvantages of the new and old templates combined with the user modification feedback in the short-term memory module to generate a difference evaluation report; According to the difference evaluation report, generate an optimized new template according to the preset fusion rule, and update it to the long-term memory module.

4. The autonomous agent-based document generation method of claim 3, wherein, The white box knowledge is called to supplement the knowledge base, specifically comprising: Based on the long-term memory module, extract the commonness of each high-quality template and the commonness of each low-quality template; According to the commonness of each high-quality template and the commonness of each low-quality template, match the structured knowledge items from the white box knowledge base; Embed the structured knowledge items into a general logic framework to generate a new template framework.

5. The autonomous agent-based document generation method of claim 4, wherein, The autonomous agent takes the memory information as reference to conduct deep demand analysis on the demand text, and generates detailed plan content including model context protocol layer tools and operation steps, specifically comprising: The autonomous agent uses memory information and natural language processing technology to conduct multi-level semantic analysis on the input demand text, and generates a demand feature vector; According to the demand feature vector, filter out the model context protocol layer tools from the preset tool library; Based on the model context protocol layer tool and the demand feature vector, a structured operation step is generated.

6. The autonomous agent-based document generation method of claim 5, wherein, The model context protocol layer tool comprises: A similar document recall tool for recalling relevant documents from a local database; An Internet search tool for performing Internet search on ambiguous text in the recalled relevant documents and obtaining results; A draft generation tool for generating a document draft by combining the recalled relevant documents and the information obtained through Internet search using natural language generation technology; A document writing tool for converting the document draft into a preset format and performing document writing; A document format conversion tool for converting the preset format document into a general document format.

7. An autonomous agent based document generation system, characterized in that, Comprise: A user interaction module for obtaining user input demand text and sending it to the autonomous agent; A memory information processing module for the autonomous agent to extract memory information related to the demand text as reference knowledge based on the demand text; A demand analysis and plan generation module for the autonomous agent to perform in-depth demand analysis on the demand text with the memory information as a reference, and generate detailed plan content including model context protocol layer tools and operation steps; A tool calling and information obtaining module for the autonomous agent to determine the model context protocol layer tool to be called according to the operation steps of the plan content, and obtain information directly related to the operation step execution through the model context protocol layer tool; An evaluation and document generation module for the autonomous agent to evaluate whether the current obtained information meets the execution requirements of the plan content, and if so, generate a final document based on the information integration, otherwise adjust the plan content or supplement information.

8. The autonomous agent-based document generation system of claim 7, wherein, The memory information processing module comprises: A demand analysis module for receiving user input demand text and analyzing the domain, theme and task target in the demand text according to natural language processing technology; A memory information retrieval module for retrieving historical document templates, case data or general knowledge matching the domain, theme and task target of the demand text from the memory module as memory information according to the analysis results; A knowledge reference module for transmitting the extracted memory information to the autonomous agent as a knowledge reference basis for demand analysis and plan generation; A memory update and white box knowledge supplement module for dynamically generating new templates or calling white box knowledge supplement knowledge base through memory update mechanism if no directly matching memory information is retrieved.

9. A computing device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the document generation method based on the autonomous agent in any one of claims 1-6 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The storage medium has a computer program stored thereon, and the computer program is executed by the processor to implement the steps of the document generation method based on the autonomous agent in any one of claims 1-6.

Images