Research report generation system
By adopting a collaborative model of overall allocation by the main agent and division of labor among sub-agents, the problem of low accuracy in research reports in multi-agent collaborative systems is solved, achieving efficient and accurate research report generation and enhancing the system's flexibility and versatility.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-26
- Publication Date
- 2026-04-14
AI Technical Summary
Existing multi-agent collaborative systems suffer from low accuracy in research report generation, primarily due to factual errors or illusory content from preceding agents being passed to subsequent agents through shared context, leading to amplified errors and a cascading propagation of illusions.
The main agent coordinates and allocates research report generation requests, which are broken down into multiple independent sub-tasks and executed by sub-agents. Each target sub-agent dynamically loads hierarchical resources and generates execution results. The main agent performs final compression processing to avoid error propagation and cascading.
It improves the accuracy and efficiency of research reports, reduces error propagation paths, enhances the system's flexibility and versatility, adapts to diverse report generation needs, and avoids the cascading spread of illusions.
Smart Images
Figure CN121860074A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and more specifically, to a research report generation system. Background Technology
[0002] With the development of artificial intelligence technology, multi-agent collaborative systems have been widely used in the field of content generation due to their advantages in distributed processing and task division. Existing multi-agent collaborative systems generally use a shared context to realize information transmission and task connection. Subsequent agents directly carry out their work based on the output of the preceding agent, and collaboration relies on the shared context flow.
[0003] However, this mechanism has inherent flaws. Factual errors or illusory content generated by preceding agents can be passed to subsequent agents through shared context; subsequent agents reason based on erroneous information, amplifying and accumulating the errors; multi-layered collaboration triggers a cascading propagation of illusions, severely reducing data credibility. This leads to a tendency for multi-agent collaborative systems to produce inaccurate research reports when applied to the field of in-depth research report generation. Therefore, how to provide an agent collaboration scheme for research report generation has become a key focus for those skilled in the art. Summary of the Invention
[0004] In view of this, this application provides a research report generation system to address the shortcomings of low accuracy in existing multi-agent collaboration technologies.
[0005] To achieve the above objectives, the following solution is proposed:
[0006] A research report generation system, comprising a main agent and multiple sub-agents;
[0007] The main agent obtains the research report generation request and the functional metadata of each sub-agent; breaks down the research report generation request into multiple independent sub-tasks; and determines the target sub-agent corresponding to each independent sub-task based on the functional metadata of each sub-agent.
[0008] Each target sub-Agent dynamically loads layered resources, executes corresponding independent sub-tasks, and generates execution results;
[0009] The main agent compresses the execution results and generates a research report.
[0010] Optionally, each target sub-Agent includes a task execution module;
[0011] The task execution module analyzes the corresponding independent subtask, calls the execution instruction set related to the independent subtask from the upper-level resource library, and executes the independent subtask in combination with the execution instruction set.
[0012] Optionally, the task execution module includes a data acquisition unit;
[0013] The data acquisition unit, in conjunction with the execution instruction set, extracts the information keywords corresponding to the independent subtask and retrieves multiple related information matching the information keywords from different data sources; and organizes each piece of related information.
[0014] Optionally, the data acquisition unit includes an information retrieval subunit;
[0015] The information retrieval subunit retrieves a list of trusted data sources from the lower-level resource library, and based on the list of trusted data sources, retrieves multiple relevant information matching the information keywords from different data sources.
[0016] Optionally, the data acquisition unit includes an information processing subunit;
[0017] The information processing subunit retrieves a data processing script from the lower-level resource library, uses the data processing script to perform integrity and consistency checks on each relevant information, and performs multi-level filtering and compression on each relevant information that passes the check to form an execution result.
[0018] Optionally, the information processing subunit includes an execution result acquisition component;
[0019] The execution result acquisition component performs multi-level filtering and compression on each relevant information that passes the inspection to generate an information digest; it also labels each relevant information that passes the inspection with a data source to form a labeling result; and based on the labeling result and the information digest, it generates the execution result of the corresponding independent subtask.
[0020] Optionally, the main agent includes a task decomposition module;
[0021] The task decomposition module extracts the research type and request keywords of the research report generation request; based on the research type and request keywords, it generates multiple independent sub-tasks.
[0022] Optionally, the main agent includes a task allocation module;
[0023] The task allocation module determines the target sub-Agent corresponding to each independent sub-task based on the information processing performance and functional type of each sub-Agent.
[0024] Optionally, the main agent includes a research report generation module;
[0025] The research report generation module integrates the various execution results into a research report based on a preset report template.
[0026] Optionally, the research report generation module includes an execution result integration unit;
[0027] The execution result integration unit performs credibility scoring and multi-level filtering on different relevant information in each execution result to form structured data, and integrates each execution result into a research report based on the structured data and the report template.
[0028] As can be seen from the above technical solution, the research report generation system provided in this application can include a main agent and multiple sub-agents. The main agent can obtain research report generation requests and functional metadata of each sub-agent; it decomposes the research report generation requests into multiple independent sub-tasks; and based on the functional metadata of each sub-agent, it determines the target sub-agent corresponding to each independent sub-task. Therefore, this application can optimize the rationality of task allocation and improve the accuracy of task execution through a collaborative mode of main agent coordination and sub-agent division of labor. Simultaneously, each target sub-agent executes its corresponding independent sub-task; therefore, the execution process of each target sub-agent is independent and will not affect each other, avoiding the cascading propagation of illusions. This system addresses the issue of low accuracy in research reports. Each target sub-agent in this system can dynamically load hierarchical resources, execute corresponding independent sub-tasks, and generate execution results. Thus, each target sub-agent can call appropriate resources as needed, avoiding the impact of resource redundancy or insufficiency on task execution, reducing the data volume of each target sub-agent, accelerating the task execution process, and improving overall task processing efficiency. Each sub-agent executes tasks independently, reducing error propagation paths in the traditional shared context mode and suppressing error amplification and cascading propagation from a mechanism perspective. Finally, the main agent can compress the execution results to generate a research report. Therefore, this application can use the main agent for final aggregation and data organization, filtering core and effective information. It is evident that this application, through the decomposition and allocation mode of the main agent's independent sub-tasks and the dynamic resource loading mode of the sub-agents, allows for flexible adaptation of sub-agents and resource configurations according to different research report generation needs, adapting to diverse report generation requirements, enhancing system versatility, avoiding illusory cascading propagation, and improving the effectiveness of generated research reports. Attached Figure Description
[0029] 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 embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0030] Figure 1 This is a schematic diagram of the architecture of a research report generation system disclosed in an embodiment of this application;
[0031] in, Figure 1 The correspondence between the reference numerals in the attached drawings and the component identifiers can be seen as follows:
[0032] Main Agent10, sub-Agent20. Detailed Implementation
[0033] 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 skilled in the art without creative effort are within the scope of protection of this application.
[0034] The research report generation system of this application can be used in a wide variety of general-purpose or special-purpose computing device environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor devices, distributed computing environments including any of the above devices, etc.
[0035] Next, combine Figure 1 This paper provides a detailed description of the research report generation system used in this application.
[0036] See Figure 1 It can be seen that this application may include a main Agent10 and multiple sub-Agents20.
[0037] Among them, the main Agent 10 can be the Agent that receives the research report generation request, and the main Agent 10 and the sub Agent 20 can be converted into each other.
[0038] The main Agent 10 and each sub-Agent 20 can be deployed in the same computing device environment or configuration, or they can be deployed in multiple computing device environments or configurations.
[0039] The main Agent 10 and each sub-Agent 20 can interact with each other.
[0040] The main agent 10 obtains the research report generation request and the functional metadata of each sub-agent; it breaks down the research report generation request into multiple independent sub-tasks; and it determines the target sub-agent corresponding to each independent sub-task based on the functional metadata of each sub-agent.
[0041] Specifically, each functional element may include the information processing performance and functional type of the corresponding sub-Agent20.
[0042] It can establish a one-to-one correspondence between each independent subtask and each target sub-Agent.
[0043] Describe independent subtasks in YAML format so that the corresponding target sub-Agent can understand the task.
[0044] Each target sub-Agent can dynamically load layered resources, execute corresponding independent sub-tasks, and generate execution results;
[0045] The main Agent10 compresses the execution results and generates a research report.
[0046] As can be seen from the above technical solution, the research report generation system provided in this application can include a main Agent 10 and multiple sub-Agents 20. The main Agent 10 can obtain research report generation requests and functional metadata of each sub-Agent 20; it decomposes the research report generation requests into multiple independent sub-tasks; and based on the functional metadata of each sub-Agent 20, it determines the target sub-Agent corresponding to each independent sub-task. Therefore, this application can optimize the rationality of task allocation and improve the accuracy of task execution through a collaborative mode of main Agent 10 coordinating and sub-Agents performing their respective tasks. Simultaneously, each target sub-Agent executes its corresponding independent sub-task; therefore, the execution process of each target sub-Agent is independent and will not affect each other, avoiding the cascading propagation of illusions. This addresses the issue of low accuracy in research reports caused by the previous system. Each target sub-Agent in this system can dynamically load hierarchical resources, execute corresponding independent sub-tasks, and generate execution results. Thus, each target sub-Agent can call appropriate resources on demand, avoiding the impact of resource redundancy or insufficiency on task execution, reducing the data volume of each target sub-Agent, accelerating the task execution process, and improving overall task processing efficiency. Each sub-Agent executes tasks independently, reducing error propagation paths in the traditional shared context mode and suppressing error amplification and cascading propagation from a mechanism perspective. Finally, the main agent 10 can compress the execution results to generate a research report. Therefore, this application can use the main agent 10 for final aggregation and data organization, filtering core and effective information.
[0047] As can be seen, this application, through the decomposition and allocation of independent subtasks of the main Agent 10 and the dynamic resource loading mode of the sub-agents, enables it to flexibly adapt sub-agents and resource configurations according to different research report generation needs. This adapts to diverse report generation requirements, enhances system versatility, avoids cascading illusion propagation, and improves the effectiveness of generated research reports. Since each subtask is independent, the various target sub-agents of this application can execute in complete parallel, improving parallelism.
[0048] This technical solution addresses the issues of cascading error propagation and insufficient accuracy in existing multi-agent systems by having a main agent (Agent10) coordinate and sub-agents collaborate, achieving multi-dimensional improvements: the main agent (Agent10) acquires functional metadata from each sub-agent, breaks down the research report generation request into multiple independent sub-tasks, accurately matches the corresponding target sub-agent, reduces initial errors and illusions, and improves execution accuracy; it achieves complete context isolation in the multi-agent system, transforming "shared collaboration" into "file-based collaboration," thus solving the problem of illusion accumulation at the architectural level.
[0049] Each sub-Agent dynamically loads layered resources to execute corresponding independent sub-tasks, breaking through the limitation of traditional systems that preload all capabilities into the context. Through layered dynamic loading, it achieves "use-as-you-go" functionality, supporting a massive number of skills while keeping the context lightweight; avoiding resource adaptation issues, supporting multi-task parallel processing, and improving overall task efficiency.
[0050] By reducing error propagation paths and suppressing error amplification and cascading propagation, the main Agent10 compresses multiple execution results, improving the report's structure and information density to ensure report quality; it also enhances system flexibility, scalability, and versatility, adapting to diverse report generation needs across various scenarios.
[0051] In some embodiments of this application, each target sub-Agent may include a task execution module.
[0052] The task execution module can analyze the corresponding independent subtask, call the execution instruction set related to the independent subtask from the upper-level resource library, and execute the independent subtask in combination with the execution instruction set.
[0053] Specifically, different types of independent subtasks can correspond to different sets of execution instructions.
[0054] For example, when the independent subtask is data collection, the execution instruction set can be used to characterize the relevant data obtained from a predefined list of data sources, primarily the data from the past 3 years and secondarily the data from the past 5 years.
[0055] When the independent subtask is data analysis, the instruction set can be used to characterize and calculate the market's compound annual growth rate (CAGR), identify key players and their market share, and analyze driving and restraining factors.
[0056] In addition, the execution instruction set may also include API call documentation, complex scenario processing procedures, common problems and solutions, and industry classification standards.
[0057] As can be seen from the above technical solution, this embodiment provides an optional composition method for a target sub-Agent. Through this method, the execution of independent sub-tasks can be completed by dynamically loading and executing instruction sets, thereby improving execution effectiveness.
[0058] In some embodiments of this application, the task execution module may include a data acquisition unit.
[0059] The data acquisition unit can combine the execution instruction set to extract the information keywords corresponding to the independent subtask, and retrieve multiple related information matching the information keywords from different data sources; and organize each piece of related information.
[0060] Specifically, different types of independent subtasks can correspond to different information keywords;
[0061] Information keywords can be extracted based on the type of independent subtask and the set of execution instructions.
[0062] As can be seen from the above technical solution, this embodiment provides an optional composition of the task execution module. Through this method, multiple relevant information matching keywords can be obtained from diverse data sources, which helps improve the quality and usability of the data, thereby enhancing the accuracy and effectiveness of the research report.
[0063] In some embodiments of this application, the data acquisition unit may include an information retrieval subunit;
[0064] The information retrieval subunit retrieves a list of trusted data sources from the lower-level resource library, and based on the list of trusted data sources, retrieves multiple relevant information matching the information keywords from different data sources.
[0065] Specifically, the list of trusted data sources can include the information acquisition methods and the degree of trustworthiness of different data sources.
[0066] Based on the information retrieval method of each data source in the list of trusted data sources, multiple relevant information matching the information keywords can be retrieved from the corresponding data source.
[0067] As can be seen from the above technical solution, this embodiment provides an optional composition of the data acquisition unit. Through this method, information can be retrieved based on a list of trusted data sources, ensuring that multiple relevant information sources originate from reliable channels, improving the authenticity and credibility of the data, and providing a solid data foundation for generating high-quality research reports.
[0068] In some embodiments of this application, the data acquisition unit may further include an information processing subunit;
[0069] The information processing subunit retrieves a data processing script from the lower-level resource library, uses the data processing script to perform integrity and consistency checks on each relevant information, and performs multi-level filtering and compression on each relevant information that passes the check to form an execution result.
[0070] Specifically, the information processing subunit can check the completeness and consistency of each piece of information by checking whether each piece of information contains required fields, verifying the rationality of the relevant information values, and detecting the existence of outliers.
[0071] The information processing sub-unit can also mark the credibility of key facts, perform secondary verification of sensitive information, and mark relevant information with uncertain authenticity as "to be verified".
[0072] Key facts can be relevant information with a wide impact.
[0073] Sensitive information can be related to sensitive allegations.
[0074] Information whose authenticity is uncertain may come from data sources with low credibility.
[0075] The information processing subunit can use a layered compression method to compress each piece of relevant information that has undergone multi-level filtering.
[0076] Specifically, layered compression can preserve core facts, halve supporting arguments, and retain a small amount of background information and process details;
[0077] It can convert layered compressed data into structured JSON.
[0078] As can be seen from the above technical solution, this embodiment provides another optional composition method for the data acquisition unit. Through this method, relevant resources can be retrieved to perform quality checks on the relevant information, and then the quality-checked information can be filtered and compressed to form the execution result, ensuring the integrity and accuracy of the execution result.
[0079] In some embodiments of this application, the information processing subunit includes an execution result acquisition component;
[0080] The execution result acquisition component performs multi-level filtering and compression on each relevant information that passes the inspection to generate an information digest; it also labels each relevant information that passes the inspection with a data source to form a labeling result; and based on the labeling result and the information digest, it generates the execution result of the corresponding independent subtask.
[0081] Specifically, the core facts of various relevant information can be integrated as an information summary.
[0082] The data source for each relevant piece of information can be identified, and the corresponding data source identifier can be marked on each relevant piece of information to form a labeling result.
[0083] The annotation results and information summaries can be integrated to form the execution results.
[0084] As can be seen from the above technical solution, this embodiment provides an optional composition method for the information processing subunit. Through this method, information summaries and data source annotations can be generated, thereby forming complete, accurate, and traceable execution results, providing high-quality data support for subsequent aggregation processing by the main Agent 10.
[0085] In some embodiments of this application, the main Agent 10 may include a task decomposition module;
[0086] The task decomposition module can extract the research type and request keywords of the research report generation request; based on the research type and request keywords, it can generate multiple independent sub-tasks.
[0087] Specifically, the task decomposition module can use a semantic analysis model to identify the intent and extract keywords from the research report generation request, and generate the research type and request keywords.
[0088] Research types can include electric vehicle market research reports, charging pile market research reports, etc.
[0089] Based on the research type and request keywords, combined with the predicted maximum context length, task focus, time span, and knowledge coverage, the task is broken down into multiple independent sub-tasks.
[0090] As can be seen from the above technical solution, this embodiment provides an optional composition method for the main Agent 10. This method improves the accuracy and efficiency of task generation, laying a solid foundation for the precise execution of subsequent sub-Agents.
[0091] In some embodiments of this application, the main Agent 10 may include a task allocation module.
[0092] The task allocation module can determine the target sub-Agent corresponding to each independent sub-task based on the information processing performance and functional type of each sub-Agent.
[0093] Specifically, the optimal target sub-agent for each independent sub-task can be determined by comprehensively considering multiple dimensions such as the sub-agent's computing power, historical task completion quality, and functional matching degree, and by using a weighted scoring mechanism.
[0094] As can be seen from the above technical solution, this embodiment provides another optional composition method for the main Agent 10. Through this method, both performance and functional type factors can be comprehensively considered to match the most suitable target sub-Agent to each independent sub-task, ensuring the scientific and reasonable allocation of tasks and further improving the task execution efficiency and accuracy of the entire research report generation system.
[0095] In some embodiments of this application, the main Agent 10 may include a research report generation module;
[0096] The research report generation module integrates the various execution results into a research report based on a preset report template.
[0097] Specifically, the report template can contain multiple areas corresponding to different data types.
[0098] The research report generation module can fill the corresponding area of the report template with the data type of each execution result to form a research report.
[0099] During the population process, the research report generation module can employ an intelligent loading strategy.
[0100] Specifically, the research report generation module can first load the information summary with the highest priority, and then load other content.
[0101] As can be seen from the above technical solution, this embodiment provides another optional composition method for the main Agent 10. Through this method, various execution results can be integrated according to a preset report template, ensuring a clear structure and accurate data in the research report, effectively improving the efficiency and quality of research report generation. At the same time, the flexibility of the report template allows the system to adapt to the generation needs of different types of research reports, enhancing the system's versatility and practicality.
[0102] In some embodiments of this application, the research report generation module may include an execution result integration unit.
[0103] The execution result integration unit can perform credibility scoring and multi-level filtering on different relevant information in each execution result to form structured data, and integrate the various execution results into a research report based on the structured data and the report template.
[0104] Specifically, the credibility score for each relevant piece of information is calculated as follows: w1 × authority score + w2 × timeliness score + w3 × relevance score + w4 × completeness score.
[0105] The authority score can be calculated based on factors such as the type of information source (e.g., academic journals, official websites, news media), expert ratings, and citation counts.
[0106] The timeliness score can be assessed based on factors such as the time difference between the information release time and the current time, and the update frequency.
[0107] Semantic similarity can be used to calculate the degree of matching between information and the research topic, and to determine the relevance score;
[0108] It can assess the structural integrity and data completeness of information and generate an integrity score.
[0109] w1, w2, w3, and w4 can be dynamically adjusted according to the research type.
[0110] Multi-stage filtration can include three stages of filtration.
[0111] The first level of filtering may include removing advertising noise, standardizing formats, and unifying language types;
[0112] Second-level filtering may include integrity and consistency checks.
[0113] The third level of filtering may include relevance ranking, importance assessment, and effectiveness assessment.
[0114] Based on relevance ranking, relevant information of the same subject is aggregated to obtain multiple aggregated subsets;
[0115] It can extract consensus content and mark divergence points in the overlapping parts of aggregate subsets, compress the content based on the importance assessment results and the effectiveness assessment results, and generate structured data.
[0116] As can be seen from the above technical solution, this embodiment provides an optional composition method for a research report generation module. Through this method, data quality can be ensured through reliable scoring and multi-level filtering. Then, structured data is integrated into a research report based on a report template, making the report content more organized, easier to read and understand, and improving the user experience.
[0117] The following section will provide a detailed introduction to this application through specific examples.
[0118] The main Agent10 received a research report generation request to "generate an in-depth research report on the global electric vehicle market, including market size, technology trends, major manufacturers, policy environment and future forecasts".
[0119] The main Agent10 can break down the research report generation request into multiple independent sub-tasks, such as market size analysis, technology trend research, competitive landscape survey, policy environment analysis, and trend prediction modeling.
[0120] Subsequently, the main Agent10 can assign corresponding target sub-Agens for market size analysis, technology trend research, competitive landscape survey, policy environment analysis, and trend prediction modeling.
[0121] The target sub-gen for market size analysis uses market, size, growth, and competition as information keywords, dynamically loads hierarchical resources, and generates execution results.
[0122] The target sub-gen of the technology trend research is executed, using technology, trend, innovation and roadmap as information keywords, dynamically loading layered resources and generating execution results;
[0123] The target sub-gen of the competitive landscape survey uses business competition and market environment as information keywords, dynamically loads layered resources, and generates execution results.
[0124] The target sub-Agen of policy environment analysis uses policies as information keywords, dynamically loads layered resources, and generates execution results.
[0125] The target sub-Agen for executing trend prediction modeling uses trend, future, and prediction as information keywords, dynamically loads hierarchical resources, and generates execution results.
[0126] The main Agent10 can compress the execution results to generate an in-depth research report on the global electric vehicle market.
[0127] A comparative experiment between the multi-agent collaboration system and this application revealed the following differences: In terms of average context length, the multi-agent collaboration system used 32K-128K tokens, while this application used 4K-8K tokens; regarding the hallucination rate, the multi-agent collaboration system had a rate of 15-30%, while this application reduced it to 3-8%; in terms of information accuracy, the multi-agent collaboration system achieved 70-80%, while this application improved it to 88-95%; the report generation time for the multi-agent collaboration system was 5-15 minutes, while this application only took 3-8 minutes; the total token consumption for the multi-agent collaboration system was 150K-500K, while this application reduced it to 50K-150K; and the system reliability (failure-free rate) for the multi-agent collaboration system was 75-85%, while this application reached 92-98%. Overall, this application outperforms the multi-agent collaboration system in all metrics.
[0128] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0129] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0130] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. The various embodiments of this application can be combined with each other. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A research report generation system, characterized in that, Includes a main Agent and multiple sub-Agents; The main agent obtains the research report generation request and the functional metadata of each sub-agent; breaks down the research report generation request into multiple independent sub-tasks; and determines the target sub-agent corresponding to each independent sub-task based on the functional metadata of each sub-agent. Each target sub-Agent dynamically loads layered resources, executes corresponding independent sub-tasks, and generates execution results; The main agent compresses the execution results and generates a research report.
2. The research report generation system according to claim 1, characterized in that, Each target sub-Agent includes a task execution module; The task execution module analyzes the corresponding independent subtask, calls the execution instruction set related to the independent subtask from the upper-level resource library, and executes the independent subtask in combination with the execution instruction set.
3. The research report generation system according to claim 2, characterized in that, The task execution module includes a data acquisition unit; The data acquisition unit, in conjunction with the execution instruction set, extracts the information keywords corresponding to the independent subtask and retrieves multiple related information matching the information keywords from different data sources; and organizes each piece of related information.
4. The research report generation system according to claim 3, characterized in that, The data acquisition unit includes an information retrieval subunit; The information retrieval subunit retrieves a list of trusted data sources from the lower-level resource library, and based on the list of trusted data sources, retrieves multiple relevant information matching the information keywords from different data sources.
5. The research report generation system according to claim 3, characterized in that, The data acquisition unit includes an information processing subunit; The information processing subunit retrieves a data processing script from the lower-level resource library, uses the data processing script to perform integrity and consistency checks on each relevant information, and performs multi-level filtering and compression on each relevant information that passes the check to form an execution result.
6. The research report generation system according to claim 5, characterized in that, The information processing subunit includes an execution result acquisition component; The execution result acquisition component performs multi-level filtering and compression on each relevant information that passes the inspection to generate an information digest; it also labels each relevant information that passes the inspection with a data source to form a labeling result; and based on the labeling result and the information digest, it generates the execution result of the corresponding independent subtask.
7. The research report generation system according to claim 1, characterized in that, The main Agent includes a task decomposition module; The task decomposition module extracts the research type and request keywords of the research report generation request; based on the research type and request keywords, it generates multiple independent sub-tasks.
8. The research report generation system according to claim 1, characterized in that, The main agent includes a task allocation module; The task allocation module determines the target sub-Agent corresponding to each independent sub-task based on the information processing performance and functional type of each sub-Agent.
9. The research report generation system according to claim 1, characterized in that, The main agent includes a research report generation module; The research report generation module integrates the various execution results into a research report based on a preset report template.
10. The research report generation system according to claim 9, characterized in that, The research report generation module includes a result integration unit; The execution result integration unit performs credibility scoring and multi-level filtering on different relevant information in each execution result to form structured data, and integrates each execution result into a research report based on the structured data and the report template.