Knowledge research system and research method, device, electronic equipment and storage medium thereof

Abstract

This application discloses a knowledge research system, its research method, apparatus, electronic device, and storage medium. The system is deployed on an electronic device and comprises a three-layer architecture: resources, agents, and flow control. The resource layer includes three types of databases: raw materials, knowledge, and projects. The agent layer configures planning, surveying, researching, and review agents, sharing underlying resources and equipped with independent, dedicated skill packages. The flow control layer standardizes the serial flow order of each agent. Relying on the serial collaboration of four types of agents, combined with paradigm-driven research method adaptation, multi-human interaction node control, a triple-source anti-counterfeiting citation mechanism, and automated quality inspection gating, the system transforms knowledge-intensive research into a streamlined, standardized, and traceable process. It solves problems such as existing technology knowledge silos, information gaps, disconnect between research methods and practical application, reliance on manual quality review, and lack of human intervention in multi-agent collaboration, achieving a balance between automated collaboration efficiency and the human controllability of key nodes.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a knowledge research system and its research methods, apparatus, electronic devices and storage media. Background Technology

[0002] In knowledge-intensive fields such as policy research, legal and regulatory research, and industry research, the complexity and specialization of research work require researchers to complete a series of steps, including data retrieval, logical analysis, manuscript writing, and quality review. With the development of large language model technology, multi-agent collaborative systems (such as AutoGen and CrewAI) have been applied to the automation of knowledge work, achieving a certain degree of automation in the research process through dialogue and task transfer between multiple agents.

[0003] However, in knowledge-intensive research scenarios, there is currently a lack of effective integration architecture between multi-agent collaborative processes and quality control mechanisms. Specifically, existing large language model multi-agent systems typically employ fully automated serial or parallel workflows, with the system autonomously deciding the direction of workflow. While this improves efficiency, it lacks mechanisms for introducing human judgment at key decision-making nodes. For high-value, high-risk knowledge output scenarios such as policy research and legal research, this fully automated workflow is prone to deviations in research plans due to agent illusions or deviations from expected directions. Furthermore, once deviations occur, they are difficult to detect and correct in a timely manner during the process.

[0004] On the other hand, while knowledge base augmented generation (RAG) technology enhances generation quality by retrieving external knowledge bases and can improve the effect of single-point generation, it lacks multi-agent collaborative process orchestration and cross-validation mechanisms, failing to solve the problems of information gaps and citation consistency in multi-stage collaboration. Traditional research management systems, although capable of allocating and tracking research tasks, lack AI-driven automated research, analysis, and writing capabilities, and essentially still rely on manual execution of core research stages.

[0005] Therefore, existing technologies are divided into two types: one is multi-agent fully automated collaboration that pursues automation efficiency but sacrifices controllability, and the other is traditional management systems that maintain controllability but lack automation capabilities. Both have their advantages and disadvantages, but neither can achieve knowledge-intensive research while maintaining human control and achieving high automation efficiency. Summary of the Invention

[0006] In view of this, this application provides a knowledge research system and its research method, apparatus, electronic device and storage medium, for achieving a balance between automated collaboration efficiency and human controllability of key nodes in a knowledge-intensive research process involving multi-agent collaboration.

[0007] To achieve the above objectives, the following solution is proposed:

[0008] A knowledge research system, applied to electronic devices, comprises a resource layer, an intelligent agent layer, and a flow control layer, wherein:

[0009] The resource layer is configured with a raw material library, a knowledge base, and a project library. The raw material library is used to store read-only raw documents, the knowledge base is used to store refined structured knowledge networks, and the project library is used to store process documents and result documents of research projects.

[0010] The intelligent agent layer is configured with a planning intelligent agent, a survey intelligent agent, a research intelligent agent, and a review intelligent agent. Each intelligent agent shares the resource layer and is configured with an independent exclusive skill package. The planning intelligent agent is used to receive research requirements and output research plans. The survey intelligent agent is used to perform material retrieval and original text extraction based on the research plan. The research intelligent agent is used to perform manuscript writing based on the retrieved materials and research plan. The review intelligent agent is used to perform quality review on the manuscript.

[0011] The flow control layer is used to define the serial flow order among the four intelligent agents. The flow control layer sets human interaction checkpoints after the planning intelligent agent outputs the research plan, after the research intelligent agent completes the first draft, and after the review intelligent agent issues review opinions, so as to wait for human confirmation before continuing the flow.

[0012] Optionally, the flow control layer also includes a verification checkpoint system, which is set at the transmission interface between adjacent agents to verify the integrity of the output data of the upstream agent.

[0013] Optionally, the system is also configured with a citation verification mechanism, which sets up independent source verification defenses at the survey agent, the research agent, and the review agent stages.

[0014] Optionally, the back-to-source verification defense includes:

[0015] The first line of defense is achieved by the survey agent locking the original text of the terms, reading the source file from the raw material library and extracting the original text of the terms word by word, and outputting the original text locking block;

[0016] The second line of defense involves the research agent performing source verification, independently reading the source files in the raw material library and comparing them word by word with the original text lock blocks, and marking the verification as passed in the manuscript;

[0017] The third line of defense involves the review agent performing independent source verification, extracting the citation list from the document, independently reading the source files in the raw material library for word-by-word comparison, and outputting a citation comparison table.

[0018] Optionally, the planning agent is configured with a paradigm recognition module and a multi-scheme output module, wherein:

[0019] The paradigm identification module is used to match a preset research paradigm classification system based on keywords in the research requirements, so as to determine the category to which the research belongs and match the corresponding soft science methodology framework.

[0020] The multi-solution output module is used to output one or more differentiated alternative research schemes based on the requirements complexity assessment results. Each scheme includes a scheme name, applicable scenario, core logic, and advantages and disadvantages analysis.

[0021] Optionally, the survey agent is configured with a paradigm-driven missing information identification module, wherein:

[0022] The missing data identification module is used to automatically check whether the required data types are complete according to the research paradigm, and output supplementary data instructions when there are data gaps.

[0023] Optionally, the review agent is configured with a methodology compliance review module, wherein:

[0024] The methodology compliance review module is used to check the correctness of research paradigm identification, the completeness of the soft science methodology framework, and the compatibility of analytical methods.

[0025] A research method, applied to the knowledge research system described above, the research method comprising the steps of:

[0026] In response to the user's project initiation request, the planning agent is controlled to receive research requirements and output a research plan;

[0027] In response to the user's confirmation of the research plan, the research agent is controlled to conduct cross-database searches in the raw material database, knowledge base, and project database according to the research plan, and to classify and organize the search results and accurately extract the original text.

[0028] The research agent is controlled to receive the source drafts from the survey agent and load the research paradigm corresponding to the research plan to write the manuscript.

[0029] The review agent is controlled to conduct a quality review of the document and issue review opinions.

[0030] In particular, after the planning agent outputs the research plan, after the research agent completes the first draft, and after the review agent issues its review opinions, there are human interaction checkpoints. These human interaction checkpoints are used to wait for user confirmation.

[0031] Optionally, controlling the planning agent to receive research requirements and output a research plan includes the following steps:

[0032] The planning agent is controlled to match a preset research paradigm classification system based on keywords in the research requirements, in order to determine the category to which the research belongs and match the corresponding soft science methodology framework.

[0033] Based on the requirements complexity assessment results, output one or more differentiated alternative research solutions. Each solution includes a solution name, applicable scenarios, core logic, and advantages and disadvantages analysis.

[0034] Optionally, after the survey agent performs a cross-database search according to the research plan, the process further includes the following steps:

[0035] The system automatically checks whether the required data types are complete based on the research paradigm, and outputs supplementary data instructions when there are data gaps.

[0036] Optionally, controlling the review agent to conduct a quality review of the manuscript and issue review opinions includes the following steps:

[0037] The control system checks the correctness of the research paradigm identification, the integrity of the soft science methodology framework, and the compatibility of the analytical methods.

[0038] The review agent is controlled to classify and label the problems found during the review, and the classification includes at least fatal errors, general flaws and optimization suggestions.

[0039] A research apparatus, applied to the knowledge research system described above, the research apparatus comprising:

[0040] The startup control module is configured to respond to the user's project startup request, control the planning agent to receive research requirements, and output a research plan;

[0041] The research control module is configured to respond to the user's confirmation of the research plan, control the research agent to conduct cross-database searches in the raw material database, knowledge base, and project database according to the research plan, and classify and organize the search results and accurately extract the original text.

[0042] The research control module is configured to control the research agent to receive the source draft of the survey agent and load the research paradigm corresponding to the research plan to write the manuscript.

[0043] The review control module is configured to control the review agent to conduct a quality review of the document and issue review opinions.

[0044] In particular, after the planning agent outputs the research plan, after the research agent completes the first draft, and after the review agent issues its review opinions, there are human interaction checkpoints. These human interaction checkpoints are used to wait for user confirmation.

[0045] An electronic device includes at least one processor and a memory connected to the processor, wherein:

[0046] The memory is used to store computer programs or instructions;

[0047] The processor is used to execute the computer program or instructions to enable the electronic device to implement the research method described above.

[0048] A computer-readable storage medium is applied to an electronic device, the storage medium carrying one or more computer programs that can be executed by the electronic device to enable the electronic device to perform the research method described above.

[0049] As can be seen from the above technical solution, this application discloses a knowledge research system and its research method, apparatus, electronic device, and storage medium. This system is applied to an electronic device and includes a resource layer, an agent layer, and a flow control layer. The resource layer is configured with a raw material library, a knowledge base, and a project library; the agent layer is configured with a planning agent, a survey agent, a research agent, and a review agent, each agent sharing the resource layer and configured with an independent, dedicated skill package; the flow control layer is used to define the serial flow order between the four agents. This system, through the serial collaboration of four specialized agents (planning, surveying, researching, and reviewing), combined with paradigm-driven knowledge research methodology adaptation, key node control with multiple human interaction checkpoints, a triple-source verification anti-counterfeiting mechanism, and automated quality gating of the verification checkpoint system, achieves the pipelined, standardized, and traceable nature of knowledge-intensive research tasks. It effectively solves the problems of knowledge silos and information gaps, disconnect between research methods and execution, reliance on human review for quality control, and lack of human interaction nodes in multi-agent collaboration in existing technologies, achieving a balance between automated collaboration efficiency and the human controllability of key nodes. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a schematic diagram of a knowledge research system according to an embodiment of this application;

[0052] Figure 2 This is a flowchart illustrating a research method according to an embodiment of this application;

[0053] Figure 3This is a flowchart of a research apparatus according to an embodiment of this application;

[0054] Figure 4 This is a block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0056] Figure 1 This is a schematic diagram of a knowledge research system according to an embodiment of this application.

[0057] like Figure 1 As shown, the multi-agent collaborative knowledge research system provided in this embodiment includes a resource layer 100, an agent layer 200, and a flow control layer 300. The resource pool, agent layer, and flow control layer cooperate with each other to complete knowledge research.

[0058] The resource layer 100 of this application is configured with at least three resource libraries, namely the raw material library, the knowledge library, and the project library.

[0059] The raw material repository stores read-only original documents, including original policy and regulatory texts, original research reports, case data, etc. Documents in the raw material repository are read-only; no intelligent agent may modify the content of the raw material repository, ensuring the authenticity and immutability of the cited sources.

[0060] The knowledge base stores a structured knowledge network extracted from a large language model, including core concept definitions, organizational information, typical cases, and comprehensive analysis. The knowledge base is organized using a systematic directory structure and supports cross-document referencing.

[0061] The project repository is used to store process and outcome documents for thematic research projects. Each research topic includes an "info" directory and a "result" directory in the project repository. The "info" directory stores process reference documents, discussion records, and meeting minutes; the "result" directory stores final and interim results.

[0062] The agent layer 200 is configured with four functionally differentiated agents. Each agent shares a resource layer and has its own independent skill set and role-specific prompt (System Prompt). The four agents are the planning agent, the research agent, the investigation agent, and the review agent.

[0063] The planning agent acts as the system's overall scheduler, and its core functions include: paradigm identification, task decomposition, process orchestration, framework matching, multiple solution output, and task distribution. The planning agent is equipped with a paradigm identification module and a multiple solution output module.

[0064] The paradigm identification module matches keywords in the research requirements with a pre-defined research paradigm classification system to determine the research category and match the corresponding soft science methodology framework. When the planning agent is working, it first performs paradigm identification on the received research requirements. The paradigm identification module matches keywords in the requirements with five pre-defined research paradigm classification systems: ICT industry research paradigm, policy research paradigm, legal and regulatory research paradigm, economic research paradigm, and comprehensive research paradigm. Each paradigm is associated with a corresponding soft science methodology framework and analytical method matrix.

[0065] After paradigm identification, the multi-solution output module determines the number of output solutions based on the complexity assessment results. The complexity assessment is based on the following dimensions: number of domains, number of stakeholders, path certainty, and output type. Assessment results are categorized into simple and complex. For simple requirements, the multi-solution output module outputs one recommended solution; for complex requirements, it outputs two to three differentiated alternative solutions. These solutions are fundamentally different, reflecting different analytical paths, framework focuses, or output directions. Each solution includes a solution name, applicable scenario, core logic, advantages and disadvantages analysis, and a recommendation index.

[0066] After outputting a solution, the planning agent pauses its workflow and waits for human confirmation at the first interaction checkpoint. Human confirmation offers options: confirm the solution, switch solutions, or reject and revise. Only after human confirmation does the planning agent continue generating standardized task orders and distributing them to downstream survey agents.

[0067] The research agent performs data retrieval and information collection functions. Its core skills include: full-text search, material classification and organization, accurate extraction of original text, identification of missing content, online search supplementation, and locking of original clause text. The research agent is also equipped with a paradigm-driven missing content identification module.

[0068] After receiving the task order, the research agent first conducts a cross-database joint search in the raw material database, knowledge base, and project database according to the research plan. The search results are then structured and organized according to material categories, with classification systems including: policies and regulations, systems and processes, typical cases, standards and definitions, and data statistics.

[0069] The missing data identification module automatically checks whether the required data types for the research paradigm are complete, and outputs supplementary information when data gaps exist. Specifically, the missing data identification module checks one by one whether the retrieved material categories cover all the data types listed in the template, based on the data requirement template corresponding to the research paradigm. For example, the data requirement template for the legal and regulatory research paradigm includes: original legal texts, judicial interpretations, typical cases, implementation rules, policy interpretations, etc.; the data requirement template for the ICT industry research paradigm includes: industry statistics, policy documents, enterprise cases, technical standards, market reports, etc. If a certain type of required data is found to be missing, supplementary information is generated and fed back to the planning agent or human. The supplementary information includes: the type of missing data, the reason for the missing data, and suggested acquisition methods, such as online search, manual supplementation, etc. After the relevant materials are supplemented manually, the research agent continues to perform accurate extraction of the original text.

[0070] When the "Citation Validation Requirement" field of the task sheet is "Yes", the research agent must activate the original text locking mechanism, i.e., the three-step locking method. First, locate the official original text of the target regulation in the raw material library and confirm that the file is the latest version. Second, use the Read tool to read the target clause in the source file, copying the complete original text character by character without any rewriting, abbreviation, or merging. Third, output the original text locked block, in the following format:

[0071] Research agents are the core content production units, and their core skills include: application of the four-step meta-framework of soft science, logical analysis and deduction, document writing, and content integration.

[0072] The research agent first loads the "Research Paradigm" field from the task list, determines the paradigm it belongs to, and initiates the corresponding soft science methodology framework. All research manuscripts are written according to the four-step meta-framework of soft science, including problem identification, system construction, layout, and methodology. Specifically, the problem identification stage requires clearly defining the research question, research boundaries, and core concepts; the system construction stage requires building an analytical framework and theoretical system; the layout stage requires designing a reasonable structural layout and argumentation logic; and the methodology stage requires selecting appropriate analytical methods and tools.

[0073] When the citation verification requirement in the task sheet is "yes," the research agent must initiate the source verification mechanism and perform a three-step verification process for each regulatory citation in the research draft: independently read the source file in the raw material library → compare the original text locked block output by the research agent word by word → mark each regulatory citation in the draft. The coverage of citation verification marks must reach 100%.

[0074] After the research agent completes the initial draft, it conducts a four-dimensional self-check to confirm that the citation verification mark coverage reaches 100%, including the logic dimension, content dimension, compliance dimension, and citation dimension. After passing the self-check, the research agent pauses the workflow and submits the manuscript for manual pre-review through the second interaction checkpoint.

[0075] The review agent serves as the final checkpoint for risk control and quality inspection. Its core skills include: framework compliance verification, rule comparison and review, risk identification, and issue labeling and rectification suggestions. The review agent is equipped with a methodology compliance review module.

[0076] The review agent examines the initial draft item by item for the following: framework fit and methodological compliance, rule consistency, terminology standardization, logical coherence, and formatting specifications. The methodological compliance review module performs the following three checks:

[0077] 1. Correctness Check of Research Paradigm Identification. The methodology compliance review module first verifies whether the research paradigm used in the initial draft is consistent with the paradigm initially determined by the planning agent. The review agent independently extracts key domain terms from the requirement text, re-determines the paradigm category to which the requirement belongs, and compares the re-determined result with the paradigm output by the planning agent. If they are consistent, the paradigm identification is considered correct; if they are inconsistent, it is marked as an issue of the corresponding level, and the deviation is noted in the review comment form.

[0078] 2. Completeness Check of the Methodological Framework of Soft Science. The review agent checks whether the manuscript fully covers the four-step meta-framework of soft science corresponding to its paradigm (problem identification → system → layout → method). The review agent verifies one by one whether the manuscript contains the corresponding chapters or content modules of the four-step framework: whether the research question is clearly defined (problem identification), whether an analytical system is constructed (system), whether a reasonable structural layout is designed (layout), and whether matching analytical methods are selected (method). If there are any missing or obviously weak links, they are marked as problems of the corresponding level.

[0079] 3. Methodological Matching Check. Review whether the analytical methods actually used in the agent verification document match the research question. For example, for a policy research paradigm, review whether appropriate policy analysis methods are used; for a legal research paradigm, review whether correct legal interpretation and application methods are used. If the analytical methods do not match the paradigm, mark it as a problem of the corresponding level and provide modification suggestions in the review comments form.

[0080] When the "Citation Verification Requirement" in the task order is "Yes", the review agent must initiate an independent source verification mechanism, executing a four-step verification process: extracting the citation list from the document → independently reading the source files in the raw material library → comparing the cited content in the document with the original text of the source file word by word → outputting a citation comparison table. The citation comparison table includes: citation location, regulation name, cited clause content, original text of the source file, whether the comparison result is consistent, and the issue level.

[0081] The review process categorizes issues into three levels: fatal errors, minor flaws, and suggestions for improvement. Fatal errors include content errors, rule violations, compliance risks, and serious methodological deviations; minor flaws include inconsistent formatting, improper citations, and imprecise wording; suggestions for improvement include structural reorganization, supplementary explanations, and enhanced readability. Based on the comparison results, any inconsistencies in citations are marked as fatal errors and cannot be downgraded.

[0082] After the reviewing agent issues a review opinion form, the process is suspended, and the process waits for manual confirmation of the rectification method through a third interactive checkpoint.

[0083] The workflow control layer 300 defines the serial workflow order and collaborative communication method among the four agents, namely, the serial workflow order of planning agent → research agent → study agent → review agent. After the planning agent outputs the research plan, the research agent completes the initial draft, and the review agent provides its review comments, the workflow control layer sets up human interaction checkpoints to await human confirmation before continuing the workflow.

[0084] Specifically, the flow control layer includes three manual interaction checkpoints, namely the first interaction checkpoint, the second interaction checkpoint, and the third interaction checkpoint.

[0085] The first interaction checkpoint is set after the planning agent outputs a solution, i.e., between the planning agent and the survey agent. After the solution is output, the process pauses, awaiting human confirmation. Human confirmation allows for the following options: confirm the solution, switch solutions, or reject and revise. Only after human confirmation does the planning agent continue generating standardized task orders and distributing them to the survey agent.

[0086] The second interaction checkpoint is set after the research agent completes its initial draft self-check, between the research agent and the review agent. Once the initial draft passes the self-check, the process pauses, awaiting manual pre-review. The reviewer verifies the four dimensions of the draft and makes a decision to approve or reject it. Upon approval, the initial draft is officially transferred to the review agent.

[0087] The third interaction point is set after the review agent issues a review opinion form, i.e., between the review agent and the archived deliverables. After the opinion form is issued, the workflow is paused, awaiting human confirmation of the rectification method. Humans can choose from four rectification decisions: full rectification, selective rectification, rejection and rewriting, or retention of the original text. In response to the rectification confirmation, the review agent executes the rectification according to the human instruction or directly archives the deliverables.

[0088] The flow control layer of this application is also configured with a verification checkpoint system. The verification checkpoint system is set at the transmission interface between adjacent intelligent agents to verify the integrity of the output data of the upstream intelligent agent. Specifically, it includes a first verification checkpoint, a second verification checkpoint, and a third verification checkpoint.

[0089] The first verification checkpoint is set at the transmission interface between the survey agent and the research agent to check the integrity of the original text lock block in the survey working paper. The trigger condition is: when the output file of the survey agent contains cited content, and the "Citation Verification Requirement" field of the task sheet is "Yes". The passing standard is: each clause involving legal citations has a corresponding original text lock block in the survey working paper. If it fails, the survey agent is returned to supplement the original text lock block.

[0090] The second verification checkpoint is set at the transmission interface between the research agent and the review agent to check the coverage of citation verification markers in the document. The trigger condition is: when the output document of the research agent contains cited content, and the "Citation Verification Requirement" field of the task sheet is "Yes". The passing standard is: each regulatory citation is marked with a source confirmation marker, and the coverage is 100%. If it fails, the research agent is returned to supplement the verification markers.

[0091] The third verification checkpoint is set at the transmission interface between the review agent and the results archive, and is used to check the number of fatal errors in the citation comparison table. The trigger condition is: when the review agent completes independent back-to-source verification and outputs the citation comparison table. The passing criterion is: the number of fatal errors in the citation comparison table is 0. If it fails, the research agent is returned to modify the corresponding citation content.

[0092] Through the above three verification checkpoints, a quality control link is formed, which involves identifying problems, locating the intelligent agent, and making closed-loop corrections, ensuring that only research results that pass all verification checkpoints can be archived.

[0093] The knowledge research system of this application is also equipped with a citation verification mechanism. The citation verification mechanism sets up independent back-to-source verification defenses in the survey agent, research agent and review agent stages, specifically including the first defense, the second defense and the third defense.

[0094] The first line of defense involves the research agent locking the original text of the cited clauses. Upon receiving the task order, the research agent checks the "Citation Verification Requirement" field. If it is "Yes," it locates the list of relevant regulatory documents specified in the task order. For each regulatory citation, the research agent finds the corresponding official original text document in the source material library, prohibiting the substitution of the original text with secondary citations from model memory, summaries, or web search results. It uses the Read tool to read the target clause in the source file, copying the complete original text character by character. It then reads the source file from the source material library and extracts the original text of the clause word by word, outputting the original text lock block. The original text lock block for each clause includes the regulatory name, cited clause number, original text content, source path, and lock flag.

[0095] The second line of defense involves source verification performed by the research agent. During the drafting process, the research agent independently uses the Read tool to read the same source file from the raw material library for each regulatory citation in the research draft. It independently reads the source file in the library and compares it word-by-word with the original text's locked blocks. After confirming accuracy, it marks the citation location in the manuscript with a source verification confirmation mark and a verification pass mark. Once the entire manuscript is completed, it checks each citation verification mark for omissions, ensuring 100% citation verification mark coverage.

[0096] The third line of defense involves independent source verification performed by the review agent. Upon receiving the initial draft, the review agent extracts all cited content from the document, generates a citation list, independently reads the source files from the source material library, and for each citation in the citation list, independently uses the Read tool to read the corresponding source file in the source material library, comparing the cited content in the document with the original text of the source file word by word, and outputs a citation comparison table. The citation comparison table includes: citation location, regulation name, cited clause content, original text of the source file, comparison result, and issue level. If inconsistencies are found based on the comparison results, they are marked as fatal errors and cannot be downgraded. Only when the number of fatal errors in the citation comparison table is zero can the review agent issue a final approval conclusion.

[0097] As can be seen from the above technical solution, this embodiment provides a knowledge research system applied to electronic devices, including a resource layer, an agent layer, and a flow control layer. The resource layer is configured with a raw material library, a knowledge base, and a project library; the agent layer is configured with a planning agent, a survey agent, a research agent, and a review agent, each agent sharing the resource layer and configured with an independent, dedicated skill package; the flow control layer is used to define the serial flow order between the four agents. This system, through the serial collaboration of four specialized agents (planning, surveying, researching, and reviewing), combined with paradigm-driven knowledge research methodology adaptation, key node control with multiple human interaction checkpoints, a triple-source verification anti-counterfeiting mechanism, and automated quality gating of the verification checkpoint system, achieves the pipelined, standardized, and traceable nature of knowledge-intensive research tasks. It effectively solves the problems of knowledge silos and information gaps, disconnect between research methods and execution, reliance on manual review for quality control, and lack of human interaction nodes in multi-agent collaboration in existing technologies, achieving a balance between automated collaboration efficiency and the human controllability of key nodes.

[0098] Figure 2 This is a flowchart illustrating a research method according to an embodiment of this application.

[0099] like Figure 2 As shown, the research method provided in this embodiment is applied to the knowledge research system described above. The research method disclosed in this embodiment includes the following steps:

[0100] S1. In response to the user's project initiation request, the control planning agent receives research requirements and outputs a research plan.

[0101] Users initiate a project launch request by inputting their research requirements into the system. Research requirements can be a description in natural language, such as "the current status and optimization suggestions for the implementation of the cross-border data security assessment system" or "analyzing the differences in the framework of domestic and international artificial intelligence regulatory regulations."

[0102] When outputting the research plan, the planning agent performs the following operations:

[0103] First, the planning agent matches the pre-defined research paradigm classification system with the keywords in the research requirements to determine the category of the research and match the corresponding soft science methodology framework.

[0104] Specifically, the paradigm identification module of the planning agent extracts key domain terms from the demand text and matches them with a pre-defined classification system of five research paradigms: ICT industry research paradigm, policy research paradigm, legal and regulatory research paradigm, economic research paradigm, and comprehensive research paradigm. Each paradigm is associated with a corresponding soft science methodology framework and analytical method matrix.

[0105] Matching rules include, for example: if the requirement contains keywords related to laws and regulations, such as legislation, regulations, system design, and appeal handling methods, it is determined to be a legal research paradigm; if the requirement contains keywords related to the development of the ICT industry, such as telecommunications, the Internet, 5G, and artificial intelligence, it is determined to be an ICT industry research paradigm; if the requirement spans multiple fields, it is determined to be a comprehensive research paradigm.

[0106] Then, based on the requirements complexity assessment results, one or more differentiated alternative research solutions are output. Each solution includes a solution name, applicable scenarios, core logic, and advantages and disadvantages analysis.

[0107] Complexity assessment is based on the following dimensions: number of domains, number of stakeholders, path certainty, and output type. Assessment results are categorized into simple and complex types.

[0108] For simple needs, such as those in a single domain with a clear path, the planning agent outputs a recommended solution. For complex needs, such as those involving multiple domains, multiple stakeholders, or uncertain paths, the planning agent outputs two or three differentiated alternative solutions. Each solution includes a solution name, applicable scenario, core logic, advantages and disadvantages analysis, and a recommendation index. These solutions are fundamentally different, reflecting different analytical paths, framework focuses, or output directions.

[0109] After the planning agent outputs a solution, the workflow control layer pauses the workflow at the first human interaction checkpoint, awaiting the user's decision to confirm, switch, or reject the research solution. The user can choose to: confirm the solution, switch the solution, or reject and revise it.

[0110] S2. In response to the user's confirmation of the research plan, control the research agent to conduct cross-database searches based on the research plan, and classify and organize the search results into materials and accurately extract the original text.

[0111] After the user confirms the research plan at the first interaction checkpoint, the planning agent generates a standardized task sheet and distributes it to the survey agent. The task sheet includes the following fields: research topic, research paradigm, research methodology framework, citation and verification requirements, list of relevant regulatory documents, and description of expected deliverables.

[0112] The research agent performs the following operations based on the task list:

[0113] First, a cross-database joint search is conducted within the raw material database, knowledge base, and project database. The search covers all three major knowledge resource databases, ensuring that the information obtained includes original texts, interpretations, and process references, thus avoiding bias caused by a single information source.

[0114] Next, the search results are categorized and organized. The categorization system includes: policies and regulations, systems and processes, typical cases, standards and definitions, and data statistics.

[0115] Next, precise extraction of the original text is performed. For the relevant regulatory clauses, the research agent finds the official original text of the target regulation from the raw material library, confirms that the document is the latest version, uses the Read tool to read the target clause in the source file, and copies the complete original text character by character without any rewriting, abbreviation or merging.

[0116] Furthermore, after cross-database retrieval, the missing data identification module of the research agent automatically checks whether the required data types are complete according to the research paradigm, and outputs supplementary information when data gaps exist. Specifically, the missing data identification module checks one by one whether the retrieved material categories cover all the data types listed in the template, based on the data requirement template corresponding to the research paradigm. For example, the data requirement template for the legal and regulatory research paradigm includes: original legal texts, judicial interpretations, typical cases, implementation rules, policy interpretations, etc.; the data requirement template for the ICT industry research paradigm includes: industry statistics, policy documents, company cases, technical standards, market reports, etc. If a certain type of required data is found to be missing, supplementary information is generated, including the missing data type, the reason for the missing data, and suggested acquisition methods, and is fed back to the planning agent or presented directly to the user. After the user supplements the relevant materials, the research agent continues to perform accurate extraction of the original text.

[0117] When the "Citation Validation Requirement" field of the task sheet is "Yes", the survey agent must activate the original text locking mechanism (three-step locking method) and perform the above-mentioned original text locking operation.

[0118] After the research agent completes the data collection and organization, it outputs the research draft. The research draft includes: a categorized and organized collection of data, a locked block of the original text, and research instructions.

[0119] The first verification checkpoint in the workflow control layer is set at the transmission interface between the survey agent and the research agent, automatically triggering a check to verify the integrity of the original text locking block in the survey worksheet. The trigger condition is: when the survey agent's output file contains cited content, and the "Citation Verification Requirement" field in the task sheet is "Yes". The passing standard is: each clause involving legal citations has a corresponding original text locking block in the survey worksheet. If the verification fails, the system blocks the workflow and returns the task to the survey agent to supplement the original text locking block.

[0120] S3. Control the research agent to receive the source material drafts from the survey agent, load the research paradigm corresponding to the research plan, and write the manuscript.

[0121] After receiving the research manuscript, the research agent first loads the "Research Paradigm" field in the task list to determine the paradigm and initiates the corresponding soft science methodology framework. All research manuscripts follow the four-step meta-framework of soft science: problem-solving → system construction → layout → methodology writing. Problem-solving stage: clearly defining the research question, research boundaries, and core concepts; System construction stage: constructing the analytical framework and theoretical system; Layout stage: designing a reasonable structural layout and argumentation logic; Methodology stage: selecting appropriate analytical methods and tools.

[0122] Based on the categorized materials in the survey drafts, the research agent, combined with the corresponding soft science methodology framework, conducted logical analysis and deduction, content integration, and manuscript writing. During the writing process, the research agent maintained item-by-item coverage of the survey drafts to ensure that no key materials were omitted.

[0123] When the "Citation Verification Requirement" on the task sheet is "Yes", the research agent must initiate the source verification mechanism and perform a three-step verification method for each regulatory citation in the research draft: independently read the source file in the raw material library → compare the original text lock block output by the research agent word by word → mark each regulatory citation in the draft with a source verification mark. The coverage of citation verification marks must reach 100%.

[0124] After the research agent completes the initial draft, it performs a self-check across four dimensions: logic, content, compliance, and citation. The self-check confirms that the citation verification mark coverage reaches 100%. Once the self-check is passed, the research agent pauses the workflow and submits the draft to the user for pre-review via a second human interaction checkpoint.

[0125] The second human interaction checkpoint is set after the research agent completes the initial draft, specifically between the research agent and the review agent. The user verifies the four-dimensional content and makes a decision to approve or reject the draft. Once the user confirms approval, the initial draft is officially transferred to the review agent.

[0126] The second verification checkpoint in the workflow control layer is set at the transmission interface between the research agent and the review agent, automatically triggering the verification of the coverage of citation verification tags in the document. The trigger condition is: when the research agent's output document contains cited content, and the "Citation Verification Requirement" field in the task sheet is "Yes". The passing standard is: each regulatory citation is marked with a source confirmation tag, and the coverage rate is 100%. If the verification fails, the system blocks the workflow and returns the document to the research agent to supplement the verification tags.

[0127] S4. The control review agent conducts quality reviews of the manuscript and issues review comments.

[0128] After receiving the initial draft, the review agent performs the following quality review operations:

[0129] 1. Framework Compliance and Content Review: The reviewing agent reviews the following aspects of the initial draft: framework matching and methodology compliance, rule consistency, terminology standards, logical coherence, and formatting specifications.

[0130] 2. Methodological Compliance Review: The methodology compliance review module of the review agent controls the review agent to check the correctness of research paradigm identification, the completeness of the soft science methodology framework, and the matching of analytical methods, specifically including:

[0131] 1) Correctness check of paradigm identification: The review agent independently extracts key domain terms from the requirement text, re-determines the paradigm category to which the requirement belongs, and compares the re-determined result with the paradigm initially determined by the planning agent. If they match, the paradigm identification is correct; if they do not match, it is marked as a problem and the deviation is noted in the review opinion form.

[0132] 2) Completeness check of the soft science methodology framework: The review agent verifies one by one whether the manuscript contains the corresponding chapters or content modules of the four-step framework—whether the research question is clearly defined (problem-solving), whether an analytical system is constructed (system), whether a reasonable structural layout is designed (layout), and whether matching analytical methods are selected. If there are any missing or obviously weak links, they are marked as problems of the corresponding level.

[0133] 3) Methodological Matching Check: Review whether the analytical methods actually used in the agent verification document match the research question. For example, for a policy research paradigm, review whether appropriate policy analysis methods were used; for a legal research paradigm, review whether correct legal interpretation and application methods were used. If the analytical methods do not match the paradigm, mark it as an issue of the corresponding level and provide modification suggestions in the review comments form.

[0134] The review also includes a focus on identifying compliance risks, such as erroneous processes, non-compliant statements, and fabricated content.

[0135] 3. Problem Classification and Labeling: The reviewing agent classifies and labels the problems it discovers, with classifications including at least fatal errors, minor flaws, and optimization suggestions. Specifically, the reviewing agent distinguishes between three problem levels:

[0136] 1) Fatal Errors: Content errors, rule violations, compliance risks, and serious methodological deviations. Examples include inconsistencies found through citation comparisons, completely incorrect paradigm identification, and substantial deviations between the document content and the original regulatory text. Fatal errors must not be downgraded.

[0137] 2) Common flaws: inconsistent formatting, non-standard citations, and inaccurate descriptions.

[0138] 3) Optimization suggestions: Restructuring, adding explanations, and improving readability.

[0139] 4. Triple Source Verification: When the "Citation Verification Requirement" in the task order is "Yes," the review agent must initiate an independent source verification mechanism, executing a four-step verification process: extracting the citation list from the document → independently reading the source files in the raw material library → comparing the citations in the document with the original text of the source files word by word → outputting a citation comparison table. The citation comparison table includes: citation location, regulation name, cited clause content, original text of the source file, comparison result, and issue level. Based on the comparison results, if inconsistencies are found, it is marked as a fatal error and cannot be downgraded.

[0140] 5. Issuance of Review Comments: The reviewing agent summarizes the above review results and issues a review comments form. The review comments form includes: a review overview, a list of issues found, rectification suggestions, and a comparison table of references.

[0141] After the reviewing agent issues a review opinion form, it waits for the user to confirm the rectification method through a third human interaction checkpoint.

[0142] The third human interaction point is set after the review agent issues its review comments, i.e., between the review agent and the archived results. Users can choose from four rectification options: full rectification, selective rectification, rejection and rewriting, or retention of the original text. Upon user confirmation of rectification completion, the review agent executes the rectification according to the user's instructions or directly archives the results.

[0143] The third verification checkpoint in the workflow control layer is set at the transmission interface between the review agent and the output archive. It automatically triggers the verification of the fatal error count in the citation comparison table. The triggering condition is: the review agent completes independent back-to-source verification and outputs the citation comparison table. The passing standard is: the number of fatal errors in the citation comparison table is 0. If the verification fails, the system blocks the archiving and returns the case to the research agent to modify the corresponding citation content. Only when the number of fatal errors in the citation comparison table is zero can the review agent issue a final approval conclusion and allow the output archive to proceed.

[0144] In this embodiment, the reference verification mechanism configured by the system sets up independent back-to-source verification defenses in the stages of surveying agents, researching agents, and reviewing agents. Specifically, it includes a first defense, a second defense, and a third defense. The details have been described in detail above and will not be repeated here.

[0145] In this application, the flow control layer also includes a verification checkpoint system to form a quality control link of problem discovery → agent localization → closed-loop correction. The verification checkpoint system includes a first verification checkpoint, a second verification checkpoint, and a third verification checkpoint, the details of which have been described in detail above and will not be repeated here.

[0146] As can be seen from the above technical solution, this embodiment provides a research method applied to the knowledge research system mentioned above. Specifically, in response to a user's project initiation request, the planning agent receives research requirements and outputs a research plan; in response to the user's confirmation of the research plan, the research agent performs cross-database searches in the raw material database, knowledge base, and project database based on the research plan, and categorizes and organizes the search results, accurately extracting the original text; the research agent receives the draft materials from the research agent, loads the research paradigm corresponding to the research plan, and writes the manuscript; the review agent conducts a quality review of the manuscript and issues review opinions; wherein, after the planning agent outputs the research plan, after the research agent completes the first draft, and after the review agent issues review opinions, there are manual interaction checkpoints set up to wait for user confirmation. This method achieves the pipelined, standardized, and traceable nature of knowledge-intensive research tasks through the serial collaboration of four specialized intelligent agents: planning, surveying, researching, and reviewing. It combines paradigm-driven knowledge research methodology adaptation, key node control with multiple human interaction checkpoints, a triple-source verification anti-counterfeiting mechanism, and automated quality gating of the verification checkpoint system. It effectively solves the problems of knowledge silos and information gaps, disconnect between research methods and execution, reliance on human review for quality control, and lack of human interaction nodes in multi-agent collaboration in existing technologies. It achieves a balance between the efficiency of automated collaboration and the human controllability of key nodes.

[0147] Although the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous.

[0148] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0149] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer.

[0150] Figure 3 This is a block diagram of a research apparatus according to an embodiment of this application.

[0151] like Figure 3 As shown, the research apparatus provided in this embodiment is applied to the knowledge research system described above, and specifically includes a startup control module 10, a survey control module 20, a research control module 30, and a review control module 40.

[0152] The startup control module is used to respond to the user's project startup request, control the planning agent to receive research requirements, and output research solutions.

[0153] Users initiate a project launch request by inputting their research requirements into the system. These requirements can be a description in natural language, such as "the current status and optimization suggestions for the implementation of the cross-border data security assessment system" or "analyzing the differences in the framework of domestic and international artificial intelligence regulatory regulations." The specific launch process has been detailed above and will not be repeated here.

[0154] The survey control module is used to respond to the user's confirmation of the research plan, control the survey agent to conduct cross-database searches based on the research plan, and classify and organize the search results and accurately extract the original text.

[0155] After the user confirms the research plan at the first interaction checkpoint, the planning agent generates a standardized task sheet and distributes it to the survey agent. The task sheet includes the following fields: research topic, research paradigm, research methodology framework, citation verification requirements (yes / no), list of relevant regulatory documents (if any), and description of expected deliverables. The specific execution process has been described in detail above and will not be repeated here.

[0156] The research control module is used to control the research agent to receive the source material drafts from the survey agent and load the research paradigm corresponding to the research plan for manuscript writing.

[0157] After receiving the research manuscript, the research agent first loads the "Research Paradigm" field in the task list to determine the paradigm and initiates the corresponding soft science methodology framework. All research manuscripts follow the four-step meta-framework of soft science: problem identification → system construction → layout → methodology writing. Problem identification stage: clearly defining the research question, research boundaries, and core concepts; System construction stage: constructing the analytical framework and theoretical system; Layout stage: designing a reasonable structural layout and argumentation logic; Methodology stage: selecting appropriate analytical methods and tools. The specific process has been detailed above and will not be repeated here.

[0158] The control review agent conducts quality reviews of the manuscript and issues review comments.

[0159] After receiving the initial draft, the review agent performs a quality review, the specific process of which has been described in detail above and will not be repeated here. As can be seen from the above technical solution, this embodiment provides a research device applied to the knowledge research system described above. Specifically, in response to a user's project initiation request, the planning agent receives research requirements and outputs a research plan; in response to the user's confirmation of the research plan, the research agent performs cross-database searches in the raw material database, knowledge base, and project database based on the research plan, classifying and organizing the search results and accurately extracting the original text; the research agent receives the source material draft from the research agent, loads the research paradigm corresponding to the research plan, and writes the manuscript; the review agent performs a quality review of the manuscript and issues review opinions. Furthermore, after the planning agent outputs the research plan, after the research agent completes the initial draft, and after the review agent issues its review opinions, there are manual interaction checkpoints for waiting for user confirmation. This method achieves the pipelined, standardized, and traceable nature of knowledge-intensive research tasks through the serial collaboration of four specialized intelligent agents: planning, surveying, researching, and reviewing. It combines paradigm-driven knowledge research methodology adaptation, key node control with multiple human interaction checkpoints, a triple-source verification anti-counterfeiting mechanism, and automated quality gating of the verification checkpoint system. It effectively solves the problems of knowledge silos and information gaps, disconnect between research methods and execution, reliance on human review for quality control, and lack of human interaction nodes in multi-agent collaboration in existing technologies. It achieves a balance between the efficiency of automated collaboration and the human controllability of key nodes.

[0160] The units described in the embodiments of this disclosure can be implemented in software or in hardware. The name of a unit does not necessarily limit the unit itself; for example, the first acquisition unit can also be described as "a unit that acquires at least two Internet Protocol addresses".

[0161] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0162] Figure 4 This is a block diagram of an electronic device according to an embodiment of this application.

[0163] The following is for reference. Figure 4 This document illustrates a structural diagram suitable for implementing the electronic device in the embodiments of this disclosure. The terminal device in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This electronic device is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this disclosure.

[0164] The electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from an input device 406 into a random access memory (RAM) 403. The RAM also stores various programs and data required for the operation of the electronic device. The processing unit, ROM, and RAM are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0165] Typically, the following devices can be connected to the I / O interface: input devices including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although electronic devices with various devices are shown in the figures, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0166] This application also provides an embodiment of a computer-readable storage medium.

[0167] The aforementioned computer-readable storage medium is applied to an electronic device and carries one or more computer programs. When the electronic device executes these programs, it causes the device to respond to a user's project initiation request by controlling the planning agent to receive research requirements and output a research plan; responding to the user's confirmation of the research plan by controlling the research agent to conduct cross-database searches in the raw material database, knowledge base, and project database based on the research plan, and to classify and organize the search results and accurately extract the original text; controlling the research agent to receive the research agent's draft materials, load the research paradigm corresponding to the research plan, and write the manuscript; and controlling the review agent to conduct a quality review of the manuscript and issue review opinions. Furthermore, after the planning agent outputs the research plan, after the research agent completes the initial draft, and after the review agent issues its review opinions, there are human interaction checkpoints for waiting for user confirmation. This method achieves the pipelined, standardized, and traceable nature of knowledge-intensive research tasks through the serial collaboration of four specialized intelligent agents: planning, surveying, researching, and reviewing. It combines paradigm-driven knowledge research methodology adaptation, key node control with multiple human interaction checkpoints, a triple-source verification anti-counterfeiting mechanism, and automated quality gating of the verification checkpoint system. It effectively solves the problems of knowledge silos and information gaps, disconnect between research methods and execution, reliance on human review for quality control, and lack of human interaction nodes in multi-agent collaboration in existing technologies. It achieves a balance between the efficiency of automated collaboration and the human controllability of key nodes.

[0168] It should be noted that the computer-readable medium described above in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0169] In this disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0170] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0171] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0172] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product.

[0173] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A knowledge research system applied to electronic devices, characterized in that, The knowledge research system comprises a resource layer, an agent layer, and a flow control layer, wherein: The resource layer is configured with a raw material library, a knowledge base, and a project library. The raw material library is used to store read-only raw documents, the knowledge base is used to store refined structured knowledge networks, and the project library is used to store process documents and result documents of research projects. The intelligent agent layer is configured with a planning intelligent agent, a survey intelligent agent, a research intelligent agent, and a review intelligent agent. Each intelligent agent shares the resource layer and is configured with an independent exclusive skill package. The planning intelligent agent is used to receive research requirements and output research plans. The survey intelligent agent is used to perform material retrieval and original text extraction based on the research plan. The research intelligent agent is used to perform manuscript writing based on the retrieved materials and research plan. The review intelligent agent is used to perform quality review on the manuscript. The flow control layer is used to define the serial flow order among the four intelligent agents. The flow control layer sets human interaction checkpoints after the planning intelligent agent outputs the research plan, after the research intelligent agent completes the first draft, and after the review intelligent agent issues review opinions, so as to wait for human confirmation before continuing the flow.

2. The knowledge research system as described in claim 1, characterized in that, The flow control layer also includes a verification checkpoint system, which is set at the transmission interface between adjacent agents to verify the integrity of the data output by the upstream agent.

3. The knowledge research system as described in claim 1, characterized in that, The system is also equipped with a citation verification mechanism, which sets up independent source verification defenses in the survey agent, the research agent and the review agent stages.

4. The knowledge research system as described in claim 3, characterized in that, The back-to-source verification defense includes: The first line of defense is achieved by the survey agent locking the original text of the terms, reading the source file from the raw material library and extracting the original text of the terms word by word, and outputting the original text locking block; The second line of defense involves the research agent performing source verification, independently reading the source files in the raw material library and comparing them word by word with the original text lock blocks, and marking the verification as passed in the manuscript; The third line of defense involves the review agent performing independent source verification, extracting the citation list from the document, independently reading the source files in the raw material library for word-by-word comparison, and outputting a citation comparison table.

5. The knowledge research system as described in claim 1, characterized in that, The planning agent is equipped with a paradigm recognition module and a multi-scheme output module, wherein: The paradigm identification module is used to match a preset research paradigm classification system based on keywords in the research requirements, so as to determine the category to which the research belongs and match the corresponding soft science methodology framework. The multi-solution output module is used to output one or more differentiated alternative research schemes based on the requirements complexity assessment results. Each scheme includes a scheme name, applicable scenario, core logic, and advantages and disadvantages analysis.

6. The knowledge research system as described in claim 1, characterized in that, The survey agent is equipped with a paradigm-driven missing information identification module, wherein: The missing data identification module is used to automatically check whether the required data types are complete according to the research paradigm, and output supplementary data instructions when there are data gaps.

7. The knowledge research system as described in claim 1, characterized in that, The review agent is configured with a methodology compliance review module, wherein: The methodology compliance review module is used to check the correctness of research paradigm identification, the completeness of the soft science methodology framework, and the compatibility of analytical methods.

8. A research method applied to a knowledge research system as described in any one of claims 1 to 7, characterized in that, The research method includes the following steps: In response to the user's project initiation request, the planning agent is controlled to receive research requirements and output a research plan; In response to the user's confirmation of the research plan, the research agent is controlled to conduct cross-database searches in the raw material database, knowledge base, and project database according to the research plan, and to classify and organize the search results and accurately extract the original text. The research agent is controlled to receive the source drafts from the survey agent and load the research paradigm corresponding to the research plan to write the manuscript. The review agent is controlled to conduct a quality review of the document and issue review opinions. In particular, after the planning agent outputs the research plan, after the research agent completes the first draft, and after the review agent issues its review opinions, there are human interaction checkpoints. These human interaction checkpoints are used to wait for user confirmation.

9. The research method as described in claim 8, characterized in that, The process of controlling the planning agent to receive research requirements and output research plans includes the following steps: The planning agent is controlled to match a preset research paradigm classification system based on keywords in the research requirements, in order to determine the category to which the research belongs and match the corresponding soft science methodology framework. Based on the requirements complexity assessment results, output one or more differentiated alternative research solutions. Each solution includes a solution name, applicable scenarios, core logic, and advantages and disadvantages analysis.

10. The research method as described in claim 8, characterized in that, After the survey agent performs a cross-database search according to the research plan, the process also includes the following steps: The system automatically checks whether the required data types are complete based on the research paradigm, and outputs supplementary data instructions when there are data gaps.

11. The research method as described in claim 8, characterized in that, The process of controlling the review agent to conduct a quality review of the document and issue review opinions includes the following steps: The control system checks the correctness of the research paradigm identification, the integrity of the soft science methodology framework, and the compatibility of the analytical methods. The review agent is controlled to classify and label the problems found during the review, and the classification includes at least fatal errors, general flaws and optimization suggestions.

12. A research apparatus, applied to a knowledge research system as described in any one of claims 1 to 7, characterized in that, The research apparatus includes: The startup control module is configured to respond to the user's project startup request, control the planning agent to receive research requirements, and output a research plan; The research control module is configured to respond to the user's confirmation of the research plan, control the research agent to conduct cross-database searches in the raw material database, knowledge base, and project database according to the research plan, and classify and organize the search results and accurately extract the original text. The research control module is configured to control the research agent to receive the source draft of the survey agent and load the research paradigm corresponding to the research plan to write the manuscript. The review control module is configured to control the review agent to conduct a quality review of the document and issue review opinions. In particular, after the planning agent outputs the research plan, after the research agent completes the first draft, and after the review agent issues its review opinions, there are human interaction checkpoints. These human interaction checkpoints are used to wait for user confirmation.

13. An electronic device, characterized in that, The electronic device includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs or instructions; The processor is used to execute the computer program or instructions to enable the electronic device to implement the research method as described in any one of claims 8 to 11.

14. A computer-readable storage medium for use in electronic devices, characterized in that, The storage medium carries one or more computer programs that can be executed by the electronic device, thereby enabling the electronic device to perform the research method as described in any one of claims 8 to 11.

Images