Audit problem qualitative and rectification tracking intelligent processing method and device based on large model, and medium

Abstract

The invention provides an audit problem qualitative and rectification tracking intelligent processing method and device based on a large model and a medium. Belongs to the technical field of auditing. The system comprises a model construction module which is used for constructing an intelligent audit knowledge cloud brain and generating knowledge base data; on the basis of the knowledge base data, key auditing element extraction is carried out on an auditing service scene, and auditing element extraction result data is generated; performing auditing problem matching model training according to the auditing element extraction result data, and constructing an auditing problem accurate matching model; based on the powerful knowledge base and the intelligent matching capability of a big language model, knowledge content closely related to the question is rapidly and automatically filled, key information such as question names, representation forms, question qualifications, qualitative bases and processing punishment bases is covered, the time for auditing personnel to manually search and sort data is greatly saved, and the auditing efficiency is improved. The accuracy and integrity of problem description are ensured, and a solid foundation is laid for subsequent audit analysis.

Description

Technical Field

[0001] This invention proposes an intelligent processing method, device, and medium for qualitative analysis and rectification tracking of audit issues based on a large model, belonging to the field of audit technology. Background Technology

[0002] Currently, internal auditing in enterprises faces a dual challenge: on the one hand, as business structures become increasingly complex, the characterization of audit issues relies more heavily on the experience and breadth of knowledge of auditors, leading to risks such as inconsistent judgment standards and incomplete citation of evidence; on the other hand, in the audit rectification phase, the lack of specific and actionable guidelines often makes rectification measures difficult to implement, or even merely a formality, hindering the effective transformation of audit results. These real pain points have created an urgent need for standardized and intelligent audit support tools in enterprises.

[0003] Against this backdrop, we have developed and implemented a knowledge-driven, "penetrating" audit management model. Integrating large language model technology and using audit knowledge, including external regulations and internal systems, as the core, we have constructed an intelligent audit support system. This system forms a two-way, closed-loop management model for audit rectification, assisting auditors in quickly and accurately identifying the essence of problems and providing structured, traceable rectification paths and material specifications. This effectively promotes the "thorough rectification and complete closure" of audit findings. Its breakthrough significance lies in transforming the two most crucial and experience-dependent aspects of the audit process—"problem characterization" and "rectification design"—into a standardized operating process driven by a knowledge engine and characterized by efficient human-machine collaboration. This provides a solid technological foundation for enterprises to achieve "penetrating" and refined management. Summary of the Invention

[0004] This invention provides a method, apparatus, and medium for intelligent processing of audit problem characterization and rectification tracking based on a large model, to solve the problems mentioned in the background art above:

[0005] The present invention proposes an intelligent processing method for qualitative analysis and rectification tracking of audit issues based on a large model, the method comprising:

[0006] S1. Construct an intelligent audit knowledge cloud brain to generate knowledge base data; based on this knowledge base data, extract key audit elements from audit business scenarios and generate audit element extraction result data; train an audit problem matching model based on the audit element extraction result data to construct an audit problem accurate matching model.

[0007] S2. Based on the audit issue precise matching model, perform second-level matching processing on issues in actual audit business to obtain preliminary audit issue qualitative results data; perform multi-dimensional verification on the preliminary audit issue qualitative results data to generate precise audit issue qualitative data;

[0008] S3. Based on the precise audit problem qualitative data, conduct a layer-by-layer penetrating analysis from three dimensions: data, process, and responsibility, trace the root cause of the problem, and generate problem root cause analysis data; based on the problem root cause analysis data, match the corresponding rectification strategies from the intelligent rectification knowledge base to generate initial rectification strategy data; conduct feasibility assessment and optimization of the initial rectification strategy data to generate precise rectification strategy data;

[0009] S4. Formulate specific rectification action plans based on precise rectification strategy data, and generate rectification action plan data; carry out rectification work based on rectification action plan data, and conduct real-time effectiveness verification during the rectification process, generating rectification effectiveness verification data; determine whether the rectification has met the standards based on the rectification effectiveness verification data; if it has not met the standards, return to the strategy generation module to re-analyze the root causes and formulate rectification strategies; if it has met the standards, proceed to the next step.

[0010] S5. For compliant rectification cases, solidify them into a system, incorporate effective rectification measures and processes into the company's internal audit system, and generate system solidification data; update the knowledge base in the intelligent audit knowledge cloud brain based on the system solidification data, and generate updated knowledge base data.

[0011] The present invention proposes an apparatus for implementing the intelligent processing method for qualitative and corrective action tracking of audit issues based on a large model, as described above. The apparatus comprises:

[0012] Model building module: Constructs an intelligent audit knowledge cloud brain to generate knowledge base data; based on this knowledge base data, extracts key audit elements from audit business scenarios to generate audit element extraction result data; trains an audit question matching model based on the audit element extraction result data to build an audit question accurate matching model;

[0013] Problem identification module: Based on the audit problem precise matching model, it performs second-level matching processing on problems in actual audit business to obtain preliminary audit problem identification results data; it performs multi-dimensional verification on the preliminary audit problem identification results data to generate accurate audit problem identification data;

[0014] Strategy Generation Module: Based on the qualitative data of accurate audit issues, it conducts a layer-by-layer penetrating analysis from three dimensions: data, process, and responsibility, tracing the root cause of the problem and generating root cause analysis data; based on the root cause analysis data, it matches the corresponding rectification strategies from the intelligent rectification knowledge base and generates initial rectification strategy data; it conducts feasibility assessment and optimization of the initial rectification strategy data to generate accurate rectification strategy data;

[0015] Rectification Verification Module: This module develops specific rectification action plans based on precise rectification strategy data, generating rectification action plan data; it conducts rectification work based on the rectification action plan data, and verifies the effectiveness in real time during the rectification process, generating rectification effectiveness verification data; it determines whether the rectification has met the standards based on the rectification effectiveness verification data; if not, it returns to the strategy generation module to re-analyze the root causes and formulate rectification strategies; if it has met the standards, it proceeds to the next step.

[0016] Data update module: Solidifies the compliant rectification cases into regulations, incorporates effective rectification measures and processes into the company's internal audit system, and generates solidified regulations data; updates the knowledge base in the intelligent audit knowledge cloud brain based on the solidified regulations data, and generates updated knowledge base data.

[0017] The present invention proposes a non-transitory computer-readable storage medium storing a computer program that is executed by a processor to implement the intelligent processing method for qualitative and remedial tracking of audit issues based on a large model, as described above.

[0018] The beneficial effects of this invention are as follows: Leveraging the powerful knowledge base and intelligent matching capabilities of a large language model, it quickly and automatically fills in knowledge content closely related to the problem, covering key information such as the problem name, manifestation, problem characterization, the basis for characterization, and the basis for handling and penalties. This greatly saves auditors' time in manually searching and organizing materials, and ensures the accuracy and completeness of the problem description, laying a solid foundation for subsequent audit analysis. The system conducts in-depth analysis of the problem from multiple dimensions such as data, processes, and responsibilities, accurately identifying abnormal indicators, comprehensively and meticulously tracing process breakpoints and vulnerabilities, clearly defining the responsible parties, and enabling rapid focus on the core of the problem and clarifying the direction of process optimization, thus laying a solid foundation for subsequent... This provides strong evidence for continued accountability and rectification. Based on the nature of the problem, the system utilizes advanced intelligent algorithms and deeply references historical rectification case libraries and industry best practices to tailor scientific, personalized, and feasible rectification plans for each specific problem. These plans include rectification requirements, rectification results, supporting documentation, and applicable laws and regulations, providing precise rectification guidance and real-time tracking of rectification progress to ensure orderly progress. The system can conduct in-depth analysis and mining of rectification feedback results, using them as an important basis for optimizing the knowledge base. Based on the feedback results, the system precisely adjusts the knowledge mapping table, continuously enriching and improving the knowledge base content, and enhancing the accuracy of knowledge matching. Through continuous dynamic optimization of the knowledge base, it better adapts to ever-changing audit needs, providing more precise and efficient support for subsequent audit work and achieving continuous improvement in audit capabilities. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method described in this invention;

[0020] Figure 2This is an interaction diagram of the method described in this invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] One embodiment of the present invention, such as Figure 1 , Figure 2 As shown, an intelligent processing method for qualitative analysis and rectification tracking of audit issues based on a large model is described, the method comprising:

[0023] S1. Integrate national laws, industry standards, and internal regulations to build an intelligent audit knowledge cloud brain, generating a knowledge base data containing comprehensive audit knowledge; based on this knowledge base data, extract key audit elements from audit business scenarios to generate audit element extraction result data; train an audit problem matching model based on the audit element extraction result data to build an audit problem accurate matching model.

[0024] S2. Based on the audit issue precise matching model, the issues in the actual audit business are matched and processed in seconds to obtain preliminary audit issue qualitative results data; the preliminary audit issue qualitative results data are verified in multiple dimensions, including comparison and verification with historical audit cases, verification with expert experience rules, etc., to generate precise audit issue qualitative data;

[0025] S3. Based on the precise audit problem qualitative data, conduct a layer-by-layer penetrating analysis from three dimensions: data, process, and responsibility, trace the root cause of the problem, and generate problem root cause analysis data; based on the problem root cause analysis data, match the corresponding rectification strategies from the intelligent rectification knowledge base to generate initial rectification strategy data; conduct feasibility assessment and optimization of the initial rectification strategy data to generate precise rectification strategy data;

[0026] S4. Formulate specific rectification action plans based on precise rectification strategy data, and generate rectification action plan data; carry out rectification work based on rectification action plan data, and conduct real-time effectiveness verification during the rectification process, generating rectification effectiveness verification data; determine whether the rectification has met the standards based on the rectification effectiveness verification data; if it has not met the standards, return to the strategy generation module to re-analyze the root causes and formulate rectification strategies; if it has met the standards, proceed to the next step.

[0027] S5. For compliant rectification cases, solidify them into a system, incorporate effective rectification measures and processes into the company's internal audit system, and generate solidified system data; update the knowledge base in the intelligent audit knowledge cloud brain based on the solidified system data, provide continuous optimization support for subsequent audit problem characterization and rectification, and generate updated knowledge base data.

[0028] The working principle and effects of the above technical solution are as follows: Relying on the intelligent audit knowledge cloud brain and precise matching model, the accuracy and efficiency of audit problem characterization are greatly improved, reducing the time cost of manual law retrieval and repeated discussions; Through three-dimensional penetration and tracing of data, processes, and responsibilities, the pertinence of rectification strategies is enhanced, the cost of rectification trial and error and the difficulty of implementation are reduced, and superficial rectification is avoided; Multi-dimensional verification and closed-loop management not only enhance the authority and standardization of audit work, but also reduce disputes over different penalties for the same mistake and the phenomenon of repeated offenses; The solidification of systems and dynamic updating of the knowledge base can not only transform the rectification results into a long-term mechanism, but also continuously enhance the organization's risk prevention and control capabilities, and avoid disputes caused by inaccurate characterization or incomplete evidence.

[0029] In one embodiment of the present invention, S1 includes:

[0030] S11. Regularly crawl policies and regulations from external authoritative channels using web crawling technology. These external channels include policy document libraries, legal databases, and official websites of industry regulatory authorities. Simultaneously, collect internal knowledge resources, including internal company rules, historical audit cases, and expert experience summaries, to form multi-source heterogeneous audit knowledge raw data.

[0031] S12. Perform clause-level slicing, semantic parsing, deduplication and cleaning, and format unification on the original data of multi-source heterogeneous audit knowledge. Classify the data according to the hierarchy of laws, administrative regulations, departmental rules, normative documents and industry standards, and perform dual-track indexing and labeling in combination with business domains, including accounting, engineering construction and bidding and procurement, to generate a standardized audit knowledge base. The standardized audit knowledge base contains a structured knowledge chain of responsibilities, issues, qualitative and legal basis.

[0032] S13. Based on natural language processing (NLP) technology, analyze the content of the standardized audit knowledge base and automatically extract key audit elements. The key audit elements include problem type, responsible party, violation, qualitative basis and penalty standard, forming a structured audit element dataset.

[0033] S14. Divide the structured audit element dataset into training set, validation set and test set, integrate the rule reasoning engine and similarity matching engine of the large language model, optimize the model weight through parameter fine-tuning strategy, and construct an audit problem accurate matching model. The audit problem accurate matching model can realize the accurate mapping between audit problems and knowledge chains.

[0034] The working principle and effects of the above technical solution are as follows: By aggregating and standardizing multi-source authoritative knowledge, the comprehensiveness and standardization of the audit knowledge base are greatly improved, avoiding the problem of missing or conflicting legal provisions leading to a lack of evidence; dual-track indexing and annotation make knowledge retrieval more efficient, reducing the time cost of manually searching for regulations; NLP automatically extracts key elements, reducing the tedious workload of manual organization and improving the accuracy of element extraction; integrating a large model with dual engines and optimizing weights enhances the accuracy of model matching, achieving efficient mapping between audit issues and knowledge chains, avoiding qualitative biases caused by insufficient human experience or professional barriers; it can not only ensure the authority and suitability of audit evidence, but also adapt to the needs of multiple business areas such as accounting and engineering construction.

[0035] In one embodiment of the present invention, S14 includes:

[0036] The structured audit element dataset is divided into training set, validation set and test set according to a preset ratio (e.g. 7:2:1). The data is then labeled and the feature dimensions are unified to generate the standardized audit element dataset.

[0037] The rule reasoning engine (used to parse legal logic and expert rules) and the similarity matching engine (used to calculate the semantic similarity between questions and knowledge chains) of the large language model are integrated into modules, and a standardized audit element dataset is accessed to generate the initial framework of the model;

[0038] Based on the training set data, the LoRA lightweight fine-tuning strategy is used to iteratively optimize the parameters of the key layers of the model. The gradient descent algorithm is combined to adjust the weight allocation and generate the optimized model parameter configuration.

[0039] The validation set iteratively verifies the model performance. The validation set is input into the initial framework of the model, and the model effect is verified by calculating the qualitative accuracy, the basis matching rate and other relevant indicators, and a model performance evaluation report is generated.

[0040] Adjust the dual-engine collaborative logic and parameter thresholds based on the performance evaluation report, complete the final performance verification using the test set, and generate an audit issue precise matching model that can accurately map audit issues to knowledge chains.

[0041] The working principle and effects of the above technical solution are as follows: Standardized processing and scientific partitioning of the dataset improves the consistency of training data, avoiding model training bias caused by chaotic data formats and inconsistent features; the integration of dual-engine modules allows for accurate parsing of legal logic using a rule-based reasoning engine, while also capturing semantic relationships through a similarity matching engine, enhancing the model's comprehensive matching ability and effectively avoiding logical biases or inaccurate semantic matching that may occur with a single engine; the LoRA lightweight fine-tuning strategy reduces the computational and time costs of model training, eliminating the need to spend significant resources retraining the entire model, while gradient descent algorithms optimize weights, making model parameter configuration more aligned with audit business needs; multiple rounds of verification and parameter adjustment significantly improve the model's qualitative accuracy and basis matching rate, avoiding unstable model performance and poor implementation results; the final constructed accurate matching model enables efficient mapping between audit issues and knowledge chains, providing reliable support for audit qualitative analysis, making model application both efficient and worry-free.

[0042] In one embodiment of the present invention, S2 includes:

[0043] S21. Receive the pending issue information entered by the auditors through the audit issue characterization and audit rectification big model system. The pending issue information includes issue description text, related business data and related voucher images. Perform data format conversion and integrity verification to generate a standardized pending issue dataset.

[0044] S22. Input the standardized dataset of undetermined issues into the audit issue precise matching model, analyze the core features of the issues through the semantic understanding module of the model, quickly retrieve the matching knowledge chain in the standardized audit knowledge base, and output preliminary audit issue qualitative result data. The preliminary audit issue qualitative result data includes issue type, qualitative conclusion, legal basis, and penalty recommendation.

[0045] S23. Conduct multi-dimensional cross-validation: On the one hand, call the historical audit case library and use a similarity algorithm to compare and verify the preliminary qualitative results with the qualitative conclusions of similar historical cases to generate case validation results; on the other hand, load the expert experience rule library to perform logical validation on the accuracy of the legal citations and the reasonableness of the liability definition of the preliminary qualitative results to generate rule validation results.

[0046] S24. By integrating case verification results and rule verification results, deviations in the preliminary qualitative results are corrected, the severity level of the problem and the attribution of responsibility are clarified, and authoritative and accurate qualitative data on audit problems are generated.

[0047] The working principle and effects of the above technical solution are as follows: By standardizing and verifying the completeness of problem information, the standardization of data to be determined is improved, avoiding qualitative deviations caused by chaotic information formats and missing key elements; the model quickly analyzes core features and matches them with knowledge chains, significantly improving qualitative efficiency and eliminating the need to spend a lot of time manually searching for regulations and sorting out the basis; multi-dimensional cross-validation is very practical, which can not only refer to historical cases to ensure consistency in qualitative analysis, but also ensure accurate citation of regulations and reasonable definition of responsibility through expert rule verification, thus enhancing the authority of the qualitative results; the precise correction of deviation items and the clear classification of levels and responsibilities effectively avoid disputes over different penalties for the same mistake and reduce the dispute rate in subsequent reviews.

[0048] In one embodiment of the present invention, S24 includes:

[0049] The case verification results (including historical case matching degree and qualitative differences) and the rule verification results (including compliance of legal citation and reasonableness of responsibility definition) are aligned and semantically associated to generate a fused verification dataset.

[0050] Based on the fusion verification dataset, the difference comparison algorithm is used to locate the deviation items in the preliminary qualitative results that conflict with historical cases and expert rules, including omissions of legal basis, ambiguity of responsible parties, and deviations in qualitative conclusions, and a list of deviation items is generated.

[0051] Based on the legal and regulatory base and expert experience base of the intelligent audit knowledge cloud brain, each problem in the deviation identification list is corrected in a targeted manner, missing evidence is supplemented, responsibility boundaries are clarified, qualitative conclusions are calibrated, and corrected qualitative intermediate data is generated.

[0052] Based on the revised qualitative intermediate data, and referring to the preset three-dimensional classification standard of violation details + scope of impact + degree of loss, the severity level of the problem is assessed; combined with the responsibility matrix, the responsible departments, responsible positions and specific responsible persons are identified, and level and responsibility definition data are generated.

[0053] The revised qualitative conclusions, complete legal basis, clear attribution of responsibility and severity level are logically integrated, and the consistency verification algorithm is used to ensure that the data is consistent, so as to generate authoritative and accurate qualitative data on audit issues.

[0054] The working principle and effects of the above technical solution are as follows: By aligning fields and semantically associating the two types of verification results, data fusion is made smoother, improving the consistency of the fused verification dataset and avoiding judgment confusion caused by conflicts in verification information from different sources; the difference comparison algorithm accurately locates deviation items, quickly identifying issues such as regulatory omissions and ambiguous responsibilities, reducing the workload of manual deviation investigation and preventing key errors from being overlooked; relying on the knowledge cloud brain for targeted correction not only supplements complete evidence but also clarifies the boundaries of responsibility, enhancing the accuracy and completeness of qualitative conclusions and leaving no room for ambiguity in the qualitative assessment of problems; the combination of the three-dimensional level classification standard and the responsibility matrix makes the determination of the severity of problems and the attribution of responsibility more objective and fair, avoiding disputes over different levels for the same mistake and the difficulty in defining responsibilities; finally, consistency verification ensures that the data is free of contradictions, clearing obstacles for subsequent root cause analysis and rectification, making the audit qualitative assessment more rigorous and efficient.

[0055] In one embodiment of the present invention, S3 includes:

[0056] S31. Based on accurate qualitative data of audit issues, automatically capture related financial data (such as expenditure details and budget execution data) and business data (such as procurement records and approval flow data), identify anomalies through data anomaly detection algorithms, including data inconsistencies and logical contradictions, and generate data dimension tracing results;

[0057] S32. Use process modeling tools to reconstruct the business processes and approval chains involved in the problem, trace the operation records of key nodes, identify process defects, including process breakpoints, permission vulnerabilities and non-compliant operations, and generate process dimension tracing results.

[0058] S33. Based on the organizational responsibility matrix and permission configuration information, identify the responsible departments, positions and specific individuals related to the problem, clarify the performance of responsibilities and boundaries of each entity, and generate responsibility dimension tracing results;

[0059] S34. Integrate the results of data, process, and responsibility tracing to uncover the institutional loopholes, management shortcomings, or cognitive biases behind the problem, and form a root cause analysis report. The root cause analysis report includes the root cause type, scope of impact, and risk level.

[0060] S35. Input the root cause analysis report data into the intelligent rectification knowledge base, and use the association algorithm to match the corresponding rectification template, standard action and implementation path to generate initial rectification strategy data. The initial rectification strategy data covers rectification goals, core measures and time nodes.

[0061] S36. Evaluate the initial rectification strategy data from four dimensions: technical feasibility, cost controllability, compliance, and long-term effectiveness. Adjust and optimize the data based on the actual business scenarios of the auditee, supplement specific implementation details and resource allocation suggestions, and generate accurate rectification strategy data.

[0062] The working principle and effects of the above technical solution are as follows: Three-dimensional source tracing delves deeply layer by layer, automatically capturing related data and identifying anomalies, quickly reconstructing business processes and identifying responsible parties. This improves the comprehensiveness of root cause analysis while reducing the tedious workload of manual investigation, avoiding situations where only surface phenomena are addressed and deeper issues are overlooked. Integrating three-dimensional source tracing information to uncover core problems such as system loopholes and management shortcomings enhances the pertinence of rectification strategies, effectively avoiding superficial and repetitive rectification. Intelligent matching of rectification templates and implementation paths reduces the time cost and trial-and-error risk of strategy development, eliminating the need to build rectification plans from scratch.

[0063] In one embodiment of the present invention, S35 includes:

[0064] The root cause analysis report data is structured and parsed to extract key features, including root cause type, scope of impact, risk level, and business areas involved, generating a standardized root cause feature dataset.

[0065] Based on a standardized root cause feature dataset, three-dimensional matching rules are set for root cause type, risk level and business scenario. The association algorithm of the intelligent rectification knowledge base is called to retrieve the appropriate rectification template and standard action library and generate a candidate rectification resource set.

[0066] By calculating semantic similarity and assigning weights to historical rectification results, the candidate rectification resource set is prioritized and the core rectification elements that are most closely related to the root cause are selected to generate the sorted rectification resource set.

[0067] Based on the sorted set of rectification resources, combined with the scope of the problem's impact and business process logic, core rectification measures are linked together, phased implementation steps and time nodes are planned, and rectification strategy framework data is generated.

[0068] The data of the rectification strategy framework is logically verified, and quantitative indicators of rectification targets and descriptions of key execution nodes are added to ensure that the strategy covers the root cause governance needs and generate initial rectification strategy data, which includes rectification targets, core measures and time nodes.

[0069] The working principle and effects of the above technical solution are as follows: By structurally analyzing the root cause report, core features are accurately extracted, improving data standardization and avoiding matching deviations caused by fragmented information and ambiguous features; the three-dimensional matching rules are highly targeted, capable of quickly retrieving suitable rectification templates and standard actions, and adapting to different root cause types, risk levels, and business scenarios, enhancing the accuracy of resource matching and reducing interference from irrelevant rectification resources; semantic similarity calculation combined with historical rectification effectiveness weighting can filter out the core elements with the highest fit, making the rectification direction clearer and reducing the trial-and-error cost of strategy formulation; measures are linked according to business logic and time nodes are planned, making the strategy framework clear and avoiding problems of chaotic rectification steps and poor connection; finally, logical integrity verification supplements quantitative indicators and execution nodes to ensure that the strategy covers the root cause governance needs, leaving no ambiguity in rectification, and the generated initial rectification strategy is both practical and easy to implement, laying a solid foundation for subsequent optimization.

[0070] In one embodiment of the present invention, step S4 includes:

[0071] S41. Decompose the precise rectification strategy data into specific and executable rectification tasks, clarify the responsible parties, execution standards, completion deadlines and assessment indicators for each task, and integrate them to form rectification execution plan data. The rectification execution plan data includes a task list, resource allocation plan and risk response plan.

[0072] S42. The audited entity promotes rectification work based on the rectification implementation plan data. The system collects rectification process data in real time through various means, including online process tracking and offline voucher uploading. The rectification process data includes the implementation status of rectification measures, phased results, and problems encountered.

[0073] S43. Based on the preset rectification effectiveness evaluation index system (including quantitative indicators such as the amount of funds recovered from violations and the number of system revisions, and qualitative indicators such as the degree of process optimization and the effect of improving personnel compliance awareness), combined with the data analysis capabilities of the big data model, the rectification process data is analyzed in real time to generate rectification effectiveness evaluation data.

[0074] S44. Compare the rectification effectiveness evaluation data with the preset compliance threshold. If the threshold is not met, trigger the backtracking mechanism and return to the strategy generation module to conduct root cause analysis and rectification strategy formulation again. If the threshold is met, generate rectification compliance confirmation data and enter the subsequent solidification process.

[0075] The working principle and effects of the above technical solution are as follows: By breaking down the precise rectification strategy into specific tasks, clarifying the responsible parties, execution standards, and time limits, the rectification goals are clearer, the division of labor is more explicit, and the operability of the rectification is improved, avoiding delays caused by vague tasks and shirking of responsibility. The combination of online tracking and offline voucher uploading allows for real-time collection of rectification process data, enabling a comprehensive grasp of the actual progress, reducing regulatory blind spots caused by information asymmetry, and preventing problems such as rectification becoming a mere formality or falsely reporting results. A pre-set evaluation indicator system combining quantitative and qualitative methods, coupled with the data analysis capabilities of large language models, makes the effectiveness evaluation more objective and accurate, avoiding judgments and one-sided assessments based solely on surface data. The compliance comparison and retrospective mechanism is crucial; if standards are not met, a return to analysis and strategy formulation can be carried out, effectively avoiding potential risks such as incomplete rectification and recurring problems. This not only ensures that the rectification process is organized and resource allocation is more reasonable, but also reduces execution risks through risk contingency plans. The generated compliance confirmation data is authentic and reliable, laying a solid foundation for subsequent system consolidation, ensuring that audit rectification is truly implemented and solves problems.

[0076] In one embodiment of the present invention, step S5 includes:

[0077] S51. Extract effective core experiences from compliance rectification cases. These core experiences include rectification measures, process specifications, responsibility definition standards, and key points of risk prevention and control, forming reusable rectification knowledge modules.

[0078] S52. Integrate the rectification knowledge module with the company's internal audit system and business management system. Through compliance review and logical integration, formulate a system revision plan, which will be incorporated into the formal system after approval and generate system update data.

[0079] S53. Synchronize the updated system data, rectification effectiveness evaluation data, and newly generated rectification knowledge modules to the intelligent audit knowledge cloud brain, update the structured knowledge chain and matching rules in the knowledge base, optimize the training parameters of the large language model, and generate updated intelligent audit knowledge cloud brain data.

[0080] S54. Synchronize the updated intelligent audit knowledge cloud brain data to the full-process tools such as the audit problem accurate matching model, rectification strategy recommendation tool, and audit process management system to form a closed-loop empowerment system of qualitative analysis, source tracing, rectification, consolidation and updating, and generate continuously optimized audit operation support data.

[0081] The working principle and effects of the above technical solution are as follows: Core experiences are extracted from compliant cases to form reusable modules, improving the reusability of rectification knowledge, avoiding the detour of starting from scratch for similar problems, and reducing the time cost of repeated rectification. The knowledge modules are integrated with internal systems, and revised plans are formed after compliance review, making system updates more aligned with actual governance needs, enhancing the practicality of the systems, and avoiding the problem of systems being disconnected from rectification practices and becoming ineffective. The structured knowledge chain and model parameters of the knowledge cloud brain are updated simultaneously, ensuring the knowledge base and audit tools remain timely, improving the accuracy of subsequent qualitative and rectification work, and avoiding deviations caused by using old knowledge to address new problems. The closed-loop empowerment system connects the entire process of qualitative analysis, source tracing, rectification, consolidation, and updating, enabling continuous optimization and upgrading of audit tools, and continuously addressing organizational management shortcomings, truly transforming audit results into long-term governance capabilities, and creating a virtuous cycle of management optimization.

[0082] An embodiment of the present invention provides an apparatus for implementing the intelligent processing method for qualitative and corrective action tracking of audit issues based on a large model as described above, the apparatus comprising:

[0083] Model building module: Constructs an intelligent audit knowledge cloud brain to generate knowledge base data; based on this knowledge base data, extracts key audit elements from audit business scenarios to generate audit element extraction result data; trains an audit question matching model based on the audit element extraction result data to build an audit question accurate matching model;

[0084] Problem identification module: Based on the audit problem precise matching model, it performs second-level matching processing on problems in actual audit business to obtain preliminary audit problem identification results data; it performs multi-dimensional verification on the preliminary audit problem identification results data to generate accurate audit problem identification data;

[0085] Strategy Generation Module: Based on the qualitative data of accurate audit issues, it conducts a layer-by-layer penetrating analysis from three dimensions: data, process, and responsibility, tracing the root cause of the problem and generating root cause analysis data; based on the root cause analysis data, it matches the corresponding rectification strategies from the intelligent rectification knowledge base and generates initial rectification strategy data; it conducts feasibility assessment and optimization of the initial rectification strategy data to generate accurate rectification strategy data;

[0086] Rectification Verification Module: This module develops specific rectification action plans based on precise rectification strategy data, generating rectification action plan data; it conducts rectification work based on the rectification action plan data, and verifies the effectiveness in real time during the rectification process, generating rectification effectiveness verification data; it determines whether the rectification has met the standards based on the rectification effectiveness verification data; if not, it returns to the strategy generation module to re-analyze the root causes and formulate rectification strategies; if it has met the standards, it proceeds to the next step.

[0087] Data update module: Solidifies the compliant rectification cases into regulations, incorporates effective rectification measures and processes into the company's internal audit system, and generates solidified regulations data; updates the knowledge base in the intelligent audit knowledge cloud brain based on the solidified regulations data, and generates updated knowledge base data.

[0088] One embodiment of the present invention provides a non-transitory computer-readable storage medium storing a computer program, characterized in that the program is executed by a processor to implement the intelligent processing method for qualitative and remedial tracking of audit issues based on a large model as described above. Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of the present invention and their equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A method for qualitative analysis and intelligent rectification tracking of audit issues based on a large-scale model, characterized in that: The method includes: S1. Construct an intelligent audit knowledge cloud brain to generate knowledge base data; based on this knowledge base data, extract key audit elements from audit business scenarios and generate audit element extraction result data; train an audit problem matching model based on the audit element extraction result data to construct an audit problem accurate matching model. S2. Based on the audit issue precise matching model, perform second-level matching processing on issues in actual audit business to obtain preliminary audit issue qualitative results data; perform multi-dimensional verification on the preliminary audit issue qualitative results data to generate precise audit issue qualitative data; S3. Based on the precise audit problem qualitative data, conduct a layer-by-layer penetrating analysis from three dimensions: data, process, and responsibility, trace the root cause of the problem, and generate problem root cause analysis data; based on the problem root cause analysis data, match the corresponding rectification strategies from the intelligent rectification knowledge base to generate initial rectification strategy data; conduct feasibility assessment and optimization of the initial rectification strategy data to generate precise rectification strategy data; S4. Formulate specific rectification action plans based on precise rectification strategy data, and generate rectification action plan data; carry out rectification work based on rectification action plan data, and conduct real-time effectiveness verification during the rectification process, generating rectification effectiveness verification data; determine whether the rectification has met the standards based on the rectification effectiveness verification data; if it has not met the standards, return to the strategy generation module to re-analyze the root causes and formulate rectification strategies; if it has met the standards, proceed to the next step. S5. For compliant rectification cases, solidify them into a system, incorporate effective rectification measures and processes into the company's internal audit system, and generate system solidification data; update the knowledge base in the intelligent audit knowledge cloud brain based on the system solidification data, and generate updated knowledge base data.

2. The intelligent processing method for qualitative and rectification tracking of audit issues based on a large model as described in claim 1, characterized in that, S1 includes: S11. Regularly crawl policies and regulations from authoritative external channels using web crawling technology, and simultaneously collect internal knowledge resources to form multi-source heterogeneous audit knowledge raw data; S12. Perform clause-level slicing, semantic parsing, deduplication and cleaning, and format unification on the original data of multi-source heterogeneous audit knowledge, and perform dual-track indexing and annotation to generate a standardized audit knowledge base. S13. Based on natural language processing technology, analyze the content of the standardized audit knowledge base, automatically extract key audit elements, and form a structured audit element dataset; S14. Divide the structured audit element dataset into training, validation and test sets, integrate the rule reasoning engine and similarity matching engine of the large language model, optimize the model weight through parameter fine-tuning strategy, and build an accurate matching model for audit issues.

3. The intelligent processing method for qualitative and rectification tracking of audit issues based on a large model as described in claim 2, characterized in that, S14 includes: The structured audit element dataset is divided into training set, validation set and test set according to the preset ratio. The data is then labeled and the feature dimensions are unified to generate the standardized audit element dataset after division. The rule reasoning engine and similarity matching engine of the large language model are integrated into a module, and a standardized audit element dataset is accessed to generate the initial framework of the model. Based on the training set data, the LoRA lightweight fine-tuning strategy is used to iteratively optimize the parameters of the key layers of the model. The gradient descent algorithm is combined to adjust the weight allocation and generate the optimized model parameter configuration. The validation set iteratively verifies the model performance. The validation set is input into the initial framework of the model. The qualitative accuracy is calculated, the model effect is verified according to the corresponding indicators, and a model performance evaluation report is generated. Adjust the dual-engine collaborative logic and parameter thresholds based on the performance evaluation report, complete the final performance verification using the test set, and generate an audit issue precise matching model that can accurately map audit issues to knowledge chains.

4. The intelligent processing method for qualitative and rectification tracking of audit issues based on a large model as described in claim 1, characterized in that, The S2 includes: S21. Receive information on pending issues entered by auditors through the audit issue characterization and audit rectification big data model system, perform data format conversion and integrity verification, and generate a standardized dataset of pending issues; S22. Input the standardized dataset of undetermined issues into the audit issue precise matching model. The model's semantic understanding module analyzes the core features of the issues, quickly retrieves matching knowledge chains in the standardized audit knowledge base, and outputs preliminary audit issue qualitative results data. S23. Perform multi-dimensional cross-validation to generate case validation results and rule validation results; S24. By integrating the case verification results and rule verification results, deviations in the preliminary qualitative results are corrected to generate accurate qualitative data on audit issues.

5. The intelligent processing method for qualitative and corrective tracking of audit issues based on a large model as described in claim 4, characterized in that, S24 includes: The case validation results and rule validation results are aligned by fields and semantically associated to generate a fused validation dataset. Based on the fusion verification dataset, the difference comparison algorithm is used to locate the deviation items in the preliminary qualitative results that conflict with historical cases and expert rules, and a list of deviation items is generated. Based on the legal and regulatory base and expert experience base of the intelligent audit knowledge cloud brain, each issue in the deviation identification list is corrected in a targeted manner, and qualitative intermediate data after correction is generated. Based on the revised qualitative intermediate data, and referring to the preset three-dimensional classification standard of violation details + scope of impact + degree of loss, the severity level of the problem is assessed; combined with the responsibility matrix, the responsible departments, responsible positions and specific responsible persons are identified, and level and responsibility definition data are generated. The revised qualitative conclusions, complete legal basis, clear attribution of responsibility and severity level are logically integrated, and the consistency verification algorithm is used to ensure that the data is consistent, so as to generate authoritative and accurate qualitative data on audit issues.

6. The intelligent processing method for qualitative and corrective tracking of audit issues based on a large model as described in claim 1, characterized in that, The S3 includes: S31. Based on accurate qualitative data of audit issues, automatically capture related financial and business data, identify anomalies through data anomaly detection algorithms, and generate data dimension tracing results; S32. Use process modeling tools to reconstruct the business processes and approval chains involved in the problem, trace the operation records of key nodes, identify process defects, and generate process dimension tracing results. S33. Based on the organizational responsibility matrix and permission configuration information, identify the responsible departments, positions and specific individuals related to the problem, clarify the performance of responsibilities and boundaries of each entity, and generate responsibility dimension tracing results; S34. Integrate the results of data, process, and responsibility tracing to uncover the institutional loopholes, management shortcomings, or cognitive biases behind the problem and form a root cause analysis report. S35. Input the root cause analysis report data into the intelligent rectification knowledge base, and use the association algorithm to match the corresponding rectification templates, standard actions and implementation paths to generate initial rectification strategy data. S36. Evaluate the initial rectification strategy data from four dimensions: technical feasibility, cost controllability, compliance, and long-term effectiveness. Adjust and optimize the data based on the actual business scenarios of the auditee, supplement specific implementation details and resource allocation suggestions, and generate accurate rectification strategy data.

7. The intelligent processing method for qualitative and rectification tracking of audit issues based on a large model as described in claim 1, characterized in that, The S4 includes: S41. Decompose the precise rectification strategy data into specific and actionable rectification tasks, clarify the responsible parties, implementation standards, completion deadlines and assessment indicators for each task, and integrate them to form rectification implementation plan data; S42. The audited entity promotes rectification work based on the rectification implementation plan data, and the system collects rectification process data in real time through multiple methods; S43. Based on the preset rectification effectiveness evaluation index system, combined with the data analysis capabilities of the large language model, the rectification process data is analyzed in real time to generate rectification effectiveness evaluation data. S44. Compare the rectification effectiveness evaluation data with the preset compliance threshold. If the threshold is not met, trigger the backtracking mechanism and return to the strategy generation module to conduct root cause analysis and rectification strategy formulation again. If the threshold is met, generate rectification compliance confirmation data and enter the subsequent solidification process.

8. The intelligent processing method for qualitative and rectification tracking of audit issues based on a large model as described in claim 1, characterized in that, The S5 includes: S51. Extract effective core experiences from compliance rectification cases and form reusable rectification knowledge modules; S52. Integrate the rectification knowledge module with the company's internal audit system and business management system. Through compliance review and logical integration, formulate a system revision plan, which will be incorporated into the formal system after approval and generate system update data. S53. Synchronize the updated system data, rectification effectiveness evaluation data, and newly generated rectification knowledge modules to the intelligent audit knowledge cloud brain, update the structured knowledge chain and matching rules in the knowledge base, optimize the training parameters of the large language model, and generate updated intelligent audit knowledge cloud brain data. S54. Synchronize the updated intelligent audit knowledge cloud brain data to the full-process tools such as the audit problem accurate matching model, rectification strategy recommendation tool, and audit process management system to form a closed-loop empowerment system of qualitative analysis, source tracing, rectification, consolidation and updating, and generate continuously optimized audit operation support data.

9. An apparatus for implementing the intelligent processing method for qualitative and corrective action tracking of audit issues based on a large model as described in claim 1, characterized in that, The device includes: Model building module: Constructs an intelligent audit knowledge cloud brain to generate knowledge base data; based on this knowledge base data, extracts key audit elements from audit business scenarios to generate audit element extraction result data; trains an audit question matching model based on the audit element extraction result data to build an audit question accurate matching model; Problem identification module: Based on the audit problem precise matching model, it performs second-level matching processing on problems in actual audit business to obtain preliminary audit problem identification results data; it performs multi-dimensional verification on the preliminary audit problem identification results data to generate accurate audit problem identification data; Strategy Generation Module: Based on the qualitative data of accurate audit issues, it conducts a layer-by-layer penetrating analysis from three dimensions: data, process, and responsibility, tracing the root cause of the problem and generating root cause analysis data; based on the root cause analysis data, it matches the corresponding rectification strategies from the intelligent rectification knowledge base and generates initial rectification strategy data; it conducts feasibility assessment and optimization of the initial rectification strategy data to generate accurate rectification strategy data; Rectification Verification Module: This module develops specific rectification action plans based on precise rectification strategy data, generating rectification action plan data; it conducts rectification work based on the rectification action plan data, and verifies the effectiveness in real time during the rectification process, generating rectification effectiveness verification data; it determines whether the rectification has met the standards based on the rectification effectiveness verification data; if not, it returns to the strategy generation module to re-analyze the root causes and formulate rectification strategies; if it has met the standards, it proceeds to the next step. Data update module: Solidifies the compliant rectification cases into regulations, incorporates effective rectification measures and processes into the company's internal audit system, and generates solidified regulations data; updates the knowledge base in the intelligent audit knowledge cloud brain based on the solidified regulations data, and generates updated knowledge base data.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the intelligent processing method for qualitative and remedial tracking of audit issues based on a large model as described in any one of claims 1 to 8.

Images