Financial research report generation method, system and device and readable storage medium

Abstract

The invention discloses a financial research report generation method, system and device and a readable storage medium, and relates to the technical field of finance, the financial research report generation method is applied to the financial research report generation system, and the financial research report generation system comprises a scheduling layer, an execution layer and an output layer which are connected in sequence. The financial research report generation method comprises the following steps: receiving a research report generation task through a task planning agent in a scheduling layer, decomposing the research report generation task into a plurality of sub-tasks, and sending each sub-task to an execution layer; each agent in the execution layer is responsible for processing the corresponding subtask and generating a corresponding task execution result; and generating a target financial research report through a report generation agent in an output layer according to task execution results output by the execution agents respectively. According to the technical scheme of the invention, through cooperation and division of labor of multiple agents, the generation efficiency and professional quality of financial research reports are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of finance, in particular to a financial research report generation method, system, device and computer readable storage medium. BACKGROUND

[0002] A financial research report is a systematic analysis report published by a financial institution or a professional research team, aiming to provide investors with in-depth research and investment recommendations on the market, industry, company or macro economy. In addition to manual writing, traditional financial research report generation methods also exist generation schemes using deep learning models. A single model undertakes all functions in the financial research report generation process, which may have the defect of insufficient integration capability for multi-source data. The expert system scheme based on the rule engine is very dependent on the preset inherent rules, is relatively rigid, and has low processing efficiency, and is difficult to cope with large-scale concurrent tasks. In summary, the above various defects of traditional various financial research report generation schemes will affect the generation efficiency and quality of the financial research report.

[0003] The above information disclosed in this BACKGROUND section is only for the purpose of understanding the background of the present application, and therefore, it can contain information that does not constitute prior art. SUMMARY

[0004] The main purpose of the present application is to provide a financial research report generation method, system, device and computer readable storage medium, aiming to improve the generation efficiency and professional quality of the financial research report through the cooperation and division of labor of multiple agents.

[0005] To achieve the above-mentioned purpose, the present application provides a financial research report generation method, which is applied to a financial research report generation system, the financial research report generation system comprising a scheduling layer, an execution layer and an output layer connected in turn, and the financial research report generation method comprising: receiving, by a task planning agent in the scheduling layer, a research report generation task, decomposing the research report generation task into a plurality of subtasks, and sending each of the subtasks to the execution layer; each agent in the execution layer is responsible for processing its corresponding subtask and generating a corresponding task execution result; a report generation agent in the output layer generates a target financial research report according to the task execution results output by each of the execution agents.

[0006] In an embodiment, the task planning agent is further configured to send a preset tool calling rule, a parameter passing mechanism and a result processing flow to each of the execution agents; The task planning agent is also configured to monitor the task execution status of each of the execution agents, and control the execution agents to perform task retry when the task execution status of the execution agents is abnormal.

[0007] In an embodiment, the execution agents in the execution layer include at least a deep research agent, a browser agent and a deep analysis agent, and the sub-tasks include at least a deep research sub-task, a browser sub-task and a deep analysis sub-task, and the task execution results include at least a target data source list, target structured data and data analysis results. The step of processing each sub-task by each agent in the execution layer and generating corresponding task execution results includes: The deep research agent determines a target data source list according to the research topic in the deep research sub-task; The browser agent acquires target structured data corresponding to the browser sub-task according to the target data source list; The deep analysis agent executes the deep analysis sub-task according to the target structured data and generates data analysis results.

[0008] In an embodiment, the step of determining a target data source list according to the research topic in the deep research sub-task includes: Based on the research topic in the deep research sub-task, a plurality of search operation steps are generated, and a search operation corresponding to the first search operation step is executed to obtain corresponding search results; The search results are analyzed to determine whether re-searching is needed; If yes, the search operation corresponding to the first search operation step is executed again; If no, a search operation corresponding to the next search operation step is executed to obtain corresponding search results; After all search operation steps are searched, a data source list in the form of structured data is generated according to the search results; It is determined whether the data source list in the form of structured data meets preset requirements; If yes, the data source list in the form of structured data is determined as the target data source list, and the browser agent is returned; If no, the search operation corresponding to the first search operation step is executed again.

[0009] In an embodiment, the step of acquiring target structured data corresponding to the browser sub-task according to the target data source list includes: Receive the browser subtask, execute the browser subtask based on the target data source list, and obtain the corresponding data content; Evaluate the data content to determine whether re-execution is necessary; If so, return to the execution steps: execute the browser subtask based on the target data source list to obtain the corresponding data content; If not, the data content will be extracted and archived to generate structured data; Determine whether the structured data meets the preset requirements; If the preset requirements are met, the structured data is identified as the target structured data and returned to the deep analysis agent; If the requirements are not met, return to the execution step: execute the browser subtask based on the target data source list to obtain the corresponding data content.

[0010] In one embodiment, the step of performing the deep analysis subtask based on the target structured data and generating data analysis results includes: Generate a data analysis plan corresponding to the browser subtask based on the target structured data, execute the operation of the first data analysis step in the data analysis plan, and obtain the corresponding analysis results; Evaluate the analysis results to determine whether a reanalysis is necessary; If so, return to the operation of executing the first data analysis step in the data analysis plan; If not, continue with the next data analysis step in the data analysis plan to obtain the corresponding analysis results; After all data analysis steps are completed, structured analysis data is generated based on the analysis results. Determine whether the structured analysis data meets preset requirements; If the preset requirements are met, the structured analysis data will be identified as the data analysis result and returned to the report generation agent. If the requirements are not met, return to the operation of the first data analysis step in the data analysis plan.

[0011] In one embodiment, the report generating agent is used to integrate the task execution results after each execution agent in the execution layer outputs the task execution results, and render the integrated task execution results based on a preset report format to obtain a target financial research report, wherein the target financial research report includes at least charts, tables and annotations.

[0012] In addition, this application also provides a financial research report generation system, which includes at least a scheduling layer, an execution layer and an output layer connected in sequence; The scheduling layer includes a task planning agent, which is used to receive a research report generation task, decompose the research report generation task into multiple sub-tasks, and send each of the sub-tasks to the execution layer. The execution layer includes at least one execution agent, which is used to execute the sub-task and obtain the corresponding task execution result; The output layer includes a report generation agent, which generates a target financial research report based on the task execution results output by each of the execution agents.

[0013] In addition, this application also provides a financial research report generation device, which includes at least: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the financial research report generation method applied to the financial research report generation device as described above.

[0014] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the financial research report generation method described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the financial research report generation method described above.

[0016] This application provides a method for generating financial research reports. This method is applied to a financial research report generation system, which includes a scheduling layer, an execution layer, and an output layer connected sequentially. The method includes: first, receiving a research report generation task through a task planning agent in the scheduling layer; decomposing the research report generation task into multiple sub-tasks; sending each sub-task to the execution layer; then, each agent in the execution layer is responsible for processing its corresponding sub-task and generating corresponding task execution results; finally, a report generation agent in the output layer generates a target financial research report based on the task execution results output by each execution agent. The financial research report generation system of this application utilizes multiple agents working collaboratively. Each agent is responsible for a portion of the tasks in the financial research report generation process. Each agent can be trained and deployed independently, enabling the integration of multi-source data through specialized agents without relying on inherent specialized system rules. Through the division of labor among specialized intelligent agents, it can effectively address various complex financial analysis and research report generation needs. The financial research report generation system, which requires no human intervention, supports continuous output around the clock and can handle the needs of large-scale concurrent tasks, effectively improving the efficiency of financial research report generation. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the financial research report generation method in the embodiments of this application; Figure 2 This is a schematic diagram of the architecture of a feasible financial research report generation system according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating the workflow of the task planning agent in the financial research report generation process in the embodiments of this application; Figure 4 This is a schematic diagram of the collaborative process of various intelligent agents in the financial research report generation system of this application embodiment; Figure 5 This is a schematic diagram illustrating the workflow of a deep research agent in the financial research report generation process in this application embodiment; Figure 6 This is a schematic diagram illustrating the workflow of the browser agent in the financial research report generation process in this application embodiment; Figure 7 This is a schematic diagram illustrating the workflow of the deep analysis agent in the financial research report generation process in this embodiment of the application; Figure 8 This is a schematic diagram of the MCP (Model Control Protocol) extended architecture of the financial research report generation system in the embodiments of this application; Figure 9 This is a schematic diagram of the hardware operating environment involved in the financial research report generation method in this application embodiment.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] The technical deficiencies to be overcome by the embodiments of this application include: low efficiency in generating traditional financial research reports: manual writing of research reports usually takes 2-3 days, which cannot meet the needs of the rapidly changing financial market; poor consistency in the quality of research reports: reliance on manual analysis has problems of strong subjectivity and inconsistent standards; difficulty in integrating multi-source data: existing systems have difficulty in efficiently integrating multi-source heterogeneous data from the National Bureau of Statistics, stock exchanges, financial media, etc.; insufficient professional analysis capabilities: traditional AI (Artificial Intelligence) systems or models lack in-depth financial professional knowledge and analysis methods; poor visualization effects: reports generated by existing systems lack professional presentation requirements such as chart integration and data traceability.

[0025] To overcome the aforementioned technical deficiencies and achieve the aforementioned beneficial effects, this application provides a method for generating financial research reports. This method is applied to a financial research report generation system, which includes a scheduling layer, an execution layer, and an output layer connected sequentially. (Refer to...) Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the financial research report generation method of this application. The financial research report generation method includes: Step S10: Receive the research report generation task through the task planning agent in the scheduling layer, decompose the research report generation task into multiple sub-tasks, and send each sub-task to the execution layer. It should be noted that in the financial research report generation system, the scheduling layer and the execution layer are interconnected. The task planning agent in the scheduling layer can schedule one or more execution agents included in the execution layer, control each execution agent to execute the corresponding task, and manage the task execution status of each execution agent. The scheduling layer directly receives research report generation tasks sent by users. The research report generation tasks are used to represent the user's research report generation needs, report type (macroeconomic financial research report / industry financial research report / corporate financial research report), and detailed report requirements.

[0026] The task planning agent in the scheduling layer can decompose complex research report generation tasks into a sequence of executable subtasks. Each subtask sequence contains multiple subtasks with sequential and dependent relationships. Thus, this embodiment simplifies and decomposes the financial research report generation task, facilitating subsequent execution by other agents and resulting in reliable financial research reports.

[0027] It should be noted that an AI agent is an intelligent entity capable of autonomously perceiving its environment, making decisions, and executing tasks, thereby automating complex tasks using artificial intelligence technology. The AI ​​agent acquires environmental information (such as text, images, and speech) through a perception module, plans action steps through a decision-making module (typically relying on a large language model), and finally completes the task by calling tools through an execution module. Its core capabilities include autonomy, perception, decision-making, and action, forming a closed loop of "perception-planning-action." Therefore, in this embodiment, the collaborative work of multiple AI agents (including various types of specialized AI agents in the scheduling layer, execution layer, and output layer) to complete the financial research report generation task enables the financial research report generation system to have stronger stability, adaptability, and self-learning capabilities. It can handle complex financial analysis tasks, possess real-time data acquisition capabilities, and have broader information coverage, thus effectively ensuring the timeliness, quality, and comprehensiveness of the output financial research reports.

[0028] In step S20, each agent in the execution layer is responsible for processing its corresponding subtask and generating the corresponding task execution result; The financial research report generation system includes an execution layer that communicates with the scheduling layer. One or more execution agents in the execution layer are used to execute sub-tasks sent by the task planning agent. The sub-tasks are the steps that must be performed in the process of generating financial research reports, such as screening target data sources, searching for relevant data, and analyzing data.

[0029] After the intelligent agent completes the corresponding subtask, it will output the corresponding task execution result. The task execution result may include one or more of the following: web page content summary, key information crawled, data in-depth analysis results, data source verification, modeling analysis results, and mined deep data. These task execution results are all information that needs to be presented in financial research reports.

[0030] Step S30: The report generation agent in the output layer generates the target financial research report based on the task execution results output by each execution agent.

[0031] In the financial research report generation system, the output layer and the execution layer are connected to each execution agent. The report generation agent in the output layer can integrate the task execution results output by each execution agent to generate a target financial research report that meets the user's expectations.

[0032] The main task of the report generation agent is to generate financial research reports in text format that conforms to users' browsing and reading habits. The financial research reports include information such as text, charts, annotations, and layout, while also taking into account the sense of design and aesthetics, so that users can understand the content presented in the financial research reports more efficiently and clearly when browsing them.

[0033] It should be noted that each execution agent and report generation agent in the execution layer is a specialized agent designed for the financial field and the characteristics of financial research reports. They can be built and trained separately. Each execution agent is responsible for a part of the work, which allows it to focus more on its own sub-task. Compared with the traditional solution of generating the entire financial research report by a single model, this approach is more professional and refined, and can better guarantee the quality of the output financial research reports.

[0034] Furthermore, in a feasible embodiment, the execution agent in the execution layer includes at least a deep research agent, a browser agent, and a deep analysis agent; the subtask includes at least a deep research subtask, a browser subtask, and a deep analysis subtask; and the task execution result includes at least a list of target data sources, target structured data, and data analysis results. The steps for each agent in the execution layer to process its corresponding subtask and generate the corresponding task execution result may include: Step S21: The deep research agent determines the list of target data sources based on the research topics in the deep research sub-tasks; Step S22: The browser agent obtains the target structured data corresponding to the browser subtask based on the target data source list; Step S23: The deep analysis agent performs deep analysis sub-tasks based on the target structured data to generate data analysis results.

[0035] The execution layer in this application embodiment includes three different types of execution agents: Deep Researcher Agent, Browser Agent, and DeepAnalyze Agent.

[0036] Specifically, the deep research agent is used to filter high-quality information sources based on research report topics (including the companies, industries, or other types of topics in the research reports), build multi-level search strategies, cover official statistics, exchange announcements, authoritative media, etc., evaluate the authority and timeliness of information sources, and provide a structured list of data sources for the browser agent.

[0037] The browser agent is used to perform automated web browsing, acquire real-time data, parse structured data such as PDF (Portable Document Format) files and Excel (spreadsheet) tables, process dynamically loaded content and CAPTCHAs using JavaScript (a just-in-time compiled programming language), and implement intelligent waiting and retry mechanisms to ensure the stability of data acquisition.

[0038] The deep analysis agent is used to implement professional financial analysis methods on the data searched by the other execution agents mentioned above, such as DCF (Discounted Cash Flow) valuation, ratio analysis, peer comparison, to build quantitative models for trend prediction and risk assessment, and to generate strategic analysis frameworks such as SWOT (Strengths, Weaknesses, Opportunities, Threats) analysis and Porter's Five Forces analysis, so as to provide users with the logical support basis for financial investment advice and risk warnings.

[0039] Combining the task planning agent and report generation agent in the aforementioned embodiments, the information interaction between the agents in the entire financial research report system is as follows: Figure 2As shown, the task planning agent receives report generation tasks such as "Generate an in-depth research report on Company XXX," "Research report on the XXX service industry," or "Macroeconomic research report on XXX," where "XXX" refers to the company name, industry name, or field name. The task planning agent includes a code interpreter, task actions, and task tools. It connects to the deep research agent, browser agent, and deep analysis agent, respectively, and can schedule these agents. The deep research agent, browser agent, and deep analysis agent demonstrate their functions by outputting observational information. Specifically, the deep research agent filters potential information sources, the browser agent performs webpage interaction and information extraction, and the deep analysis agent performs financial analysis, comparative analysis, and valuation analysis based on financial data, company information, and competitor identification. The report generation agent generates HTML (Hyper Text Markup Language) reports by chapter based on the content output by the deep research agent, browser agent, and deep analysis agent, and highlights the outputs. It also supports HTML to DOCX (a document format).

[0040] In one feasible embodiment, the task planning agent is also used to send preset tool invocation rules, parameter passing mechanisms, and result processing procedures to each execution agent; The task planning agent is also used to monitor the task execution status of each execution agent and to control the execution agent to retry the task when there is an anomaly in the task execution status of the execution agent.

[0041] Specifically, the task planning agent can also be used to dynamically schedule execution agents of various specialties in the execution layer, control the execution flow and dependencies of the agents, monitor the execution status of each agent, handle abnormal situations, and control agents to retry tasks. For example, if an execution agent encounters an exception during the execution of a subtask, stopping or failing to output the task execution result, the task planning agent can detect the abnormal state of the execution agent and schedule the execution agent to re-execute the subtask until the execution agent outputs a complete task execution result.

[0042] For example, communication and control of action steps between different agents are achieved within the task planning agent using JSON (a lightweight data exchange format). The JSON format for tool calls, parameter passing mechanisms, and result processing flows are explicitly defined. The protocol within the task planning agent mandates that each tool call must include a precise tool name and parameter values, rather than variable names, effectively solving the problem of ambiguous parameter passing in traditional systems. The passing format can include the following: { "name": "tool_name", "arguments" }” Here, name represents the name, tool_name represents the tool name, and arguments represent the parameters.

[0043] For example, the workflow of the task planning agent in the financial research report generation process is as follows: Figure 3 As shown, the system first receives the user task (i.e., the research report generation task), then performs task analysis, planning, and generates a detailed execution plan and action generation (equivalent to decomposing sub-tasks). Then, it makes a tool call to schedule each agent to execute the sub-tasks, obtains the observation results, and performs result verification and decision-making to determine whether more steps are needed. If so, it returns to execute the action generation step and further decomposes the sub-tasks; otherwise, it directly outputs the results to the report generation agent to output the target financial research report.

[0044] Furthermore, in the process of scheduling various execution agents (including deep research agents, browser agents, and deep analysis agents) through the task planning agent, the collaborative process of each agent is as follows: Figure 4 As shown. First, in the scheduling layer, the task planning agent schedules the agents in the execution layer, including: the browser use agent, the deep researcher agent, and the data analyze agent. The sub-tasks received by each agent can be in the following form: the task "original task description" must include: a complete task, detailed context, and a clear expected output. The browser use agent's output includes: webpage content summaries and key information extraction. The deep researcher agent's output includes: deep research results and information source verification. The data analyze agent's output includes: professional modeling analysis and deep data mining. Finally, the output of the observer layer (i.e., the execution layer) is sent to the report generation agent in the output layer (final report). After verifying the consistency of information, the final answer (i.e., the target financial research report) is generated. Here, verifying the consistency of information means verifying whether the searched data content describes the same topic or object.

[0045] In one feasible embodiment, when the research report generation task involves source retrieval, the execution process includes the following steps: 1. The planning agent analyzes the task requirements and determines that the browser use agent needs to be invoked; 2. Generate a detailed task description, including the original task instructions and necessary context; 3. Call the browser use agent, passing in a complete task description; 4. Receive and analyze the results to determine whether a deep researcher agent is needed for further research; 5. Integrate the results to generate the final answer.

[0046] Furthermore, in one feasible embodiment, the step of determining the target data source list based on the research report topics in the in-depth research sub-task may include: Step S211: Based on the research report topics in the in-depth research subtask, generate multiple corresponding search operation steps, and execute the search operation corresponding to the first search operation step to obtain the corresponding search results; Step S212: Analyze the search results to determine whether a re-search is needed; Step S213: If yes, then return to execute the search operation corresponding to the first search operation step; Step S214: If not, execute the search operation corresponding to the next search operation step to obtain the corresponding search result; Step S215: After all search operations are completed, generate a list of data sources in structured data format based on each search result. Step S216: Determine whether the list of data sources in the form of structured data meets the preset requirements; Step S217: If the preset requirements are met, the list of data sources in the form of structured data is determined as the target data source list and returned to the browser agent; Step S218: If the preset requirements are not met, return to execute the search operation corresponding to the first search operation step.

[0047] This application specifically discloses the process by which a deep research agent performs deep research sub-tasks, as follows: Figure 5 As shown, the scheduling layer schedules the deep researcher agent to perform deep network search and information verification. The execution layer of the deep research agent specifically includes an input processing layer (used to receive deep research sub-tasks, analyze tasks, and generate execution plans), a research execution layer (used for network search, archive search, historical record analysis, and result parsing), and a result processing layer (used for information extraction, information verification, cross-validation, multi-source integration, and structured output).

[0048] Furthermore, the specific execution steps (Action / Observation) include: first, receiving the deep research subtask sent by the task planning agent; decomposing it into multiple search operation steps based on the research topic; then, performing the search operation on the first of the search operation steps; analyzing the results of the search operation and deciding on the next action; the analysis includes: Is more step needed? Is the information reliable? Is more data needed? The aforementioned analysis steps mainly judge based on the completeness and information quality of the results of the existing search operations; if more step is needed, the information is unreliable, or more data is needed, then return to execute the search operation; if no more step is needed, the information is reliable, or more data is not needed, then continue searching and analyzing until all search operation steps are completed and a structured final answer is generated; then determine whether the structured final answer includes the following parts (corresponding to preset requirements): 1. Task result (brief version), 2. Task result (extremely detailed version), 3. Additional context; if so, then return the structured answer through the final_answer tool (equivalent to returning the target data source list to the browser agent); otherwise, return to execute the search operation.

[0049] In one feasible embodiment, the step of obtaining the target structured data corresponding to the browser subtask based on the target data source list may include: Step S221: Receive browser subtasks, execute the browser subtasks based on the target data source list, and obtain the corresponding data content; Step S222: Evaluate the data content and determine whether it needs to be re-executed; Step S223, if yes, then return to the execution step: execute the browser subtask based on the target data source list to obtain the corresponding data content; Step S224: If not, extract and archive the data content to generate structured data; Step S225: Determine whether the structured data meets the preset requirements; Step S226: If the preset requirements are met, the structured data is identified as the target structured data and returned to the deep analysis agent; Step S227: If the requirements are not met, return to the execution step: Execute the browser subtask based on the target data source list to obtain the corresponding data content.

[0050] This application specifically discloses the process by which a browser agent executes browser subtasks, as follows: Figure 6As shown, the scheduling layer schedules the browser use agent to perform automatic browser browsing and content analysis. The execution layer of the browser agent specifically includes an input processing layer (used to receive user tasks, analyze tasks, and generate execution plans), a browser execution layer (used to start the search engine, MCP (Model Control Protocol) server, web page interaction, and file browsing), and a content analysis and processing layer (used for content extraction, screenshot analysis, code analysis, and data analysis).

[0051] Furthermore, the specific execution steps (Action / Observation) include: first, receiving the source browsing task (i.e., browser subtask); then, combining the target data source list (providing data sources), executing the browser subtask to perform webpage interaction / file browsing, extracting content, and obtaining the corresponding data content; based on the completeness and information quality of this data content, determining whether more steps are needed; determining whether the iteration limit has been reached; if yes, returning to execute the webpage interaction / file browsing steps; if no, extracting and archiving information from the data content to generate a structured final answer; determining whether the structured final answer includes the following parts (corresponding to preset requirements): 1. Task result (brief version), 2. Task result (extremely detailed version), 3. Additional context; if yes, returning the structured answer through the final_answer tool (equivalent to returning the target structured data to the deep analysis agent); otherwise, returning to execute the webpage interaction / file browsing steps.

[0052] In one feasible embodiment, the step of performing a deep analysis subtask based on the target structured data and generating data analysis results may include: Step S231: Generate a data analysis plan corresponding to the browser subtask based on the target structured data, execute the operation of the first data analysis step in the data analysis plan, and obtain the corresponding analysis results; Step S232: Evaluate the analysis results and determine whether reanalysis is necessary; Step S233: If yes, return to the operation of executing the first data analysis step in the data analysis plan; Step S234: If not, continue to execute the next data analysis step in the data analysis plan to obtain the corresponding analysis results; Step S235: After all the data analysis steps are completed, structured analysis data is generated based on the analysis results. Step S236: Determine whether the structured analysis data meets the preset requirements; Step S237: If the preset requirements are met, the structured analysis data is determined as the data analysis result and returned to the report generation agent; If the requirements are not met in step S238, return to the operation of the first data analysis step in the data analysis plan.

[0053] This application specifically discloses the process by which a deep analysis agent performs deep analysis subtasks, as follows: Figure 7 As shown, the scheduling layer schedules deep analyze to perform deep network search and information verification. The execution layer of the deep analysis agent specifically includes an input processing layer (used to receive user tasks, analyze tasks, and generate execution plans), an analysis execution layer (used for data loading, code execution, chart drawing, and data statistics), and a result processing layer (used for data cleaning, feature extraction, cross-validation, and data mining).

[0054] Furthermore, the specific execution steps (Action / Observation) include: first, generating a corresponding analysis plan based on the deep analysis subtask and the target structured data. The analysis plan includes multiple data analysis steps for deep analysis of the target structured data. Then, executing the first data analysis step yields the analysis results, which are then evaluated. Specifically, this includes: are more steps needed? Has the iteration limit been reached? If yes, then return to executing the first data analysis step; otherwise, continue with data mining analysis, i.e., continue processing other data analysis steps until all data analysis steps are completed and a structured final answer is generated. It is then determined whether the structured final answer includes the following parts (corresponding to preset requirements): 1. Task result (brief version), 2. Task result (extremely detailed version), 3. Additional context. If yes, then return the structured answer via the final_answer tool (equivalent to returning the data analysis results to the report generation agent); otherwise, return to executing the first data analysis step.

[0055] In one feasible embodiment, the report generation agent is used to integrate the task execution results after each execution agent in the execution layer outputs the task execution results, and render the integrated task execution results based on a preset report format to obtain the target financial research report, wherein the target financial research report includes at least charts, tables and annotations.

[0056] Specifically, the report generation agent integrates the analysis results of various agents, constructs a complete report structure, renders it in HTML format, including charts, tables, and highlights, and performs quality assessments to ensure logical consistency and professional standards. It also supports multiple output format conversions (PDF, Word, HTML) to provide users with a visual financial research report.

[0057] In one feasible embodiment, the agents in the financial research report generation system of this application communicate using a standardized Call JSON (a lightweight data exchange format) format to ensure the accuracy and reliability of information transmission. Because a standardized JSON communication protocol and exception handling mechanism are established, the collaboration between agents is more stable and reliable, supporting 24 / 7 uninterrupted operation, thus meeting the stringent timeliness requirements of the financial market.

[0058] Moreover, because of the introduction of specialized intelligent agents, each agent focuses on professional tasks in a specific field. The Deep Analyze agent has built-in professional methods such as DCF valuation and SWOT analysis, while the Final Report agent implements unified quality assessment standards, thereby ensuring the professionalism and consistency of the target financial research reports and eliminating the subjective bias of human analysis.

[0059] The financial research report generation system employs a multi-agent parallel processing mechanism, enabling simultaneous execution of tasks such as data acquisition, analysis and modeling, and content generation. This reduces report generation time from the traditional 2-3 days to 2-3 hours, improving efficiency by over 90%. The unified scheduling of the Planning Agent avoids resource conflicts, ensuring efficient task execution.

[0060] In one feasible embodiment, the financial research report generation system of this application adopts an MCP server architecture, which supports dynamic tool expansion. Therefore, the system can dynamically access new data sources, analysis tools, and rendering methods, adapting to market changes and business needs without modifying the core code, thus possessing excellent scalability and adaptability.

[0061] For example, the MCP server extended architecture is as follows: Figure 8As shown, it includes a financial MCP server, a web crawler MCP server, and a data analysis MCP server. The financial MCP server can include information such as Wind (a financial information website), Askshare (an open-source financial data interface library), and the CSI 300 Shanghai Stock Exchange Composite Index. The web crawler MCP server includes search sources such as web search, database search, and file browser. The data analysis MCP server includes models or tools such as DCF Model (Discounted Cash Flow Model), SWOT tool, and ML model (Machine Learning Model).

[0062] It should be noted that this example is only for the purpose of assisting in understanding this application and does not constitute a limitation on the financial research report generation method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0063] This application also provides a financial research report generation system, which includes at least a scheduling layer, an execution layer, and an output layer connected in sequence. The scheduling layer includes a task planning agent, which receives the research report generation task, decomposes the research report generation task into multiple sub-tasks, and sends each sub-task to the execution layer. The execution layer includes at least one execution agent, which is used to execute the sub-tasks and obtain the corresponding task execution results; The output layer includes a report generation agent, which generates target financial research reports based on the task execution results output by each execution agent.

[0064] Specifically, the execution agents in the execution layer include at least a deep research agent, a browser agent, and a deep analysis agent; the subtasks include at least a deep research subtask, a browser subtask, and a deep analysis subtask; and the task execution results include at least a list of target data sources, target structured data, and data analysis results.

[0065] For example, the end-to-end workflow of a financial research report generation system may include the following stages: 1. Request reception stage: The Planning Agent receives user research report generation tasks; 2. Task decomposition stage: The task is decomposed into specific execution tasks according to the report type; 3. Data acquisition stage: The Deep Researcher Agent filters information sources, and the Browser Agent acquires real-time data; 4. Deep analysis stage: The DeepAnalyze Agent performs professional analysis and modeling; 5. Report generation stage: The Final Report Agent integrates the results and generates the final target financial research report; 6. Quality control stage: Multiple rounds of quality assessment and necessary revisions are performed.

[0066] The composition structure of the financial research report generation system provided in this application embodiment can be referred to... Figure 2 and / or Figure 4 The connection methods and functional details of various intelligent agents in the financial research report generation system can be found in the embodiments of the aforementioned financial research report generation method, and will not be repeated here.

[0067] The financial research report generation system provided in this application adopts the financial research report generation method in the above embodiments, which can improve the generation efficiency and professional quality of financial research reports through the cooperation and division of labor of multiple intelligent agents. Compared with the prior art, the beneficial effects of the financial research report generation system provided in this application are the same as the beneficial effects of the financial research report generation method provided in the above embodiments, and other technical features of this financial research report generation system are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0068] This application also provides a financial research report generation device, which includes at least: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the financial research report generation method in the above embodiments.

[0069] The following is for reference. Figure 9 It shows a schematic diagram of a structure suitable for implementing the financial research report generation device of the present application embodiments. Figure 9 The financial research report generation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0070] like Figure 9As shown, the financial research report generation device may include a processing unit 101 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 102 or a program loaded from a storage device 103 into a random access memory (RAM) 104. The RAM 104 also stores various programs and data required for the operation of the financial research report generation device. The processing unit 101, ROM 102, and RAM 104 are interconnected via a bus 105. An input / output (I / O) interface 106 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 106: input devices 107 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 108 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 103 including, for example, magnetic tape, hard disk, etc.; and communication devices 109. Communication device 109 allows the financial report generation device to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows a financial report generation device with various systems, it should be understood that implementing or having all of the systems shown is not required. More or fewer systems may be implemented alternatively.

[0071] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 103, or installed from ROM 102. When the computer program is executed by processing device 101, it performs the functions defined in the methods of the embodiments of this application.

[0072] The financial research report generation device provided in this application, employing the financial research report generation method described in the above embodiments, can improve the generation efficiency and professional quality of financial research reports through the collaboration and division of labor among multiple intelligent agents. Compared with the prior art, the beneficial effects of the financial research report generation device provided in this application are the same as those of the financial research report generation method provided in the above embodiments, and other technical features in this financial research report generation device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0073] It should be understood that various parts of the embodiments of this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0074] The above description is merely a specific implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of the embodiments of this application. Therefore, the protection scope of the embodiments of this application should be determined by the protection scope of the above claims.

[0075] This application also provides a computer-readable storage medium storing a computer program that can run on a processor. The computer program is used to execute the financial research report generation method in the above embodiments.

[0076] The computer-readable storage medium provided in this application embodiment may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0077] The aforementioned computer-readable storage medium may be included in the financial research report generation device; or it may exist independently and not be assembled into the financial research report generation device.

[0078] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the financial research report generation device, the financial research report generation device: receives a research report generation task through a task planning agent in the scheduling layer, decomposes the research report generation task into multiple sub-tasks, and sends each sub-task to the execution layer; each agent in the execution layer is responsible for processing its corresponding sub-task and generating the corresponding task execution result; and the report generation agent in the output layer generates the target financial research report based on the task execution results output by each execution agent.

[0079] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming 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 (e.g., via the Internet using an Internet service provider).

[0080] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0081] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0082] The computer-readable storage medium provided in this application embodiment stores computer-readable program instructions for executing the above-described financial research report generation method. Through the cooperation and division of labor among multiple agents, it can improve the generation efficiency and professional quality of financial research reports. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application embodiment are the same as the beneficial effects of the financial research report generation method provided in the above embodiments, and will not be repeated here.

[0083] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the financial research report generation method described above.

[0084] The computer program product provided in this application embodiment can improve the efficiency and professional quality of financial research report generation through the cooperation and division of labor among multiple intelligent agents. Compared with the prior art, the beneficial effects of the computer program product provided in this application embodiment are the same as the beneficial effects of the financial research report generation method provided in the above embodiments, and will not be repeated here.

[0085] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A method for generating financial research reports, characterized in that, The financial research report generation method is applied to a financial research report generation system, which includes a scheduling layer, an execution layer, and an output layer connected in sequence. The financial research report generation method includes: The task planning agent in the scheduling layer receives the research report generation task, decomposes the research report generation task into multiple sub-tasks, and sends each sub-task to the execution layer. Each agent in the execution layer is responsible for processing its corresponding sub-task and generating the corresponding task execution result; The report generation agent in the output layer generates the target financial research report based on the task execution results output by each of the execution agents.

2. The financial research report generation method as described in claim 1, characterized in that, The task planning agent is also used to send preset tool invocation rules, parameter passing mechanisms, and result processing procedures to each of the execution agents. The task planning agent is also used to monitor the task execution status of each execution agent, and to control the execution agent to retry the task when there is an abnormality in the task execution status of the execution agent.

3. The financial research report generation method as described in claim 1, characterized in that, The execution agent in the execution layer includes at least a deep research agent, a browser agent, and a deep analysis agent; the subtask includes at least a deep research subtask, a browser subtask, and a deep analysis subtask; and the task execution result includes at least a list of target data sources, target structured data, and data analysis results. The step of having each agent in the execution layer handle its corresponding subtask and generate the corresponding task execution result includes: The deep research agent determines a list of target data sources based on the research topics in the deep research sub-tasks; The browser agent obtains the target structured data corresponding to the browser subtask based on the target data source list; The deep analysis agent executes the deep analysis sub-tasks based on the target structured data to generate data analysis results.

4. The financial research report generation method as described in claim 3, characterized in that, The step of determining the target data source list based on the research report topics in the in-depth research sub-task includes: Based on the research report topics in the in-depth research sub-task, generate multiple corresponding search operation steps, and execute the search operation corresponding to the first search operation step to obtain the corresponding search results; Analyze the search results to determine if a re-search is needed; If so, return to execute the search operation corresponding to the first search operation step; If not, then execute the search operation corresponding to the next search operation step to obtain the corresponding search results; After all search operations are completed, a list of data sources in structured data format is generated based on the search results. Determine whether the list of data sources in the structured data format meets the preset requirements; If the preset requirements are met, the list of data sources in the form of structured data will be determined as the target data source list and returned to the browser agent; If the preset requirements are not met, return to the first search operation step and execute the search operation corresponding to it.

5. The financial research report generation method as described in claim 3, characterized in that, The step of obtaining the target structured data corresponding to the browser subtask based on the target data source list includes: Receive the browser subtask, execute the browser subtask based on the target data source list, and obtain the corresponding data content; Evaluate the data content to determine whether re-execution is necessary; If so, return to the execution steps: execute the browser subtask based on the target data source list to obtain the corresponding data content; If not, the data content will be extracted and archived to generate structured data; Determine whether the structured data meets the preset requirements; If the preset requirements are met, the structured data is identified as the target structured data and returned to the deep analysis agent; If the requirements are not met, return to the execution step: execute the browser subtask based on the target data source list to obtain the corresponding data content.

6. The financial research report generation method as described in claim 3, characterized in that, The step of executing the deep analysis subtask based on the target structured data and generating data analysis results includes: Generate a data analysis plan corresponding to the browser subtask based on the target structured data, execute the operation of the first data analysis step in the data analysis plan, and obtain the corresponding analysis results; Evaluate the analysis results to determine whether a reanalysis is necessary; If so, return to the operation of executing the first data analysis step in the data analysis plan; If not, continue with the next data analysis step in the data analysis plan to obtain the corresponding analysis results; After all data analysis steps are completed, structured analysis data is generated based on the analysis results. Determine whether the structured analysis data meets preset requirements; If the preset requirements are met, the structured analysis data will be identified as the data analysis result and returned to the report generation agent. If the requirements are not met, return to the operation of the first data analysis step in the data analysis plan.

7. The financial research report generation method as described in any one of claims 1 to 6, characterized in that, The report generating agent is used to integrate the task execution results after each execution agent in the execution layer outputs the task execution results, and render the integrated task execution results based on a preset report format to obtain a target financial research report. The target financial research report includes at least charts, tables and annotations.

8. A financial research report generation system, characterized in that, The financial research report generation system includes at least a scheduling layer, an execution layer, and an output layer connected in sequence. The scheduling layer includes a task planning agent, which is used to receive a research report generation task, decompose the research report generation task into multiple sub-tasks, and send each of the sub-tasks to the execution layer. The execution layer includes at least one execution agent, which is used to execute the sub-task and obtain the corresponding task execution result; The output layer includes a report generation agent, which generates a target financial research report based on the task execution results output by each of the execution agents.

9. A financial research report generation device, characterized in that, The financial research report generation device includes at least: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the financial research report generation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for implementing a financial research report generation method, which is executed by a processor to implement the steps of the financial research report generation method as described in any one of claims 1 to 7.

Images