Multi-agent automatic analysis method, system, device and apparatus

Abstract

The application provides a multi-agent automatic analysis method, system, device and storage medium, the method comprises the following steps: according to the input question and the complex report engine industry conversion decomposition into multiple analysis indexes, and the analysis indexes of conversion decomposition are respectively input into the configuration template of the multi-agent response system; each analysis index in the configuration template is disassembled into the corresponding task list; after the task list disassembled by each analysis index is sequentially executed, the task list result is output to the corresponding configuration template according to the preset standard output format, and the index conclusion of the analysis index is obtained; the index conclusions corresponding to each analysis index in the configuration template are integrated, and then the answer corresponding to the input question is output by the preset output template. The application can automatically analyze the numerous table data in the complex report engine and quickly output a relatively accurate and comprehensive problem conclusion.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically to a method, system, device, and apparatus for automatic analysis of multiple agents. Background Technology

[0002] A complex reporting engine, comprising forms (actual business data) and custom reports, can automatically generate reports with complex structures and formats based on predefined rules, templates, and data sources. Custom reports feature form aggregation, report association, multi-sheet association, and multi-region association. As enterprises grow and their businesses diversify, they accumulate vast amounts of data. This data is stored in different systems and databases, such as Customer Relationship Management (CRM) systems and Enterprise Resource Planning (ERP) systems. To better manage and utilize this data, enterprises need a tool to integrate this scattered data and present it in an intuitive and complex reporting format. For example, a large manufacturing enterprise stores its production data in a production management system, sales data in a sales system, and financial data in a financial system. To comprehensively understand the enterprise's operations, such as analyzing the production cost, sales profit, and inventory turnover rate of each product, a complex reporting engine is needed to integrate this data and generate comprehensive reports. A complex reporting engine is equivalent to Excel with Elastic Search (a distributed search engine) capabilities.

[0003] Reporting systems typically display key metrics that managers care about (e.g., performance summary tables, employee productivity statistics, sales funnel conversion statistics, etc.). Managers can quickly and directly understand these key metrics and even see comparisons (e.g., year-on-year / month-on-month). However, as the number of reports increases, managers often find it overwhelming to review them all. This is especially true in complex reporting engines, where reports are often intricate, and analyzing a single metric may involve multiple reports, multiple sheets within an Excel spreadsheet, and multiple data forms. Managers struggle to quickly find the answers they need. Furthermore, for some reports, identifying the key factors causing changes in key metrics requires understanding the relationships between the report data, which is time-consuming for users. Therefore, the urgent problem to be solved in the field of complex reporting engines is how to automatically analyze numerous tables to quickly obtain comprehensive answers to questions input by users (such as managers and business owners). Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a multi-agent automatic analysis method, system, device, and apparatus that can automatically analyze numerous tables and quickly output relatively comprehensive problem conclusions.

[0005] The technical solution of this invention is implemented as follows:

[0006] On the one hand, the present invention provides an automatic multi-agent analysis method, comprising the following steps:

[0007] Based on the input question and the industry to which the complex report engine belongs, the analysis indicators are decomposed into multiple analytical indicators, and the decomposed analytical indicators are input into the configuration template of the multi-agent response system.

[0008] Each analysis indicator in the configuration template is broken down into a corresponding task list, which includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks.

[0009] After executing the tasks in sequence according to the task list decomposed from each analysis indicator, the results of each task list are output to the corresponding configuration template according to the preset standard output format to obtain the indicator conclusion of that analysis indicator.

[0010] The system integrates the conclusions of each analysis indicator in the configuration template and then outputs the answer corresponding to the input question using a preset output template.

[0011] Preferably, the step of breaking down each analysis indicator in the configuration template into a corresponding task list includes:

[0012] Determine the indicator type of the analysis indicators in the configuration template. The indicator type includes, but is not limited to, standard causal type, custom causal type, standard prediction type, and custom prediction type.

[0013] The indicator analysis process is determined based on the identified indicator type, and the corresponding task list is decomposed based on the preset task list of the identified indicator analysis process; the task list includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks.

[0014] Preferably, the step of sequentially executing tasks according to the task list derived from each analysis indicator, and outputting the results of each task list to the corresponding configuration template according to a preset standard output format to obtain the indicator conclusion includes:

[0015] Based on the extracted task list, corresponding category prompts are constructed sequentially based on the task list questions and the output history of the current task. The data intelligence agent determines the programming task type of the task list based on the category prompts.

[0016] Based on the task code requirements of the programming task type, the corresponding tools are indexed and combined with historical memory information to construct a coding prompt. The Python agent then writes code based on the coding prompt.

[0017] Based on the task code requirements of the programming task type and common problems of indexing tools, a code verification prompt is constructed. The verification agent verifies the code written by the Python agent according to the code verification prompt.

[0018] The Python code executor executes the code that verifies the agent's credentials.

[0019] Based on the formatting requirements of the corresponding programming task type and the code execution results of the Python code executor agent, a formatting prompt is constructed, and the output agent outputs the corresponding task list results in a structured manner based on the formatting prompt.

[0020] The extracted task list is processed through the above steps in sequence to obtain the results corresponding to each task list. These results are then output to the corresponding configuration template according to the preset standard output format to obtain the indicator conclusion of the analysis indicator.

[0021] Preferably, the step of the data intelligence agent determining the programming task type of the task list based on the classification prompt is as follows:

[0022] The programming task type includes node function category information, node query information, node change calculation information, and node factor evaluation information. A coding prompt is constructed based on the relevant information in the task list.

[0023] On the other hand, the present invention also provides a multi-agent automatic analysis system, including

[0024] The analysis indicator conversion module is used to decompose the input question and the industry to which the complex report engine belongs into multiple analysis indicators, and input the decomposed analysis indicators into the configuration template of the multi-agent response system respectively;

[0025] The task list decomposition module is used to decompose each analysis indicator in the configuration template into a corresponding task list. The task list includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks.

[0026] The indicator conclusion output module is used to execute tasks sequentially according to the task list decomposed for each analysis indicator, and then output the results of each task list to the corresponding configuration template according to the preset standard output format to obtain the indicator conclusion of the analysis indicator.

[0027] The standard answer output module is used to integrate the conclusions of each analysis indicator in the configuration template, and then output the answer corresponding to the input question using a preset output template.

[0028] Preferably, the task list breakdown module includes:

[0029] The indicator type determination unit is used to determine the indicator type of the analysis indicators in the configuration template. The indicator type includes, but is not limited to, standard causal type, customized causal type, standard prediction type and customized prediction type.

[0030] The task list decomposition unit is used to determine the indicator analysis process of the analysis indicator based on the determined indicator type, and to decompose the corresponding task list according to the preset task list of the determined indicator analysis process; the task list includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks and formatted output tasks.

[0031] Preferably, the indicator conclusion output module includes:

[0032] The data intelligence agent processing unit is used to construct corresponding classification prompts based on the task list issues and the output history of the current task, according to the split task list. The data intelligence agent determines the programming task type of the task list based on the classification prompts.

[0033] The Python agent processing unit is used to index the corresponding tools and construct a coding prompt based on the task code requirements of the programming task type and historical memory information. The Python agent writes code according to the coding prompt.

[0034] The agent processing unit is used to construct a code verification prompt based on the task code requirements of the programming task type and common problems of indexing tools, and the agent verifies the code written by the Python agent according to the code verification prompt.

[0035] The executor unit is used to execute the code verified by the agent through the Python code executor.

[0036] The output agent processing unit is used to construct a formatted prompt based on the formatting requirements of the corresponding programming task type and the code execution result of the Python code executor agent. The output agent then performs structured output of the corresponding task list result based on the formatted prompt.

[0037] The indicator conclusion output unit is used to sequentially repeat the above steps to obtain the results corresponding to each task list from the split task list, and then output these results to the corresponding configuration template according to the preset standard output format, so as to obtain the indicator conclusion of the analysis indicator.

[0038] Preferably, in the data intelligence agent processing unit, the programming task type includes node function category information, node query information, node change calculation information, and node factor evaluation information, and a coding prompt is constructed based on the relevant information in the task list.

[0039] In another aspect, the present invention also provides a computer electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the multi-agent automatic analysis method described above.

[0040] In another aspect, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the multi-agent automatic analysis method described above.

[0041] Compared with existing technologies, the present invention has the following advantages: The present invention analyzes complex user-input questions by breaking them down into multiple analytical indicators, and then further breaks down each indicator into a series of task lists, which are then executed separately to output the results of the corresponding task lists. This allows for a comprehensive and accurate answer to the user's input questions. Moreover, the present invention outputs the results of each task list to the corresponding configuration template according to a preset standard output format, and the conclusions of each analytical indicator are integrated and output in a preset output template to provide the answer corresponding to the input question. This makes it easy for users to quickly obtain the information they want from numerous reports. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of the multi-agent automatic analysis method of the present invention;

[0044] Figure 2 This is a structural block diagram of the multi-agent automatic analysis system of the present invention;

[0045] Figure 3 This is a block diagram of the multi-agent architecture;

[0046] Figure 4 This is a structural block diagram of a computer electronic device according to the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] See Figure 1 This invention discloses an automatic multi-agent analysis method, comprising the following steps:

[0049] S1, based on the input question and the industry of the complex report engine, decompose it into multiple analytical indicators, and input the decomposed analytical indicators into the configuration template of the multi-agent response system respectively;

[0050] S2, decompose each analysis indicator in the configuration template into a corresponding task list, the task list including but not limited to classification tasks, coding tasks, verification tasks, code execution tasks and formatted output tasks;

[0051] S3, after executing the tasks in sequence according to the task list decomposed from each analysis indicator, output the results of each task list to the corresponding configuration template according to the preset standard output format, and obtain the indicator conclusion of the analysis indicator.

[0052] S4 integrates the conclusions of each analysis indicator in the configuration template, and then outputs the answer corresponding to the input question using the preset output template.

[0053] Correspondingly, see Figure 2 The present invention also discloses a multi-agent automatic analysis system, including...

[0054] The analysis indicator conversion module is used to decompose the input question and the industry to which the complex report engine belongs into multiple analysis indicators, and input the decomposed analysis indicators into the configuration template of the multi-agent response system respectively;

[0055] The task list decomposition module is used to decompose each analysis indicator in the configuration template into a corresponding task list. The task list includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks.

[0056] The indicator conclusion output module is used to execute tasks sequentially according to the task list decomposed for each analysis indicator, and then output the results of each task list to the corresponding configuration template according to the preset standard output format to obtain the indicator conclusion of the analysis indicator.

[0057] The standard answer output module is used to integrate the conclusions of each analysis indicator in the configuration template, and then output the answer corresponding to the input question using a preset output template.

[0058] In this embodiment, the multi-agent automatic analysis method uses a multi-agent automatic analysis system as the execution object of the steps. Specifically, step S1 uses the analysis index transformation module as the execution object, step S2 uses the task list decomposition module as the execution object, step S3 uses the index conclusion output module as the execution object, and step S4 uses the standard answer output module as the execution object.

[0059] A complex reporting engine, comprising forms (actual business data) and custom reports. Custom reports feature form aggregation, report association, multi-sheet association, and multi-region association. The definitions of form aggregation, report association, multi-sheet association, and multi-region association are as follows:

[0060] 1) Form aggregation: Various statistical aggregations based on original forms (database level);

[0061] 2) Report Linking: Linking and referencing data from multiple analytical reports;

[0062] 3) Multi-sheet association: Linking and referencing data across multiple sheets in a single report;

[0063] 4) Multi-region linking: Data linking and referencing across multiple regions within a single sheet;

[0064] Simply put, a complex reporting engine is essentially an Excel file with an embedded database application.

[0065] The structural block diagram of the multi-agent response system is as follows: Figure 3 As shown, when performing specific single-indicator analysis, data from a complex reporting engine is used as the basis.

[0066] In step S1, since the complex reporting engines built by different industries will differ, the analytical indicators decomposed based on the relevant industry characteristics of the complex reporting engine and the user's input question in this embodiment of the invention will also differ. For example, the focus of analysis is completely different for supermarket performance decline and manufacturing performance decline, which may involve various different factors such as stores, people, products, and channels. The content of the complex reporting engine will also differ. Therefore, it is necessary to decompose the complex reporting engine into multiple analytical indicators according to the industry to which it belongs and the user's input question during the specific implementation process. The analytical indicators decomposed from different industries will differ. These multiple analytical indicators will involve multiple aspects of factors. For example, when a user asks "Why did the performance decline this month?", step 1 will divide the question into multiple analytical indicators, such as performance indicators, sales type indicators, sales process indicators, etc. These indicators involve different factors in various aspects. Only by analyzing the analytical indicators of different aspects of factors can a comprehensive and accurate answer be given to the input question.

[0067] In this embodiment of the invention, corresponding analytical indicators can be pre-set according to the complex reporting engines of different industries. When a user asks a question, the specific analytical indicator is determined from the preset analytical indicators based on the question asked. This can be specifically determined through the constructed model.

[0068] Furthermore, this invention inputs these analytical indicators into the configuration template, and the corresponding conclusions of each analytical indicator are ultimately filled back into the configuration template, thereby facilitating the formation of a standard output result so that users can quickly obtain the information they want.

[0069] In step 2, each analysis indicator in the configuration template also needs to be broken down into a series of task lists for task execution. The output results after execution form the analysis conclusion corresponding to that analysis indicator.

[0070] Specifically, in step 2, the step of breaking down each analysis indicator in the configuration template into a corresponding task list includes:

[0071] S201, determine the indicator type of the analysis indicators in the configuration template, wherein the indicator type includes, but is not limited to, standard causal type, customized causal type, standard prediction type and customized prediction type;

[0072] S202, determine the indicator analysis process for the determined indicator type, and decompose the corresponding task list according to the preset task list of the determined indicator analysis process; the task list includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks and formatted output tasks.

[0073] Correspondingly, in a multi-agent automatic analysis system, the task list decomposition module includes:

[0074] The indicator type determination unit is used to determine the indicator type of the analysis indicators in the configuration template. The indicator type includes, but is not limited to, standard causal type, customized causal type, standard prediction type and customized prediction type.

[0075] The task list decomposition unit is used to determine the indicator analysis process of the analysis indicator based on the determined indicator type, and to decompose the corresponding task list according to the preset task list of the determined indicator analysis process; the task list includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks and formatted output tasks.

[0076] Step 2 in the multi-agent automatic analysis method uses the task list decomposition module in the multi-agent automatic analysis system as the execution object. Specifically, step S201 uses the indicator type determination unit as the execution object, and step S202 uses the task list decomposition unit as the execution object.

[0077] In step S201, after breaking down the problem into multiple analytical indicators, it is also necessary to determine whether each indicator type is a standard causal type, a custom causal type, a standard prediction type, a custom prediction type, or another type. For example, taking the performance indicator broken down from "Why did performance decline this month?" as an example, to analyze this month's performance, this is a result that can be directly queried in the reporting engine, therefore it belongs to the standard causal type. Similarly, the indicator type of other analytical indicators can also be determined based on the constructed model, implementation method, or pre-set parameters.

[0078] In step S202, since the standard causal type, customized causal type, standard prediction type, customized prediction type, or other types have all had their corresponding indicator analysis processes pre-set in the system, after determining the indicator type, a corresponding task list will be decomposed according to the preset task list of the corresponding indicator analysis process. Taking performance indicators as an example, to analyze August's performance, it's necessary to know which function to use, which analysis method to employ, which tools to use, and how the results should be presented. Therefore, this analysis indicator needs to be decomposed into a corresponding task list for analysis. The task list includes, but is not limited to, classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks.

[0079] Specifically, in step S3, the step of sequentially executing tasks based on the task list derived from each analysis indicator, and outputting the results of each task list to the corresponding configuration template according to a preset standard output format to obtain the indicator conclusion includes:

[0080] S301, based on the split task list, construct corresponding classification prompts based on the task list questions and the output history of the current task, and the data intelligence agent determines the programming task type of the task list based on the classification prompts;

[0081] S302, according to the task code requirements of the programming task type, index the corresponding tools and combine them with historical memory information to construct a coding prompt, and the Python agent writes code according to the coding prompt;

[0082] S303, a code verification prompt is constructed based on the task code requirements of the programming task type and common problems of indexing tools, and the verification agent verifies the code written by the Python agent according to the code verification prompt;

[0083] S304, the Python code executor agent executes the code that verifies the agent's verification;

[0084] S305, construct a formatting prompt based on the formatting requirements of the corresponding programming task type and the code execution result of the Python code executor agent, and output the corresponding task list result based on the formatting prompt;

[0085] S306, the above steps are repeated sequentially to obtain the results corresponding to each task list. Then, these results are output to the corresponding configuration template according to the preset standard output format to finally obtain the indicator conclusion of the analysis indicator.

[0086] Correspondingly, in a multi-agent automated analysis system, the indicator conclusion output module includes:

[0087] The data intelligence agent processing unit is used to construct corresponding classification prompts based on the task list issues and the output history of the current task, according to the split task list. The data intelligence agent determines the programming task type of the task list based on the classification prompts.

[0088] The Python agent processing unit is used to index the corresponding tools and construct a coding prompt based on the task code requirements of the programming task type and historical memory information. The Python agent writes code according to the coding prompt.

[0089] The agent processing unit is used to construct a code verification prompt based on the task code requirements of the programming task type and common problems of indexing tools, and the agent verifies the code written by the Python agent according to the code verification prompt.

[0090] The executor unit is used to execute the code verified by the agent through the Python code executor.

[0091] The output agent processing unit is used to construct a formatted prompt based on the formatting requirements of the corresponding programming task type and the code execution result of the Python code executor agent. The output agent then performs structured output of the corresponding task list result based on the formatted prompt.

[0092] The indicator conclusion output unit is used to sequentially repeat the above steps to obtain the results corresponding to each task list from the split task list, and then output these results to the corresponding configuration template according to the preset standard output format, so as to obtain the indicator conclusion of the analysis indicator.

[0093] Similarly, in the multi-agent automatic analysis method, step 3 uses the index conclusion output module in the multi-agent automatic analysis system as the execution object. Specifically, step S301 uses the data agent processing unit as the execution object, step S302 uses the Python agent processing unit as the execution object, step S303 uses the verification agent processing unit as the execution object, step S304 uses the executor agent execution unit as the execution object, step S305 uses the output agent processing unit as the execution object, and step S306 uses the index conclusion output unit as the execution object.

[0094] In this embodiment, after the corresponding task list is extracted, a corresponding classification prompt, coding prompt, code verification prompt, and formatting prompt are constructed for each item in the task list. This allows the data agent to determine the programming task type of the task list based on the classification prompt. Once the programming task type is determined, the Python agent knows which functions to use for coding. Therefore, the Python agent then writes code based on the prompts provided in the coding prompt. The written code is verified by the verification agent based on the code verification prompt, and then executed by the Python code executor agent. Finally, the output agent outputs the structured task list results to the configuration template based on the formatting prompt. In this embodiment, the extracted task lists are iterated through steps S301-S305 to obtain the results corresponding to each task list. These results are then backfilled into the configuration template as required, forming the indicator conclusion for this analysis metric.

[0095] In step S4, after obtaining the conclusions of all analysis indicators, the conclusions of all indicators are integrated, and finally the answers to the user's questions are output using a preset output template, so that the user can quickly obtain the relevant information for their questions from a large number of reports.

[0096] Specifically, in step S301, the data agent determines the programming task type of the task list based on the classification prompt:

[0097] The programming task type includes node function category information, node query information, node change calculation information, and node factor evaluation information. A coding prompt is constructed based on the relevant information in the task list.

[0098] Correspondingly, in the data intelligence agent processing unit, the programming task type includes node function category information, node query information, node change calculation information, and node factor evaluation information, and a coding prompt is constructed based on the relevant information in the task list.

[0099] In this embodiment, the programming task type includes information such as node function category information, node query information, node change calculation information, and node factor evaluation information, so as to guide the Python agent in writing code.

[0100] In summary, this invention provides a comprehensive and accurate answer to user-inputted questions by breaking down complex questions into multiple analytical indicators. Each indicator is further broken down into a series of task lists, which are then executed to output the results of the corresponding task lists. Furthermore, this invention outputs the results of each task list to a corresponding configuration template according to a preset standard output format. The conclusions of each analytical indicator are also integrated and output using a preset output template, providing the answer to the input question. This allows users to quickly obtain the information they need from numerous reports.

[0101] In another aspect, the present invention also provides a computer electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the multi-agent automatic analysis method described above.

[0102] In another aspect, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the multi-agent automatic analysis method described above.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-agent automatic analysis method applied to a complex report engine, characterized in that, Includes the following steps: Based on the input question and the industry to which the complex report engine belongs, the analysis indicators are decomposed into multiple analytical indicators, and the decomposed analytical indicators are input into the configuration template of the multi-agent response system. Each analysis indicator in the configuration template is broken down into a corresponding task list, which includes classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks. After executing the tasks in sequence according to the task list decomposed from each analysis indicator, the results of each task list are output to the corresponding configuration template according to the preset standard output format to obtain the indicator conclusion of that analysis indicator. The conclusions corresponding to each analysis indicator in the configuration template are integrated, and then the answers corresponding to the input questions are output using the preset output template. The step of breaking down each analysis indicator in the configuration template into a corresponding task list includes: Determine the indicator type of the analysis indicators in the configuration template. The indicator types include standard causal type, custom causal type, standard prediction type, and custom prediction type. The indicator analysis process for the determined indicator is determined based on the determined indicator type, and the corresponding task list is decomposed based on the preset task list of the determined indicator analysis process; the task list includes classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks. Based on the extracted task list, corresponding category prompts are constructed sequentially based on the task list questions and the output history of the current task. The data intelligence agent determines the programming task type of the task list based on the category prompts. Based on the task code requirements of the programming task type, the corresponding tools are indexed and combined with historical memory information to construct a coding prompt. The Python agent then writes code based on the coding prompt. Based on the task code requirements of the programming task type and common problems of indexing tools, a code verification prompt is constructed. The verification agent verifies the code written by the Python agent according to the code verification prompt. The Python code executor executes the code that verifies the agent's credentials. Based on the formatting requirements of the corresponding programming task type and the code execution results of the Python code executor agent, a formatting prompt is constructed, and the output agent outputs the corresponding task list results in a structured manner based on the formatting prompt. The extracted task list is processed through the above steps in sequence to obtain the results corresponding to each task list. These results are then output to the corresponding configuration template according to the preset standard output format to obtain the indicator conclusion of the analysis indicator.

2. The method of claim 1, wherein, The steps by which the data intelligence agent determines the programming task type of the task list based on the classification prompt are as follows: The programming task type includes node function category information, node query information, node change calculation information, and node factor evaluation information. A coding prompt is constructed based on the relevant information in the task list.

3. A multi-agent automatic analysis system applied to a complex report engine, characterized in that, include: The analysis indicator conversion module is used to decompose the input question and the industry to which the complex report engine belongs into multiple analysis indicators, and input the decomposed analysis indicators into the configuration template of the multi-agent response system respectively; The task list decomposition module is used to decompose each analysis indicator in the configuration template into a corresponding task list. The task list includes classification tasks, coding tasks, verification tasks, code execution tasks, and formatted output tasks. The indicator conclusion output module is used to execute tasks sequentially according to the task list decomposed for each analysis indicator, and then output the results of each task list to the corresponding configuration template according to the preset standard output format to obtain the indicator conclusion of the analysis indicator. The standard answer output module is used to integrate the indicator conclusions corresponding to each analysis indicator in the configuration template, and then output the answer corresponding to the input question using a preset output template; The task list breakdown module includes: The indicator type determination unit is used to determine the indicator type of the analysis indicators in the configuration template. The indicator types include standard causal type, customized causal type, standard prediction type and customized prediction type. The task list decomposition unit is used to determine the indicator analysis process of the analysis indicator based on the determined indicator type, and to decompose the corresponding task list according to the preset task list of the determined indicator analysis process; the task list includes classification tasks, coding tasks, verification tasks, code execution tasks and formatted output tasks. The indicator conclusion output module includes: The data intelligence agent processing unit is used to construct corresponding classification prompts based on the task list issues and the output history of the current task, according to the split task list. The data intelligence agent determines the programming task type of the task list based on the classification prompts. The Python agent processing unit is used to index the corresponding tools and construct a coding prompt based on the task code requirements of the programming task type and historical memory information. The Python agent writes code according to the coding prompt. The agent processing unit is used to construct a code verification prompt based on the task code requirements of the programming task type and common problems of indexing tools, and the agent verifies the code written by the Python agent according to the code verification prompt. The executor unit is used to execute the code verified by the agent through the Python code executor. The output agent processing unit is used to construct a formatted prompt based on the formatting requirements of the corresponding programming task type and the code execution result of the Python code executor agent. The output agent then outputs the corresponding task list result in a structured manner based on the formatted prompt. The indicator conclusion output unit is used to sequentially repeat the above steps to obtain the results corresponding to each task list from the split task list, and then output these results to the corresponding configuration template according to the preset standard output format, so as to obtain the indicator conclusion of the analysis indicator.

4. The multi-agent automatic analysis system of claim 3, wherein, In the data intelligence agent processing unit, the programming task type includes node function category information, node query information, node change calculation information, and node factor evaluation information. Based on the relevant information in the task list, a coding prompt is constructed.

5. A computer electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the multi-agent automatic analysis method according to any one of claims 1-2.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the multi-agent automatic analysis method according to any one of claims 1-2.

Images