Report generation method and system based on rule engine and large model cooperation

Abstract

This invention discloses a report generation method and system based on the collaboration of a rule engine and a large model, belonging to the field of information processing technology. Its main purpose is to solve the problem of low report output efficiency caused by reliance on manual report processing. The method mainly includes: responding to a report generation request, invoking a target dialogue agent matching the target business operation type to perform a report request interaction and obtain report request interaction information; decomposing the report generation task based on the report request interaction information to obtain multiple report generation sub-tasks and target report templates covering the corresponding report component modules of each sub-task; invoking a target generation agent matching the report generation sub-task under the target business operation type, and generating the report content of each report component module through the target generation agent; integrating the report content based on the target report templates, generating and outputting the target report in response to the report generation request. This method is mainly used for report generation.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a report generation method and system based on the collaboration of a rule engine and a large model. Background Technology

[0002] In the field of asset custody and operational risk control, generating accurate, timely, and insightful risk reports amidst massive amounts of heterogeneous business data from multiple sources, including clearing, accounting, and valuation, is a core and demanding task. These reports are key deliverables for fulfilling fiduciary responsibilities and meeting regulatory compliance and internal control requirements. Current industry practice indicates that report generation efficiency and decision support levels have become key indicators for measuring the operational capabilities and modernization level of risk management of custodian institutions.

[0003] Currently, generating such reports mainly relies on manual processing. This manual approach requires experts to manually extract, verify, and integrate data from various scattered business subsystems, and then write and analyze the reports based on their personal experience. This method suffers from high labor costs and poor timeliness. Summary of the Invention

[0004] In view of this, the present invention provides a report generation method and system based on the collaboration of a rule engine and a large model, the main purpose of which is to solve the problem of low report output efficiency caused by the reliance on manual report processing.

[0005] According to one aspect of the present invention, a report generation method based on the collaboration of a rule engine and a large model is provided, comprising: In response to a report generation request, the target dialogue agent matching the target business operation type is invoked to perform the report request interaction and obtain the report request interaction information; Based on the report requirement interaction information, the report generation task is decomposed to obtain multiple report generation sub-tasks, and a target report template covering the report component modules corresponding to each of the report generation sub-tasks; Under the target business operation type, the target generating agent of the report generation subtask is matched, and the report content of each of the report component modules is generated by the target generating agent. The report content is integrated based on the target report template to generate and output the target report in response to the report generation request.

[0006] Further, before invoking the target dialogue agent matching the target business operation type to perform the report request interaction and obtain the report request interaction information, the method includes: For any dialogue agent, construct multiple versions of user guidance words that match the business operation type corresponding to the dialogue agent; Using the report generation task decomposition and evaluation parameters as the comparison object, the user guidance words of different versions are compared and tested, and the user guidance words are iteratively updated according to the comparison and test results to obtain the latest version of the user guidance words, so as to assist the dialogue agent in interacting with the report request through the latest version of the user guidance words; For any generating agent, construct multiple versions of basic prompt words that match the report generation subtask corresponding to the generating agent, wherein the basic prompt words include role prompt words and task goal prompt words; Using the report generation evaluation parameters as a comparison object, the basic prompt words of different versions are compared and tested. Based on the comparison and test results, the basic prompt words are iteratively updated to obtain the latest version of the basic prompt words, so as to assist the execution of the report generation subtask.

[0007] Furthermore, after generating the target report, the method also includes: The report generates evaluation parameters that include structure hit rate, content hit rate, and compliance rate; If the structure hit rate is less than the preset structure hit rate threshold, then the target dialogue agent is retrained and / or the user guidance words of the target dialogue agent are iteratively updated through comparison tests. If the content hit rate is less than the preset content hit rate threshold, the target generated agent will be retrained and / or the basic prompt words of the target generated agent will be iteratively updated through comparison testing. If the compliance rate is less than a preset compliance rate threshold, the target business rule engine of the target generating agent will be retrained.

[0008] Furthermore, any of the target generation agents includes a business rule engine for matching report generation subtasks and a large model inference engine fine-tuned based on historical data from the report generation subtasks. The process of generating report content by any of the target generating agents includes: The database is queried to retrieve pending business data that matches the target business operation type, and the target business rule engine of the target generating agent is retrieved. The pending business data is calculated using the pre-configured business rules in the target business rule engine to generate a deterministic output containing quantitative parameters and compliance constraints. The process employs enhanced generation techniques to retrieve associated business data from the business knowledge base that is related to the business data to be processed. The deterministic output and the associated business data are fused with the basic prompt words corresponding to the target generating agent to generate structured prompt words; The structured prompts are input into the large model inference engine of the target generating agent to generate report content through the large model inference engine.

[0009] Furthermore, before querying the database for pending business data matching the target business operation type, the method further includes: Receive and parse risk warning messages, pending tasks, and business progress data sent by the business operation monitoring system, and associate the parsed data with different business operation types and store it in the managed operation system database for querying pending business data; The generation process for risk warning messages, to-do items, and business progress data includes: The business operation monitoring system subscribes to real-time business operation process data from the managed business system through a message queue, and calculates monitoring indicators and generates early warning messages within a preset period based on the real-time business operation process data, thereby obtaining risk warning messages, to-do items, and business progress data.

[0010] Furthermore, the step of breaking down the report generation task based on the report requirement interaction information includes: Analyze the report request interaction information to determine the report task; The reporting task is broken down into multiple independent report generation sub-tasks according to the progressive hierarchy of business data statistics, core indicator monitoring, business analysis, and decision-making suggestions. The report component modules corresponding to each of the different report generation subtasks are determined, and the structural deployment of each report component module in the report is determined based on the execution dependencies of each report generation subtask, thus obtaining the target report template; The report consists of a preset business cycle overview module, a core business data statistics module, and a pending business statistics module, as well as a business analysis module, an early warning status module, and a decision suggestion module that match the target business operation type.

[0011] Furthermore, after outputting the target report, the method also includes: The target dialogue agent listens for and parses user feedback information. If the parsing result of the user feedback information triggers any report update condition, the target report is updated according to the update strategy corresponding to the triggered report update condition. The update strategy includes single module content update, report structure update, and full report update.

[0012] According to another aspect of the present invention, a report generation system based on the collaboration of a rule engine and a large model is provided, comprising: The dialogue agent invocation module is used to respond to report generation requests by invoking the target dialogue agent that matches the target business operation type to perform the report request interaction and obtain the report request interaction information. The task decomposition module is used to decompose the report generation task according to the report requirement interaction information to obtain multiple report generation sub-tasks and target report templates covering the report component modules corresponding to each of the report generation sub-tasks. The content generation module is used to call the target generation agent that matches the report generation subtask under the target business operation type, and generate the report content of each of the report component modules through the target generation agent; The content integration module is used to integrate the report content according to the target report template, generate and output the target report in response to the report generation request.

[0013] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform operations corresponding to the above-described report generation method based on rule engine and large model collaboration.

[0014] According to another aspect of the present invention, a terminal is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operations corresponding to the above-described report generation method based on the collaboration between the rule engine and the large model.

[0015] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages: This invention provides a report generation method and system based on the collaboration of a rule engine and a large model. In this embodiment, in response to a report generation request, a target dialogue agent matching the target business operation type is invoked to perform a report request interaction, obtaining report request interaction information. Based on this information, the report generation task is decomposed into multiple report generation sub-tasks, and a target report template covering the corresponding report component modules for each sub-task is generated. The target generation agent matching the report generation sub-task under the target business operation type is invoked, and the report content for each report component module is generated through the target generation agent. The report content is integrated based on the target report templates to generate and output the target report responding to the report generation request. This significantly reduces manual intervention and repetitive work in the report generation process, minimizing report quality fluctuations and time consumption caused by differences in human experience. Simultaneously, the intelligent task decomposition and professional agent scheduling mechanism ensure the comprehensiveness and format standardization of the report, thereby greatly improving report generation efficiency.

[0016] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the specification. Furthermore, in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This diagram illustrates a report generation method based on the collaboration between a rule engine and a large model, as provided in an embodiment of the present invention. Figure 2 This invention provides a flowchart of another report generation method based on the collaboration of a rule engine and a large model, as illustrated in an embodiment of the present invention. Figure 3 This diagram illustrates the interaction between multiple systems in an asset custody business according to an embodiment of the present invention. Figure 4 This diagram illustrates a block diagram of a report generation system based on the collaboration of a rule engine and a large model, as provided in an embodiment of the present invention. Figure 5 A schematic diagram of the structure of a terminal provided in an embodiment of the present invention is shown. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0019] To address the problem of low report output efficiency caused by reliance on manual report processing, this invention provides a report generation method based on the collaboration of a rule engine and a large-scale model. This method utilizes a report generation system based on this collaboration to perform corresponding operations. This system can be deployed within an enterprise system to collect relevant data from different business operations and generate assessment reports for different business operations, including but not limited to asset custody and operation risk control reports. Figure 1 As shown, the method includes: 101. In response to the report generation request, the target dialogue agent matching the target business operation type is invoked to perform the report request interaction and obtain the report request interaction information.

[0020] In this embodiment of the invention, when a user, such as a managed business manager, needs to generate a risk report through the report generation system, the current system first identifies the report generation request initiated by the user. This request can be initiated through natural language input, such as "Please generate a risk warning report for last month's valuation and accounting business," or triggered by interface operations, such as triggering a report generation button. Based on the implicit or specified business operation type in the request, such as liquidation instructions, valuation and accounting, or account management, the system invokes a target dialogue agent matching that type. For example, if the user inputs, "I need a monthly analysis report on the abnormal execution of liquidation instructions," the system identifies the business type as liquidation instructions and then invokes a target dialogue agent specifically trained for that business line. The target dialogue agent simulates expert dialogue, guiding the user to clarify the specific report requirements. For example, through multiple rounds of interactive questioning, such as "Which institutions should the report cover? Should it focus on anomalies in a specific time period? Should it include comparative analysis with historical data?" the system ultimately obtains structured report requirement information. This report requirement information may include the report scope, time window, and analysis dimensions. The report scope refers to the organization to which the report is issued; the time window refers to the time period covered by the report; and the analysis dimensions are the output items, such as statistical data early warning, comparison of processing timeliness, trend prediction, and decision-making recommendations.

[0021] 102. Based on the report requirement interaction information, the report generation task is decomposed to obtain multiple report generation sub-tasks and target report templates covering the corresponding report component modules of all the report generation sub-tasks.

[0022] In this embodiment of the invention, after obtaining clear report requirement information, the system intelligently parses and decomposes the report requirements. Based on preset task management logic and a report structure knowledge base, the complex report generation requirements are broken down into a series of relatively independent report generation sub-tasks that can be executed in parallel or sequentially. Simultaneously, according to the business type and requirement characteristics, the system selects or dynamically assembles a target report template from the template library. This template defines the overall framework of the report and includes report component modules corresponding to each sub-task. In the risk control report, the report component modules may include a monthly risk control overview, core indicator comparison, daily tasks, risk warnings, thematic analysis of core settlement and accounting scenarios, improvement measures, and optimization suggestions.

[0023] It should be noted that the intelligent task decomposition and template matching mechanism not only solves problems such as misunderstanding of requirements, time-consuming content framework design, and easy omission of key points in traditional report production, but also lays the foundation for subsequent automated generation through structured task output.

[0024] 103. Under the target business operation type, the target generating agent of the report generation subtask is called, and the report content of each of the report component modules is generated by the target generating agent.

[0025] In this embodiment of the invention, the system intelligently schedules a series of target generation agents based on the decomposed report generation sub-tasks and their corresponding report component modules. Each generation agent is optimized and trained for different report generation sub-tasks. Different agents specialize in different generative functions; for example, some agents excel at numerical statistics and chart descriptions, while others excel at analysis and decision-making. These agents, based on retrieval-enhanced generation technology, access a unified knowledge base, which may include a business rule base, a historical case base, a contract clause base, etc., and generate high-quality, compliant report content to populate the various modules of the report template based on iteratively optimized dedicated prompts.

[0026] It's worth noting that by invoking target-generating agents that match the business type and task to generate content for corresponding modules, the system can generate highly adapted, accurate, and compliant paragraph content tailored to the professional requirements of different report modules. For example, when generating report content for the valuation and accounting discrepancy analysis module, the system invokes a specially trained valuation and accounting agent. This agent can not only accurately calculate the discrepancy values ​​but also, by combining accounting standards and regulatory requirements from the knowledge base, generate recommendations that meet compliance constraints. Through targeted optimization of the agents, the accuracy, compliance, and operability of the reports are significantly improved. Simultaneously, the parallel generation of paragraphs by multiple agents also significantly shortens the overall report generation time, effectively improving report generation efficiency.

[0027] 104. Integrate the report content according to the target report template, generate and output the target report in response to the report generation request.

[0028] In this embodiment of the invention, after the report content of all report components is generated, the system automatically integrates, arranges, and formats it according to the structure of the target report template to generate a complete risk report that conforms to industry standards. The generated report can be delivered to users through various channels such as the front-end interface, email, or system messages; this embodiment of the invention does not impose specific limitations. During the integration process, the report content of the different report components generated above can be sequentially filled into the target report template, and necessary titles, headers and footers, formatted tables, and trend charts with textual descriptions can be added to generate a complete report. In the asset custody business system, the generated report can also be displayed in the risk dashboard of the custody operation platform for users to download or share via a link.

[0029] In one embodiment of the present invention, for further illustration and limitation, such as Figure 2As shown, before the step of invoking the target dialogue agent matching the target business operation type to perform the report request interaction and obtaining the report request interaction information, the method includes: 201. For any dialogue agent, construct multiple versions of user guidance words that match the business operation type corresponding to the dialogue agent.

[0030] 202. Using the report generation task decomposition and evaluation parameters as the comparison object, conduct comparative tests on different versions of user guidance words, and iteratively update the user guidance words based on the comparison test results to obtain the latest version of user guidance words, so as to assist the dialogue agent in interacting with report requests through the latest version of user guidance words.

[0031] 203. For any generating agent, construct base prompt words that match multiple versions of the report generation subtask corresponding to the generating agent.

[0032] 204. Using the report generation evaluation parameters as a comparison object, conduct comparative tests on different versions of basic prompt words, and iteratively update the basic prompt words based on the comparison test results to obtain the latest version of basic prompt words, so as to assist the execution of the report generation subtask with the latest version of basic prompt words.

[0033] In this embodiment of the invention, during the testing phase, dialogue agents were trained for different business operation types, such as clearing, valuation, and account management. Furthermore, multiple versions of user guidance text were constructed for each dialogue agent. These user guidance texts are used to guide users to clearly and completely express their reporting needs at the beginning of the interaction. Constructing multiple versions of user guidance text allows for iterative optimization through A / B testing. Specifically, different versions of user guidance text are used to serve random user groups in actual interaction scenarios. The effectiveness of each version of guidance text is evaluated by analyzing the report generation task breakdown evaluation parameters generated by the backend, such as the completeness of the task breakdown, the accuracy of subtask matching with requirements, and the efficiency of interaction rounds. The best-performing user guidance text is used to optimize the current user guidance text, thus performing iterative optimization. For example, based on the test results, the system determines the best-performing V2 version as the base version, incorporates the advantages of V3 version, generates up to one V4 version, and then compares the effectiveness of V4 version with V2 version, thereby achieving iterative optimization of the user guidance text.

[0034] For each agent performing a specific report generation subtask, multiple versions of basic prompts are also constructed. These basic prompts include role prompts and task objective prompts. Role prompts define the agent's identity, such as "You are an asset custody and liquidation business expert, skilled in root cause analysis and clear description of operational anomalies." Task objective prompts define specific output requirements, such as "Please strictly classify the provided anomaly data according to the 'Liquidation Business Anomaly Classification Standard' in the knowledge base." A / B testing is used to iteratively optimize the basic prompts. For example, using the same set of test data, the "Liquidation Instruction Anomaly Cause Classification and Description Generation Agent" configured with P1 and P2 prompts generates analysis content. Evaluation reveals that while the hit rate and compliance rate of version P1 are lower than version P2, the language of version P2 is slightly stiff. Therefore, multiple versions are adjusted based on version P2, and then compared and tested to obtain the optimal version P3. This version is then determined as the latest version of the basic prompts for this agent, guiding and constraining the execution of all subsequent related subtasks.

[0035] It should be noted that the above-described process of iteratively updating user guidance words and basic prompts based on A / B testing is also applicable to dynamic updates during actual system application. This involves building new versions of user guidance words or basic prompts based on user feedback and actual application effect feedback, then comparing and testing this new version with the currently used version to update the prompts for the current version. For user guidance words, actual application effect feedback can include average number of dialogue rounds, task breakdown time, accuracy of requirement understanding inferred from subsequent task execution results, and task completion rate. For basic prompts, actual application effect feedback can include content accuracy, compliance, the proportion of generated content directly adopted by business personnel, the proportion of generated false information, and the average time spent on content generation. Through iterative optimization of guidance words and prompts, a closed-loop optimization mechanism based on objective evaluation indicators is established, enabling the agent's interaction and generation capabilities to continuously evolve with business development and feedback accumulation, maintaining the system's cutting-edge nature and applicability.

[0036] In one embodiment of the present invention, for further explanation and limitation, after generating the target report, the method further includes: The report generates evaluation parameters that include structure hit rate, content hit rate, and compliance rate; If the structure hit rate is less than the preset structure hit rate threshold, then the target dialogue agent is retrained and / or the user guidance words of the target dialogue agent are iteratively updated through comparison tests. If the content hit rate is less than the preset content hit rate threshold, the target generated agent will be retrained and / or the basic prompt words of the target generated agent will be iteratively updated through comparison testing. If the compliance rate is less than a preset compliance rate threshold, the target business rule engine of the target generating agent will be retrained.

[0037] In this embodiment of the invention, the structure hit rate characterizes whether the framework structure of the generated assessment report conforms to the preset template requirements. It can be obtained by calculating the ratio of the number of necessary modules in the template to the number of predefined necessary modules in the template. The content hit rate characterizes the accuracy and completeness of the assessment report content, and can be obtained by calculating the ratio of the number of core information points in the report that are consistent with the facts in the knowledge base and meet the user's needs to the total number of core information points expected to be included. The compliance rate characterizes whether the assessment report content conforms to regulatory requirements and internal compliance standards, and can be obtained by calculating the ratio of the number of paragraphs that pass the compliance check to the total number of paragraphs in the report. When the assessment finds that the structure hit rate is below the standard, it indicates that the dialogue agent has failed to adequately guide the user to clarify their needs, or that there are defects in the task decomposition process, requiring targeted optimization of the dialogue interaction capabilities. The structure hit rate threshold can be set to 95%, or adjusted according to business importance, without specific limitation. For example, the structure hit rate of an asset allocation report was 92%, 89%, and 90% for three consecutive times, all below the threshold. Furthermore, the "asset correlation risk analysis" module is often neglected. This allows for two approaches: firstly, retraining the dialogue agent to reinforce the importance of the module; and secondly, updating the guidance words by adding keywords that cross asset classes to the user guidance words.

[0038] When the content hit rate is below standard, it indicates a problem in the content generation process of the generative agent, requiring improvement in the accuracy of its information extraction and the completeness of its output. The content hit rate threshold can be set differently based on the importance of the information points, for example, 90%. Retraining the generative agent involves collecting relevant source data and knowledge base fragments as reinforcement learning samples for missing or erroneous information points, and then retraining the corresponding generative agent. Iteratively updating the basic prompts involves optimizing the task objective descriptions in the prompts to emphasize the necessity of specific information points. For example, the prompts could explicitly state: "The current period's change value for subject XX, the reasons for the change, and industry comparison data must be included in the 'Key Subject Analysis' section." Furthermore, retrieval capabilities can be enhanced by checking the retrieval enhancement generation process and optimizing retrieval strategies to ensure the generative agent can obtain more relevant knowledge fragments.

[0039] When the compliance rate fails to meet the standard, it indicates that the generated content poses regulatory or internal compliance risks and must be corrected immediately. Due to the extremely high compliance requirements in the financial sector, the compliance rate threshold can be set to 98%, and this is the highest priority for optimization. In this case, retraining the business rule engine is the core measure. All paragraphs that failed the compliance check and the specific compliance rules they triggered are collected, and patterns of rule misuse or omission are analyzed. The rule engine's rule base and inference logic are updated, and the collaborative mechanism between the rule engine and the generating agent is retrained. During model processing, the weight of compliance constraints is increased.

[0040] It should be noted that the comparison of the aforementioned structure hit rate, content hit rate, and compliance rate with their corresponding thresholds can be performed after each report is generated, or a unified evaluation of reports from historical periods can be conducted based on preset time points and report generation volume triggers. By monitoring these three independent indicators, weaknesses can be identified in a timely manner, and targeted optimizations can be made to the models or prompts that need to be optimized, effectively controlling resource consumption during the optimization process while ensuring accuracy.

[0041] In one embodiment of the present invention, for further explanation and limitation, the process of any of the target generating agents generating report content includes: The system queries the database for pending business data that matches the target business operation type and the target business rule engine of the target generating agent. It then calculates the pending business data using the pre-configured business rules in the target business rule engine to generate a deterministic output containing quantitative parameters and compliance constraints. The process employs enhanced generation techniques to retrieve associated business data from the business knowledge base that is related to the business data to be processed. The deterministic output and the associated business data are fused with the basic prompt words corresponding to the target generating agent to generate structured prompt words; The structured prompts are input into the large model inference engine of the target generating agent to generate report content through the large model inference engine.

[0042] In this embodiment of the invention, each target generation agent consists of two core engines (a business rule engine and a large-model inference engine), forming a hybrid decision-making architecture that combines rule determinism with intelligent flexibility. Specifically, each target generation agent includes a business rule engine that matches report generation sub-tasks and a large-model inference engine trained by fine-tuning based on historical data from these sub-tasks. The business rule engine represents rule determinism, used to match specific report generation sub-tasks, and incorporates precise calculation rules, compliance clauses, and standardized processing logic for that business domain. In the asset custody business scenario, this engine is built upon a knowledge base of asset custody operation business rules, ensuring that the output of basic facts and judgments fully complies with business specifications. The large-model inference engine represents intelligent flexibility, trained by fine-tuning based on historical high-quality report data and interaction records of the corresponding report generation sub-task, and possesses professional domain language understanding, multi-source information fusion, and natural language generation capabilities.

[0043] When the system schedules a target-generating agent to execute its corresponding report generation subtask, it first extracts the business data to be processed related to the task from the business database or message queue. Then, it calls the business rule engine corresponding to the agent to process this data according to pre-configured business rules, such as valuation formulas, risk monitoring thresholds, and regulatory compliance clauses. This produces quantitative parameters such as risk indicator values ​​and statistical results, as well as compliance constraints such as mandatory disclosure formats, prohibited expressions, and mandatory contractual clauses, resulting in a deterministic output as an intermediate result.

[0044] To enrich the analytical dimensions and depth of the report, the system simultaneously employs Retrieval-Augmented Generation (RAG) technology to retrieve highly relevant background information, historical similar cases, regulatory guidelines, or professional interpretations from a vectorized business knowledge base. This step provides rich contextual information for the subsequent report generation by the large-scale model inference engine. The system then intelligently integrates the deterministic output of the business rules engine, the relevant business data retrieved by RAG, and the latest version of the basic prompts for the generating agent, assembling them into a complete and clearly defined structured prompt. This allows the large-scale model to generate the corresponding report content modules under the guidance of these prompts. The business rules engine ensures the determinism, accuracy, and compliance of the output; the large-scale model inference engine provides analytical flexibility, depth of insight, and adaptability of expression, ensuring the accuracy of the report content while making it standardized and easy to understand.

[0045] In one embodiment of the present invention, for further explanation and limitation, before querying pending business data matching the target business operation type from the database, the method further includes: Receive and parse risk warning messages, pending tasks, and business progress data sent by the business operation monitoring system, and associate the parsed data with different business operation types and store it in the managed operation system database for querying pending business data.

[0046] In this embodiment of the invention, at the beginning of the report generation process, the system queries the database of the managed operation system for real-time data pushes from the upstream business operation monitoring system. This data includes risk warning messages used to identify specific risk events or abnormal indicators; pending tasks generated based on risks or business processes that require manual intervention or confirmation; and business progress data reflecting the current processing status and timeliness of various business processes. The real-time data pushes from the business operation monitoring system are generated in real time through an automated monitoring process. Specifically, the business operation monitoring system subscribes to real-time business operation process data from the managed business system via message queues such as Kafka and RocketMQ. The monitoring system internally deploys indicator calculation services and risk rule engines. These engines listen to and process the real-time business operation process data flowing through the message queues: calculating the current value in real time based on predefined business indicators, such as average instruction confirmation time and valuation calculation error rate. The calculated indicator values ​​are compared with preset risk thresholds and business rules. Once a rule is triggered, a structured risk warning message is immediately generated. Simultaneously, based on the business process model, when the status of a certain link changes or the waiting time is too long, pending tasks are automatically generated. The system will also continuously aggregate this process data to form business progress dashboards categorized by organization, personnel, and product. Data subscription and transmission through these systems ensure the accuracy and timeliness of the reports. In the context of asset custody risk assessment, the interaction relationships between these systems are as follows: Figure 3 As shown.

[0047] In one embodiment of the present invention, for further explanation and limitation, the step of decomposing the report generation task based on the report requirement interaction information includes: Analyze the report request interaction information to determine the report task; The reporting task is broken down into multiple independent report generation sub-tasks according to the progressive hierarchy of business data statistics, core indicator monitoring, business analysis, and decision-making suggestions. The report component modules corresponding to each of the different report generation subtasks are determined, and the structural deployment of each report component module in the report is determined based on the execution dependencies of each report generation subtask, thus obtaining the target report template.

[0048] In this embodiment of the invention, the system performs semantic parsing on the user's confirmed request, identifying core elements such as report topic, business scope, time period, and focus of attention, thereby clarifying the specific reporting task. Subsequently, following a progressive logic of "describing phenomena, quantifying monitoring, analyzing root causes, and proposing countermeasures," the system decomposes the macro-task into a series of logically related yet relatively independent sub-tasks. For example, it first summarizes basic business data, then calculates key risk indicators, followed by multi-dimensional business analysis, and finally forms decision recommendations. Simultaneously, based on the content of the decomposed sub-tasks, the system maps them to preset standardized report modules and arranges these modules in an orderly manner according to the natural logic of information presentation. For example, from overview to details, from current status to recommendations, it dynamically generates a customized report template with a clear structure and complete content, providing accurate task objectives for subsequent modular content generation and intelligent agent scheduling. The report components include a preset business cycle overview module, a core business data statistics module, and a pending business statistics module, as well as a business analysis module, an early warning status module, and a decision recommendation module that match the target business operation type. The preset business cycle overview module includes: the report coverage period, such as the third quarter of 2023; the covered business lines, such as securities clearing, fund clearing, and cross-border clearing; and the institutions involved, such as public fund segregated accounts. The core business data statistics module presents core operational indicators within the target business cycle in quantitative data form, such as categorized statistical data, year-on-year / month-on-month trends, and total business volume statistics. The pending business statistics module includes a list of tasks to be executed. The business analysis module includes in-depth analysis of warnings and abnormal situations, which may include multi-dimensional analysis of abnormality types, identification of bottlenecks in key processes, historical comparisons, and trend predictions. The warning status module displays the triggering status of preset risk monitoring rules. The decision-making suggestion module includes measures and suggestions for responding to warnings, which may include short-term emergency measures and medium-term process optimization suggestions.

[0049] In one embodiment of the present invention, for further explanation and limitation, after outputting the target report, the method further includes: The target dialogue agent listens for and parses user feedback information. If the parsing result of the user feedback information triggers any report update condition, the target report will be updated according to the update strategy corresponding to the triggered report update condition.

[0050] In this embodiment of the invention, after the target report is generated and presented to the user, the system does not immediately terminate the service. Instead, it switches the corresponding target dialogue agent to a continuous listening state, keeping the conversation window or dedicated feedback channel active. This allows the system to receive and parse various forms of user feedback regarding the generated report. The agent utilizes its natural language understanding capabilities to interpret the intent of this feedback, the specific report module it refers to, and the desired adjustment direction. The agent matches the parsed feedback results with a preset mapping relationship between report update conditions and update strategies. If any condition is triggered, the update process is automatically initiated, invoking the corresponding resources and agents to revise the original target report according to the predetermined update strategy.

[0051] The update strategies include single-module content updates, report structure updates, and full-text report updates. When feedback only concerns a specific module in the report, a single-module content update strategy is used. This involves locating and scheduling the target agent responsible for that module's content, generating new instructions based on the feedback information, or querying new data to regenerate the module's content, replacing the original module's content while keeping the rest of the report unchanged. When feedback indicates that the overall report structure, module order, or presentation logic needs adjustment, a report structure update strategy is used. This involves reassessing the report requirements, potentially involving task decomposition and template construction to generate a new report structure template. Then, based on the new template, the corresponding agents are scheduled to reorganize and migrate the original module content, and may need to generate entirely new content for new modules. When feedback indicates significant changes to the report's underlying data, core conclusions, or application scenarios, such as sudden changes in regulatory policies requiring a reassessment of compliance, a full-text report update strategy is used. This treats user feedback as a new and more important input, combining it with the original report request, and re-executes the entire process from the initial report request interaction to generate a completely new report version. The original report may be archived as a historical version. Based on the above process, the generated reports can be dynamically updated, allowing the reports to adapt to user feedback and improve the accuracy and flexibility of report generation. This also avoids the need to resubmit a request if some parts of the report are non-compliant, thus avoiding a waste of resources.

[0052] This invention provides a report generation method based on the collaboration of a rule engine and a large model. In this embodiment, in response to a report generation request, a target dialogue agent matching the target business operation type is invoked to perform a report request interaction, obtaining report request interaction information. Based on this information, the report generation task is decomposed into multiple report generation sub-tasks, and a target report template covering the corresponding report components of each sub-task is generated. The target generation agent matching the report generation sub-task under the target business operation type is invoked, and the report content of each report component is generated by the target generation agent. The report content is integrated based on the target report templates to generate and output the target report responding to the report generation request. This significantly reduces manual intervention and repetitive work in the report generation process, minimizing report quality fluctuations and time consumption caused by differences in human experience. Simultaneously, the intelligent task decomposition and professional agent scheduling mechanism ensure the comprehensiveness and format standardization of the report, thereby greatly improving the efficiency of report generation.

[0053] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this embodiment of the invention provides a report generation system based on the collaboration between a rule engine and a large model, such as... Figure 4 As shown, the system includes: The dialogue agent invocation module 31 is used to respond to the report generation request, invoke the target dialogue agent that matches the target business operation type to perform the report request interaction, and obtain the report request interaction information; The task decomposition module 32 is used to decompose the report generation task according to the report requirement interaction information to obtain multiple report generation sub-tasks and a target report template covering the report composition modules corresponding to each of the report generation sub-tasks. Content generation module 33 is used to call the target generation agent that matches the report generation subtask under the target business operation type, and generate the report content of each of the report component modules through the target generation agent; The content integration module 34 is used to integrate the report content according to the target report template, generate and output the target report in response to the report generation request.

[0054] Furthermore, the system also includes: The first construction module is used to construct multiple versions of user guidance words that match the business operation type corresponding to any dialogue agent; The first optimization module is used to compare and test different versions of user guide words with the evaluation parameters of the report generation task decomposition as the comparison object, and to iteratively update the user guide words according to the comparison and test results to obtain the latest version of user guide words, so as to assist the dialogue agent in interacting with the report request through the latest version of user guide words; The second construction module is used to construct, for any generating agent, multiple versions of basic prompt words matching the report generation subtask corresponding to the generating agent, wherein the basic prompt words include role prompt words and task target prompt words; The second optimization module is used to compare and test different versions of basic prompt words using the report generation evaluation parameters as the comparison object, and to iteratively update the basic prompt words based on the comparison test results to obtain the latest version of basic prompt words, so as to assist the execution of the report generation subtask with the latest version of basic prompt words.

[0055] Furthermore, the system also includes: The calculation module is used to calculate report generation evaluation parameters that include structure hit rate, content hit rate, and compliance rate; The first training module is used to retrain the target dialogue agent and / or iteratively update the user guidance words of the target dialogue agent through comparison tests if the structure hit rate is less than a preset structure hit rate threshold. The second training module is used to retrain the target generated agent and / or iteratively update the basic prompt words of the target generated agent through comparison tests if the content hit rate is less than a preset content hit rate threshold. The third training module is used to retrain the target business rule engine of the target generating agent if the compliance rate is less than a preset compliance rate threshold.

[0056] Furthermore, the content generation module 33 includes: The rule calculation unit is used to query the pending business data that matches the target business operation type from the database, and the target business rule engine of the target generating agent. The pending business data is calculated using the pre-configured business rules in the target business rule engine to generate a deterministic output containing quantitative parameters and compliance constraints. The retrieval unit is used to query related business data associated with the business data to be processed from the business knowledge base through retrieval enhancement generation technology; The fusion unit is used to fuse the deterministic output and the associated business data with the basic prompt words corresponding to the target generating agent to generate structured prompt words; The generation unit is used to input the structured prompts into the large model inference engine of the target generating agent, so as to generate report content through the large model inference engine.

[0057] Furthermore, the system also includes: The storage module is used to receive and parse risk warning messages, pending tasks, and business progress data sent by the business operation monitoring system, and associate the parsed data with different business operation types and store it in the managed operation system database for querying pending business data. The generation process of risk warning messages, pending tasks, and business progress data includes: the business operation monitoring system subscribes to real-time business operation process data from the managed business system through a message queue, and calculates monitoring indicators and generates warning messages within a preset period for the real-time business operation process data to obtain risk warning messages, pending tasks, and business progress data.

[0058] Furthermore, the task breakdown module 32 includes: The parsing unit is used to parse the report request interaction information and determine the report task; The decomposition unit is used to decompose the reporting task into multiple independent report generation sub-tasks according to the progressive hierarchy of business data statistics, core indicator monitoring, business analysis, and decision-making suggestions. The determining unit is used to determine the report component modules corresponding to each of the different report generation subtasks, and to determine the structural deployment of each report component module in the report based on the execution dependency of each report generation subtask, so as to obtain the target report template; The report consists of a preset business cycle overview module, a core business data statistics module, and a pending business statistics module, as well as a business analysis module, an early warning status module, and a decision suggestion module that match the target business operation type.

[0059] Furthermore, the system also includes: The listening module is used to listen to and parse user feedback information through the target dialogue agent; The update module is used to update the target report according to the update strategy corresponding to the triggered report update condition if the parsing result of the user feedback information triggers any report update condition. The update strategy includes single module content update, report structure update and full report update.

[0060] This invention provides a report generation system based on the collaboration of a rule engine and a large model. In this embodiment, in response to a report generation request, a target dialogue agent matching the target business operation type is invoked to perform a report request interaction, obtaining report request interaction information. Based on this information, the report generation task is decomposed into multiple report generation sub-tasks, and a target report template covering the corresponding report components of each sub-task is generated. The target generation agent matching the report generation sub-task under the target business operation type is invoked, and the report content of each report component module is generated by the target generation agent. The report content is integrated based on the target report templates to generate and output the target report responding to the report generation request. This significantly reduces manual intervention and repetitive work in the report generation process, minimizing report quality fluctuations and time consumption caused by differences in human experience. Simultaneously, the intelligent task decomposition and professional agent scheduling mechanism ensure the comprehensiveness and format standardization of the report, thereby greatly improving the efficiency of report generation.

[0061] According to one embodiment of the present invention, a storage medium is provided, the storage medium storing at least one executable instruction, which is capable of executing the report generation method based on the collaboration of rule engine and large model in any of the above method embodiments.

[0062] Figure 5 The diagram shows a structural schematic of a terminal according to an embodiment of the present invention. The specific implementation of the present invention does not limit the specific implementation of the terminal.

[0063] like Figure 5 As shown, the terminal may include: a processor 402, a communication interface 404, a memory 406, and a communication bus 408.

[0064] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408.

[0065] Communication interface 404 is used for network communication with other devices such as clients or other servers.

[0066] The processor 402 is used to execute program 410, specifically to execute the relevant steps in the above-described embodiment of the report generation method based on the collaboration of a rule engine and a large model.

[0067] Specifically, program 410 may include program code that includes computer operation instructions.

[0068] Processor 402 may be a central processing unit (CPU), a specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The terminal may include one or more processors of the same type, such as one or more CPUs; or it may include processors of different types, such as one or more CPUs and one or more ASICs.

[0069] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0070] Specifically, program 410 can be used to cause processor 402 to perform the following operations: In response to a report generation request, the target dialogue agent matching the target business operation type is invoked to perform the report request interaction and obtain the report request interaction information; Based on the report requirement interaction information, the report generation task is decomposed to obtain multiple report generation sub-tasks, and a target report template covering the report component modules corresponding to each of the report generation sub-tasks; Under the target business operation type, the target generating agent of the report generation subtask is matched, and the report content of each of the report component modules is generated by the target generating agent. The report content is integrated based on the target report template to generate and output the target report in response to the report generation request.

[0071] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0072] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A report generation method based on the collaboration of a rule engine and a large model, characterized in that, include: In response to a report generation request, the target dialogue agent matching the target business operation type is invoked to perform the report request interaction and obtain the report request interaction information; Based on the report requirement interaction information, the report generation task is decomposed to obtain multiple report generation sub-tasks, and a target report template covering the report component modules corresponding to each of the report generation sub-tasks; Under the target business operation type, the target generating agent of the report generation subtask is matched, and the report content of each of the report component modules is generated by the target generating agent. The report content is integrated based on the target report template to generate and output the target report in response to the report generation request.

2. The method according to claim 1, characterized in that, Before invoking the target dialogue agent matching the target business operation type to perform the report request interaction and obtain the report request interaction information, the method includes: For any dialogue agent, construct multiple versions of user guidance words that match the business operation type corresponding to the dialogue agent; Using the report generation task decomposition and evaluation parameters as the comparison object, the user guidance words of different versions are compared and tested, and the user guidance words are iteratively updated according to the comparison and test results to obtain the latest version of the user guidance words, so as to assist the dialogue agent in interacting with the report request through the latest version of the user guidance words; For any generating agent, construct multiple versions of basic prompt words that match the report generation subtask corresponding to the generating agent, wherein the basic prompt words include role prompt words and task goal prompt words; Using the report generation evaluation parameters as a comparison object, the basic prompt words of different versions are compared and tested. Based on the comparison and test results, the basic prompt words are iteratively updated to obtain the latest version of the basic prompt words, so as to assist the execution of the report generation subtask.

3. The method according to claim 2, characterized in that, After generating the target report, the method further includes: The report generates evaluation parameters that include structure hit rate, content hit rate, and compliance rate; If the structure hit rate is less than the preset structure hit rate threshold, then the target dialogue agent is retrained and / or the user guidance words of the target dialogue agent are iteratively updated through comparison tests. If the content hit rate is less than the preset content hit rate threshold, the target generated agent will be retrained and / or the basic prompt words of the target generated agent will be iteratively updated through comparison testing. If the compliance rate is less than a preset compliance rate threshold, the target business rule engine of the target generating agent will be retrained.

4. The method according to claim 1, characterized in that, Any of the target generation agents includes a business rule engine for matching report generation subtasks and a large model inference engine fine-tuned based on historical data from the report generation subtasks; The process of generating report content by any of the target generating agents includes: The database is queried to retrieve pending business data that matches the target business operation type, and the target business rule engine of the target generating agent is retrieved. The pending business data is calculated using the pre-configured business rules in the target business rule engine to generate a deterministic output containing quantitative parameters and compliance constraints. The process employs enhanced generation techniques to retrieve associated business data from the business knowledge base that is related to the business data to be processed. The deterministic output and the associated business data are fused with the basic prompt words corresponding to the target generating agent to generate structured prompt words; The structured prompts are input into the large model inference engine of the target generating agent to generate report content through the large model inference engine.

5. The method according to claim 1, characterized in that, Before querying pending business data matching the target business operation type from the database, the method further includes: Receive and parse risk warning messages, pending tasks, and business progress data sent by the business operation monitoring system, and associate the parsed data with different business operation types and store it in the managed operation system database for querying pending business data; The generation process for risk warning messages, to-do items, and business progress data includes: The business operation monitoring system subscribes to real-time business operation process data from the managed business system through a message queue, and calculates monitoring indicators and generates early warning messages within a preset period based on the real-time business operation process data, thereby obtaining risk warning messages, to-do items, and business progress data.

6. The method according to claim 1, characterized in that, The step of breaking down the report generation task based on the report requirement interaction information includes: Analyze the report request interaction information to determine the report task; The reporting task is broken down into multiple independent report generation sub-tasks according to the progressive hierarchy of business data statistics, core indicator monitoring, business analysis, and decision-making suggestions. The report component modules corresponding to each of the different report generation subtasks are determined, and the structural deployment of each report component module in the report is determined based on the execution dependencies of each report generation subtask, thus obtaining the target report template; The report consists of a preset business cycle overview module, a core business data statistics module, and a pending business statistics module, as well as a business analysis module, an early warning status module, and a decision suggestion module that match the target business operation type.

7. The method according to claim 1, characterized in that, After outputting the target report, the method further includes: The target dialogue agent listens for and parses user feedback information. If the parsing result of the user feedback information triggers any report update condition, the target report is updated according to the update strategy corresponding to the triggered report update condition. The update strategy includes single module content update, report structure update, and full report update.

8. A report generation system based on the collaboration of a rule engine and a large model, characterized in that, include: The dialogue agent invocation module is used to respond to report generation requests by invoking the target dialogue agent that matches the target business operation type to perform the report request interaction and obtain the report request interaction information. The task decomposition module is used to decompose the report generation task according to the report requirement interaction information to obtain multiple report generation sub-tasks and target report templates covering the report component modules corresponding to each of the report generation sub-tasks. The content generation module is used to call the target generation agent that matches the report generation subtask under the target business operation type, and generate the report content of each of the report component modules through the target generation agent; The content integration module is used to integrate the report content according to the target report template, generate and output the target report in response to the report generation request.

9. A storage medium, characterized in that, The storage medium stores at least one executable instruction that causes the processor to perform the operation corresponding to the report generation method based on rule engine and large model collaboration as described in any one of claims 1-7.

10. A terminal, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation corresponding to the report generation method based on rule engine and large model collaboration as described in any one of claims 1-7.

Images