A financial fraud identification method integrated into a corporate governance thought chain

Abstract

This invention discloses a method for identifying financial fraud by integrating a corporate governance thinking chain, comprising: receiving publicly disclosed texts from listed companies; constructing a corporate governance thinking chain consisting of seven corporate governance dimensions, sixteen key governance issues, sixty-four atomic governance indicators, a fraud hexagonal element mapping, and a governance defect to risk signal transmission template; extracting all indicators from the governance text, identifying and annotating situations where information is undisclosed, unmentioned, has only a title but no content, or is insufficient for calculation; generating core analytical conclusions based on the reasoning sequence of "governance facts - governance defects - fraud conditions - risk transmission - financial fraud identification conclusions"; calculating dimensional risk scores and overall risk scores, and outputting financial fraud identification results and structured results. This invention couples corporate governance theory, indicator extraction rules, and the fraud hexagonal mapping into a unified reasoning chain, aiming to improve the completeness, interpretability, and verifiability of financial fraud identification.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, natural language processing, corporate governance analysis, and financial fraud identification. Specifically, it relates to a financial fraud identification method that integrates a corporate governance thinking chain into a unified approach, which couples full extraction of corporate governance text, judgment of governance defects, fraud hexagonal mapping, and risk transmission reasoning into a unified method. Background Technology

[0002] With the continuous development of the capital market, regulatory agencies and investors are increasingly concerned about the authenticity and completeness of listed companies' financial information and the effectiveness of their corporate governance. Traditional methods of identifying financial fraud mainly rely on manual reading of annual reports, audit reports, internal control reports, and interim announcements, combined with experience to judge abnormal transactions, abnormal disclosures, and abnormal financial performance. This method has limited efficiency and is easily affected by subjective experience, the scope of reading, and the ability to integrate across documents. In particular, it is difficult to conduct continuous and systematic correlation analysis of corporate governance clues scattered across different chapters, announcements, and periods.

[0003] Against the backdrop of deepening reforms and increasingly stringent regulations in my country's capital market, institutional requirements related to financial fraud and corporate governance are rapidly being improved. On April 4, 2024, the "Opinions of the State Council on Strengthening Supervision, Preventing Risks, and Promoting High-Quality Development of the Capital Market" explicitly proposed strict ongoing supervision of listed companies, the construction of a comprehensive system for preventing and combating fraud in the capital market, and severe crackdowns on illegal and irregular activities such as financial fraud. On July 5, 2024, the "Opinions on Further Improving the Comprehensive Prevention and Punishment of Financial Fraud in the Capital Market" further emphasized the need to improve regulatory coordination mechanisms, build a long-term mechanism for preventing and combating financial fraud, and focus on cracking down on systemic fraud, fictitious transactions by related parties, collusion in fraud, and the abuse of accounting policies to manipulate profits. On August 1, 2023, the "Measures for the Management of Independent Directors of Listed Companies" clarified that independent directors should participate in board decision-making, supervise matters involving potential major conflicts of interest, and may independently engage intermediaries in accordance with the law. On March 26, 2025, the "Administrative Measures for Information Disclosure of Listed Companies" and the "Guidelines for the Content and Format of Information Disclosure by Companies Issuing Securities to the Public No. 2—Content and Format of Annual Reports" (March 21, 2025) further required listed companies to continuously disclose important information such as the basic status of corporate governance, the situation of controlling shareholders and actual controllers, non-operating fund occupation by related parties, the supervisory activities of the audit committee, and bonds and debt repayment. On October 16, 2025, the "Corporate Governance Guidelines for Listed Companies" further strengthened the requirements for incentives and constraints on directors and senior executives, the code of conduct for controlling shareholders and actual controllers, and the responsibility for reviewing related-party transactions.

[0004] Publicly disclosed information from listed companies serves as a crucial data source for identifying financial fraud, exhibiting characteristics of multi-dimensionality, multi-format nature, large scale, and strong correlation. For instance, information on equity structure and actual controllers is typically scattered throughout the corporate governance section of annual reports and interim announcements; related-party transactions, external guarantees, major litigation and arbitration, and administrative penalties are often distributed across significant matters, special announcements, and responses to inquiries in annual reports; audit opinions, key audit matters, and internal control deficiencies are reflected in audit reports, internal control evaluation reports, and related special explanations, respectively; and bonds and debt repayment arrangements may also appear in bond matters, prospectuses, and credit rating documents in annual reports. The large amount of key information stored in various formats, including text descriptions, tables, notes, announcements, and PDF documents, presents significant challenges for automated extraction, cross-text correlation, and consistency identification. For large enterprise groups or batch identification scenarios involving multiple listed companies, issues such as differences in disclosure standards, complex formats, scattered chapters, and difficulties in continuous tracking across time periods are particularly prominent.

[0005] In recent years, research on corporate governance, text analysis, and machine learning has provided a new technological foundation for financial fraud identification. Patrick Velte (2023) pointed out that there is a systematic link between corporate governance and financial restatements, law enforcement activities, and fraud, but the impact of different governance variables on financial misconduct is significantly heterogeneous. Jing Li (2025) showed that introducing corporate governance indicators into financial fraud identification models can significantly improve model accuracy. Ludivia HernandezAros (2024), after a systematic review of 104 studies, concluded that although research on financial fraud identification has developed rapidly, it remains relatively fragmented in terms of models, datasets, and feature construction paths. Khrystyna Bochkay (2023) further pointed out that the construct validity of text metrics remains a key issue in accounting text analysis research.

[0006] Meanwhile, publicly disclosed texts themselves have become important evidentiary carriers for identifying fraud and misstatements. Jingyu Li et al. (2024) showed that unusual tone of management plays an important role in identifying fraud and misstatements in a comprehensive indicator system that includes financial, non-financial, and textual features. Xin Huang et al. (2024) showed that the readability, tone, and level of detail of key audit matter texts are significantly correlated with the probability of financial restatements. Chao Zhang et al. (2024) showed that the textual features of regulatory inquiry letters have predictive value for financial restatements. Xiaoqian Zhu et al. (2025) showed that annual report risk disclosure texts can improve the accuracy of accounting fraud identification and play an early warning role.

[0007] The development of big data, natural language processing, multimodal document understanding, and large language models has provided new possibilities for the automated analysis of complex governance texts. Yinheng Li et al. (2023) and Yaxuan Kong et al. (2024) showed that large language models have strong application potential in financial text understanding, question answering, and analysis tasks. Dongsheng Wang et al. (2024) and Yuliang Liu et al. (2024) indicated that layout awareness and OCR-free document understanding capabilities are helpful in processing complex visual documents such as reports and tables. However, Yifei Dong et al. (2025) pointed out that the application of large language models in financial scenarios still faces challenges related to professional adaptation and real-world business constraints; Darren Yow-Bang Wang et al. (2025) indicated that the consistency of structured output remains a key technical challenge.

[0008] However, existing methods still have significant shortcomings: First, existing research often revolves around single financial indicators, single governance variables, or single text fragments, lacking a unified corporate governance logic chain that follows the progression of "control basis—channels for transferring benefits—tools for risk transfer—signals of violations—incentive and constraint mechanisms—supervision and checks and balances mechanisms—external contractual pressure." This results in an unclear transmission relationship between governance anomalies and financial fraud risks, and insufficient interpretability.

[0009] Second, information related to corporate governance and financial fraud of listed companies is scattered across multiple sources of text, such as annual reports, audit reports, internal control reports, independent directors' performance materials, responses to regulatory inquiries, and bond documents. These are mostly complex PDF documents containing a large number of tables, notes, and page-spanning structures. Traditional plain text methods are difficult to reliably achieve consistent extraction and correlation analysis across documents, chapters, and periods.

[0010] Third, even with the introduction of large language models, the lack of a stable governance knowledge framework, inference order constraints, and structured output mechanisms can easily lead to problems such as missing fields, inconsistent conclusions, unverifiable reasoning processes, and inconsistent identification standards across different text scenarios. This makes it difficult to meet the requirements of completeness, consistency, and verifiability for financial fraud identification. These shortcomings are a comprehensive assessment based on recent research in corporate governance, text analysis, document understanding, and structured output.

[0011] Therefore, there is an urgent need for a method that can automatically, accurately, and structurally identify financial fraud in listed companies. This method would integrate corporate governance thinking into the identification process, focusing on key dimensions such as equity structure and actual controllers, related-party transactions, external guarantees, litigation, arbitration and major penalties, incentive and constraint mechanisms, audit information disclosure and compliance, and bonds and debt repayment arrangements. It would organize, extract, correlate, and reason about the multi-source public disclosure texts of listed companies, thereby making up for the shortcomings of existing technologies, such as insufficient utilization of governance information, lack of an interpretable reasoning chain between governance anomalies and fraud risks, failure to include missing disclosures in identification features, unverifiable output results, and inconsistent identification standards in different text scenarios. Summary of the Invention

[0012] The purpose of this invention is to provide a financial fraud identification method that integrates corporate governance thinking, in order to solve the problems existing in financial fraud identification, such as insufficient utilization of governance information, lack of an interpretable reasoning chain between governance anomalies and fraud risks, failure to include missing disclosures in identification features, unverifiable output results, and inconsistent identification standards in different text scenarios.

[0013] To achieve the objective of this invention, the technical solution adopted is as follows: A method for identifying financial fraud that integrates corporate governance thinking includes the following steps: S1. Receive the company name of the target to be analyzed and the publicly disclosed text of the listed company, document type, analysis period and input scope, perform layout parsing, paragraph segmentation, table recognition and plain text cleaning on the publicly disclosed text of the listed company, and obtain the cleaned plain text. S2. Construct a corporate governance thinking chain framework, which includes seven corporate governance dimensions, sixteen key governance issues, sixty-four atomic governance indicators, a fixed mapping relationship between corporate governance dimensions and the six-corner element of fraud, and a risk transmission template for each corporate governance dimension corresponding to governance defects to financial fraud risk signals. S3. Take the cleaned plain text from step S1 as the input text, extract all of the input text according to the sixty-four atomic governance indicators, generate indicator records for each atomic governance indicator, and retain indicator records that are not disclosed, not mentioned, have only a title, or have insufficient information to be calculated. S4. Identify the occurrence status of each indicator record, and determine the abnormal feature flag, abnormal feature type, indicator value, indicator value acquisition method, severity, confidence level, and corresponding fraud hexagonal elements for each indicator record based on the occurrence status. S5. Following the corporate governance thinking chain sequence of “governance fact extraction → key governance issue determination → governance defect identification → fraud hexagonal element mapping → risk transmission reasoning → financial fraud identification conclusion generation”, reason about the indicator records to generate dimensional summary results and core analytical conclusions from governance defects to financial fraud risk signals. S6. Based on the corporate governance dimension to which each indicator belongs, the severity of each indicator recorded in step S4, and the dimension summary results generated in step S5, the sixty-four atomic governance indicators are grouped into the corresponding corporate governance dimensions according to G1 to G7, where G1 to G7 are the numbers of the seven corporate governance dimensions. The severity of all indicators recorded within each corporate governance dimension is averaged, and this average is multiplied by 100 / 3 and rounded to two decimal places to obtain the dimension risk score; the dimension risk scores of the seven corporate governance dimensions are averaged and rounded to two decimal places to obtain the overall risk score. Based on the overall risk score and preset threshold, the system outputs financial fraud identification results, and outputs structured results including indicator record set, dimension summary results, core analysis conclusions, overall risk score, final risk level, risk judgment basis, document summary and integrity verification results. S7. Based on step S6, perform an integrity check to verify whether all indicator records cover all sixty-four atomic governance indicators, and verify whether the actual output indicator quantity of each corporate governance dimension is consistent with the preset quantity. After the check passes, output the final financial fraud identification result and structured result.

[0014] As a preferred method, the publicly disclosed texts of listed companies in step S1 of the financial fraud identification method are input from external data sources, including: one or more of the following: the corporate governance section of the annual report in PDF format, notes to the annual report, audit report and audit opinion, internal control or compliance self-inspection report, exchange inquiry letter and inquiry response, announcement of major events, bond prospectus and credit rating documents.

[0015] Preferably, in step S1 of the financial fraud identification method, the document type is used to identify the disclosure document category of the publicly disclosed text of the listed company; the analysis period is used to mark the fiscal year, semi-annual, quarterly, specific reporting period or announcement date range covered by the publicly disclosed text of the listed company; and the input range is used to identify the coverage of the input text in step S2, including three categories: full document, excerpt, and unknown range.

[0016] As a preferred approach, the seven corporate governance dimensions in step S2 of the financial fraud identification method include: equity structure and actual controller; related-party transactions; external guarantees; litigation, arbitration and major penalties; incentive and constraint mechanisms; audit, information disclosure and compliance; and bonds and debt repayment arrangements.

[0017] As a preferred approach, the seven corporate governance dimensions in step S2 of the financial fraud identification method sequentially cover the basis of control, channels for transferring benefits, risk transfer tools, signs of violations, mechanisms of behavioral motivation, mechanisms of supervision and checks and balances, and external contractual pressure.

[0018] As a preferred approach, the sixteen key governance issues in step S2 of the financial fraud identification method include: whether control is excessively concentrated or insufficiently checked; whether the control structure is unbalanced; whether the actual controller has special governance risks; whether the scale of related-party transactions is abnormal; whether the structure of related-party transactions is abnormal; whether there are abnormal related-party transactions; whether the scale of guarantees is too large; whether there are procedural or substantive risks in guarantees; whether there are major lawsuits or arbitrations; whether regulatory penalties are significant; whether the incentive intensity is too high; whether the constraint mechanism is weak; whether audit supervision is effective; whether there are problems with internal control and disclosure compliance; whether debt pressure is concentrated; and whether financing and rating pressures are increasing.

[0019] As a preferred embodiment, the sixty-four atomic governance indicators in step S2 of the financial fraud identification method are configured as follows: G1 = 8, G2 = 11, G3 = 8, G4 = 9, G5 = 9, G6 = 11, and G7 = 8. Among them, G1 to G7 are the numbers of seven corporate governance dimensions: G1 corresponds to equity structure and actual controller; G2 corresponds to related-party transactions; G3 corresponds to external guarantees; G4 corresponds to litigation, arbitration and major penalties; G5 corresponds to incentive and restraint mechanisms; G6 corresponds to audit, information disclosure and compliance; and G7 corresponds to bonds and debt repayment arrangements.

[0020] As a preferred option, the sixty-four atomic governance indicators in step S2 of the financial fraud identification method are as follows: 8 G1 items: G1-Q1-I1 Shareholding Ratio of the Largest Shareholder; G1-Q1-I2 Concentration of the Top Ten Shareholders; G1-Q1-I3 Herfindahl-Hirschman Index (HHI); G1-Q2-I1 Control-Cash Flow Dispersion; G1-Q2-I2 Dual-Class Shares / Special Voting Rights / Voting Rights Arrangements; G1-Q3-I1 Type of Actual Controller; G1-Q3-I2 Change of Actual Controller / Instable Control / No Actual Controller; G1-Q3-I3 Control Risks such as Share Pledge, Freezing, and Judicial Auction. 11 G2 items: G2-Q1-I1 Related Party Transaction Amount; G2-Q1-I2 Related Party Transaction Amount / Operating Revenue; G2-Q1-I3 Related Party Transaction Amount / Total Assets; G2-Q2-I1 Related Party Sales Ratio; G2-Q2-I2 Related Party Procurement Ratio; G2-Q2-I3 Related Party Fund Transactions Ratio; G2-Q2-I4 Other Receivables / Related Party Occupation; G2-Q2-I5 Related Party Guarantee Ratio; G2-Q3-I1 Abnormal Related Party Transactions; G2-Q3-I2 Unfair Pricing / Circular Transactions / Year-End Sudden Transactions; G2-Q3-I3 Abnormal Approval and Disclosure Procedures; 8 G3 items: G3-Q1-I1 Guarantee Balance; G3-Q1-I2 Guarantee Balance / Net Assets; G3-Q1-I3 Guarantee Balance / Total Assets; G3-Q1-I4 Number of Guarantees; Whether G3-Q2-I1 Has Any Irregular Guarantees; Whether G3-Q2-I2 Has Any Implicit Guarantees; Whether G3-Q2-I3 Lacks Counter-Guarantees; Whether G3-Q2-I4 Has Any Guarantees Involved in Compensation or Litigation; 9 G4 items: G4-Q1-I1: Are there any major lawsuits / arbitrations? G4-Q1-I2: Number of major lawsuits / arbitrations; G4-Q1-I3: Amount involved; G4-Q1-I4: Amount involved / Net assets; G4-Q1-I5: Amount involved / Total assets; G4-Q2-I1: Number of regulatory penalties; G4-Q2-I2: Severity of penalties; G4-Q2-I3: Is there any case under investigation? G4-Q2-I4: Are there any major compliance incidents? 9 G5 items: G5-Q1-I1 Sensitivity to Stock Option Incentives; G5-Q1-I2 Total Executive Compensation; G5-Q1-I3 Bonus Ratio; G5-Q1-I4 Existence of Equity Incentive Plan; G5-Q2-I1 Board Independence; G5-Q2-I2 Audit Committee Characteristics; G5-Q2-I3 Board Meeting Frequency; G5-Q2-I4 Supervisory Board Oversight; G5-Q2-I5 Internal Oversight. 11 items in G6: G6-Q1-I1 Auditor size; G6-Q1-I2 Audit opinion type; G6-Q1-I3 Exceptions in emphasis of matter / key audit matters; G6-Q1-I4 Auditor change; G6-Q1-I5 Audit fees; G6-Q1-I6 Exceptional audit fees; G6-Q2-I1 Whether there are material weaknesses in internal control; G6-Q2-I2 Internal control audit opinion type; G6-Q2-I3 Information disclosure evaluation; G6-Q2-I4 Frequency of regulatory inquiries; G6-Q2-I5 Whether there are supplementary disclosures / corrected disclosures / delayed disclosures? 8 G7 items: G7-Q1-I1 Debt-to-equity ratio; G7-Q1-I2 Short-term debt ratio; G7-Q1-I3 Interest coverage ratio; G7-Q1-I4 Bond maturity concentration; G7-Q2-I1 Credit rating; G7-Q2-I2 Rating outlook change; G7-Q2-I3 Bond spread / yield; G7-Q2-I4 Whether there is a put option / rollover / default / cross-default. In the atomic governance indicator names, the slash " / " indicates a ratio relationship or parallel items according to the indicator semantics: in ratio-type atomic governance indicators, it means "divided by", and in parallel item-type atomic governance indicators, it means parallel selection or parallel listing, without changing the number of individual atomic governance indicators.

[0021] As a preferred embodiment, the corporate governance thinking chain framework in step S2 of the financial fraud identification method consists of the following layers in sequence: governance facts layer, governance problem layer, governance defect judgment layer, fraud condition mapping layer, risk transmission reasoning layer, and identification output layer.

[0022] In the risk transmission reasoning layer, regarding the equity structure and actual controller dimension, the focus is on identifying the weakening of checks and balances and the increase in the dominance of insiders due to the imbalance of control; regarding the related-party transaction dimension, the focus is on identifying situations where related-party purchases and sales, fund transfers, and related-party guarantees become vehicles for the transfer of benefits and profit adjustment; regarding the external guarantee dimension, the focus is on identifying contingent liabilities, compensation risks, and internal control failures; regarding the litigation, arbitration, and major penalties dimension, the focus is on identifying compliance failures, reputational damage, and profit and financing pressures; regarding the incentive and constraint mechanism dimension, the focus is on identifying the motivation for performance embellishment caused by the weakening of short-term assessments and supervision; regarding the audit, information disclosure, and compliance dimension, the focus is on identifying hidden opportunities formed by the weakening of audit, internal control, and disclosure defenses; regarding the bond and debt repayment arrangement dimension, the focus is on identifying the pressure to manipulate financial statements caused by concentrated maturities, rating downgrades, and liquidity shortages.

[0023] As a preferred approach, the fixed mapping relationship between the corporate governance dimension and the hexagonal elements of fraud in step S2 of the financial fraud identification method is as follows: equity structure and actual controller correspond to opportunity, capability, collusion, and arrogance; related-party transactions correspond to opportunity, collusion, and rationalization; external guarantees correspond to pressure, opportunity, and collusion; litigation, arbitration, and major penalties correspond to pressure, opportunity, and rationalization; incentive and constraint mechanisms correspond to pressure, rationalization, arrogance, and capability; auditing, information disclosure, and compliance correspond to opportunity, collusion, and capability; bonds and debt repayment arrangements correspond to pressure, rationalization, and capability. Among these, pressure, opportunity, rationalization, capability, arrogance, and collusion are the six types of behavioral conditions in the hexagonal theory of fraud. Each corporate governance dimension corresponds to one or more hexagonal elements of fraud through a fixed mapping relationship, which are used to generate a set of hexagonal elements of fraud in step S4 and to explain the behavioral conditions for the transmission of governance defects to financial fraud risk signals in step S5.

[0024] The aforementioned fixed mapping relationship is determined based on the role of each governance dimension in the formation of financial fraud: equity structure and actual controllers determine the strength of insider control and checks and balances, thus mainly corresponding to opportunity, capability, collusion, and arrogance; related-party transactions are characterized by non-marketization, concealment, and the ability to be manipulated by related parties, thus corresponding to opportunity, collusion, and rationalization; external guarantees may create contingent liabilities and risk transfer pressure, while exposing gaps in approval and disclosure opportunities, thus corresponding to pressure, opportunity, and collusion; litigation, arbitration, and major penalties reflect compliance failures, reputational damage, and increased financing constraints, thus corresponding to pressure, opportunity, and rationalization; incentive and constraint mechanisms affect management's short-term performance motivation, accountability constraints, and ability to manipulate financial statements, thus corresponding to pressure, rationalization, arrogance, and capability; auditing, information disclosure, and compliance constitute external supervision and internal control defenses, and their weakening can create opportunities to hide fraud and may lead to collusion to distort disclosures, thus corresponding to opportunity, collusion, and capability; bonds and debt repayment arrangements reflect debt contracts, rating maintenance, and refinancing pressures, and require management to have the ability to allocate funds and package financial statements, thus corresponding to pressure, rationalization, and capability.

[0025] Preferably, the indicator status in step S4 of the financial fraud identification method is selected from the following seven categories: observed_value, explicit_none, partial_disclosure_but_not_computable, section_present_but_blank, explicit_not_applicable, not_found_in_input, and missing_required_disclosure (if the entire document should be disclosed). Wherein, when the input scope is the entire document... When a full document of an indicator is not found, it is identified as a missing document that should be disclosed (missing_required_disclosure); when the input range is an excerpt or an unknown range of text and no relevant content is found, it is identified as not found in the current input (not_found_in_input); when relevant content is found but the indicator calculation cannot be completed, it is identified as partially disclosed but not computable (partial_disclosure_but_not_computable); when only a title, table header, or chapter name exists but no substantive content, it is identified as a chapter that exists but the content is blank (section_present_but_blank).

[0026] Preferably, the abnormal feature flags and abnormal feature types in step S4 of the financial fraud identification method are determined according to the following rules: When the indicator's status is a clearly observed value and no clear governance defect is observed, the abnormal feature flag is 0, and the abnormal feature type is none; when the indicator's status is a clearly observed value and a governance defect, procedural anomaly, or high-risk fact is observed, the abnormal feature flag is 1, and the abnormal feature type is observed governance defect; when the indicator's status is partially disclosed but not computable, section exists but content is blank, or current input is not found... When a complete document should disclose something but is missing (missing_required_disclosure), the anomaly flag is 1, and the anomaly type is missing_disclosure_or_not_mentioned; when the indicator status is explicitly none (explicit_none) or explicitly not applicable (explicit_not_applicable), the anomaly flag is 0, and the anomaly type is disclosed_none_or_not_applicable; among them, the indicator value acquisition methods include three categories: direct extraction (direct_extract), computed from text (computed_from_text), and not_computable_in_current_input (not_computable_in_current_input).

[0027] As a preferred method, the dimension risk score of each corporate governance dimension in step S6 of the financial fraud identification method is calculated according to the following formula: dimension_score = round(average of severity of all indicators within the dimension × 100 / 3, 2).

[0028] As a preferred embodiment, the overall risk score in step S6 of the financial fraud identification method is calculated according to the following formula: overall_risk_score = round(average of the risk scores of the seven corporate governance dimensions, 2); when overall_risk_score is greater than or equal to 45, the financial fraud identification result is output as "moderately high"; when overall_risk_score is less than 45, the financial fraud identification result is output as "moderately low".

[0029] Preferably, the structured result in step S6 of the financial fraud identification method includes all of the following fields: Indicator record set (indicator_records), used to store the indicator code, indicator name, dimension, key governance issues, indicator occurrence status, anomaly characteristic flag, anomaly characteristic type, indicator value, indicator value acquisition method, supporting text, severity, confidence level, and fraud hexagonal elements for the sixty-four atomic governance indicators; Dimension summary result (dimension_summary), used to store the dimension number, dimension name, number of indicators to be output, number of indicators actually output, number of missing or undisclosed indicators, number of observed governance defect anomalies, dimension risk score, and dimension... The core analysis conclusions (core_analysis_conclusions) are used to store governance deficiencies, risk signals, the transmission chain from deficiencies to risk signals, correlation dimensions, correlation indicators, fraud hexagonal elements, supporting evidence, overall severity and confidence level; overall risk score; final risk level; risk judgment basis; document summary; and completeness check result.

[0030] Preferably, the integrity check in step S7 of the financial fraud identification method includes all of the following check items: Checking whether the indicator record set (indicator_records) covers all sixty-four atomic governance indicators and whether the indicator codes are consistent with the fixed indicator dictionary; Checking whether the dimension summary results (dimension_summary) cover all seven corporate governance dimensions; Checking the output of 8 items for the equity structure and actual controller dimension, 11 items for the related party transaction dimension, 8 items for the external guarantee dimension, 9 items for the litigation, arbitration and major penalty dimension, 9 items for the incentive and constraint mechanism dimension, 11 items for the audit, information disclosure and compliance dimension, and 8 items for the bond and debt repayment arrangement dimension; Checking whether each indicator record contains the indicator code (indicator_code), indicator name (indicator_name), dimension (dimension), and key governance issue (governance_query). The system outputs the following parameters: indicator presence status, abnormal feature flag, abnormal feature type, indicator value, indicator value acquisition method, supporting source evidence, severity, confidence, and hexagonal elements. The final risk level is verified to be either "low to medium" or "high to medium". If any verification item fails, a verification failure flag and a list of missing indicator codes or fields are output. Upon successful verification, the final financial fraud identification result and structured results are output.

[0031] The beneficial effects of this invention are as follows: This invention takes seven corporate governance dimensions, sixteen key governance issues and sixty-four atomic governance indicators as its core, and establishes a fixed thinking chain of "governance facts - governance defects - fraud conditions - risk transmission - financial fraud identification conclusions", so that corporate governance texts can directly serve the identification of financial fraud, rather than just extracting local indicators or general risk warnings. This invention includes situations such as undisclosed, unmentioned, only title without content, and insufficient information to calculate within the scope of abnormal feature identification, so that the governance disclosure gap itself can be stably transformed into financial fraud identification features, thereby improving governance transparency and the ability to utilize abnormal information. This invention improves the interpretability of financial fraud identification conclusions by establishing a fixed mapping relationship between corporate governance dimensions and the hexagonal elements of fraud, as well as risk transmission templates corresponding to the seven dimensions. This enables the identification results to not only indicate the "level of risk" but also to explain "how governance deficiencies are transformed into risk signals along the thought chain." This invention improves the standardization, comparability, and verifiability of analysis results from different companies, different text types, and different periods by using an integrity verification mechanism, unified status classification rules, and unified severity and overall risk assessment rules. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall process structure of the financial fraud identification method that incorporates corporate governance thinking into the present invention. Detailed Implementation

[0033] The present invention will be further illustrated below using an example of the publicly disclosed documents from Kangmei Pharmaceutical Co., Ltd. in 2018. These documents include publicly disclosed materials such as annual reports, audit reports, internal control reports, responses to inquiry letters, and administrative penalty decisions.

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0035] In the description of this invention, it should be understood that the terms "upper", "lower", "horizontal", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship of the technical solution, and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0036] like Figure 1 As shown, a financial fraud identification method integrating corporate governance thinking chain includes seven steps: text input and scope identification, corporate governance thinking chain framework construction, full indicator extraction, indicator status identification, corporate governance thinking chain reasoning, risk assessment and structured output, and integrity verification. S1. Text Input and Scope Identification: This function receives the company name and publicly disclosed text of the listed company, document type, analysis period, and input scope. It performs format parsing, paragraph segmentation, table recognition, and plain text cleaning on the publicly disclosed text to obtain the cleaned plain text. The publicly disclosed text is input from external data sources, including: one or more of the following: corporate governance chapters in PDF annual reports, notes to the annual report, audit reports and audit opinions, internal control or compliance self-inspection reports, exchange inquiry letters and responses, announcements of major events, bond prospectuses, and credit rating documents. The document type identifies the disclosure document category of the publicly disclosed text. The analysis period marks the fiscal year, semi-annual, quarterly, specific reporting period, or announcement date range covered by the publicly disclosed text. The input scope identifies the coverage level of the input text, including three categories: full document (full_document), excerpt (excerpt), and text with unknown scope (unknown), to support subsequent risk level identification. S2. Construction of Corporate Governance Thinking Chain Framework: Construct a corporate governance thinking chain framework, which includes seven corporate governance dimensions, sixteen key governance issues, sixty-four atomic governance indicators, fixed mapping relationships between corporate governance dimensions and the six-corner element of fraud, and risk transmission templates for governance defects in each corporate governance dimension to financial fraud risk signals. S3. Full extraction and status identification of indicators: The cleaned plain text in step S1 is used as the input text. The input text is fully extracted according to the sixty-four atomic governance indicators. An indicator record is generated for each atomic governance indicator, and the indicator records that are not disclosed, not mentioned, have only a title or are too inaccurate to be calculated are retained. S4. Indicator Status Identification: Identify the status of each indicator record and determine the abnormal feature flag, abnormal feature type, indicator value, indicator value acquisition method, severity, confidence level, and corresponding fraud hexagonal elements for each indicator record based on the indicator status. S5. Corporate Governance Thinking Chain Reasoning: Following the corporate governance thinking chain sequence of "Governance Fact Extraction → Key Governance Issue Judgment → Governance Deficiency Identification → Fraud Hexagonal Element Mapping → Risk Transmission Reasoning → Financial Fraud Identification Conclusion Generation", the indicator records are reasoned to generate dimensional summary results and core analytical conclusions from governance deficiencies to financial fraud risk signals. This step is not a simple restatement of the identified indicators, but rather a layer-by-layer reasoning process based on the governance risk main chain across seven dimensions. First, at the governance fact layer, observed values, missing disclosures, and explicit anomalies of each indicator are identified. Second, at the governance issue layer, sixteen key governance questions are answered. Third, at the governance defect judgment layer, descriptions of governance defects such as imbalance of control, abnormal related-party transactions, failure of guarantee procedures, compliance breaches, incentive imbalances, weak auditing and internal control, and rising debt pressure are formed. Subsequently, at the fraud condition mapping layer, governance defects are mapped to one or more behavioral conditions among pressure, opportunity, rationalization, capability, arrogance, and collusion. Finally, at the risk transmission reasoning layer, an explanatory chain of "governance defect → financial fraud risk signal" is generated based on the dimension template, and core_analysis_conclusions are formed at the identification output layer. S6. Risk Assessment and Structured Output: Based on the corporate governance dimensions to which each indicator belongs, the severity of each indicator record determined in step S4, and the dimension summary results generated in step S5, the sixty-four atomic governance indicators are grouped into the corresponding corporate governance dimensions according to G1 to G7, where G1 to G7 are the numbers of the seven corporate governance dimensions; the average severity of all indicator records within each corporate governance dimension is taken, and this average is multiplied by 100 / 3 to obtain the dimension risk score; the average of the dimension risk scores of the seven corporate governance dimensions is taken to obtain the overall risk score; based on the overall risk score and a preset threshold, the financial fraud identification result is output, along with a structured result containing the indicator record set, dimension summary results, core analysis conclusions, overall risk score, final risk level, risk judgment basis, document summary, and integrity verification results. This step outputs a dimension summary for each dimension. The dimension summary includes all the following fields: dimension code, dimension name, expected number of indicators, actual number of indicators, missing or non-disclosed number, observed number of governance deficiencies, dimension risk score, and dimension core conclusion. The dimension risk score is calculated based on the severity field in step S4. After being aggregated to the corresponding dimension in step S5, the score is calculated by multiplying the average severity of all indicators in that dimension by 100 / 3 and rounding to two decimal places. S7. Integrity Verification: Based on step S6, perform integrity verification to check whether all indicator records cover all sixty-four atomic governance indicators, and verify whether the actual output indicator quantity of each corporate governance dimension is consistent with the preset quantity. After the verification is passed, output the final financial fraud identification result and structured result.

[0037] Furthermore, the publicly disclosed texts of listed companies in step S1 of the financial fraud identification method include: one or more of the following: the corporate governance section of the annual report, audit report and audit opinion, internal control or compliance self-inspection report, response to inquiries, announcement of major events, bond prospectus, and credit rating documents; the document type is used to identify the category of disclosure documents of the publicly disclosed texts, including company annual reports, announcements, audit reports, internal control reports, response to inquiries, bond documents, or credit rating documents; the analysis period is the fiscal year, semi-annual, quarterly, specific reporting period, or announcement date range covered by the publicly disclosed texts; the input range is used to identify the degree of coverage of the input texts, including three categories: full document (full_document), excerpt (excerpt), and text with unknown range (unknown).

[0038] Furthermore, the seven corporate governance dimensions in step S2 of the financial fraud identification method include: G1 equity structure and actual controller; G2 related party transactions; G3 external guarantees; G4 litigation, arbitration and major penalties; G5 incentive and constraint mechanisms; G6 audit, information disclosure and compliance; G7 bonds and debt repayment arrangements. These seven corporate governance dimensions sequentially cover the basis of control, channels for transferring benefits, risk transfer tools, signs of violations, behavioral motivation mechanisms, supervisory and checks and balances mechanisms, and external contractual pressure.

[0039] These seven dimensions are not isolated but form a main chain of governance risk along the lines of "basic control → channels for transferring benefits → tools for risk transfer → signals of violations → mechanisms for behavioral motivation → mechanisms for supervision and checks and balances → external contractual pressure." This main chain not only covers the core aspects of corporate governance influencing financial fraud but also provides sequential constraints for subsequent risk transmission reasoning.

[0040] Furthermore, the sixteen key governance issues in step S2 of the financial fraud identification method include: whether control is excessively concentrated or insufficiently checked; whether the control structure is unbalanced; whether the actual controller has special governance risks; whether the scale of related-party transactions is abnormal; whether the structure of related-party transactions is abnormal; whether there are abnormal related-party transactions; whether the scale of guarantees is too large; whether there are procedural or substantive risks in guarantees; whether there are major lawsuits or arbitrations; whether regulatory penalties are significant; whether the incentive intensity is too high; whether the constraint mechanism is weak; whether audit supervision is effective; and whether there are problems with internal control and disclosure compliance. Whether debt pressure is concentrated; whether financing and rating pressure is increasing; the sixty-four atomic governance indicators are configured as follows: G1 has 8 indicators, G2 has 11 indicators, G3 has 8 indicators, G4 has 9 indicators, G5 has 9 indicators, G6 has 11 indicators, and G7 has 8 indicators. Among them, G1 to G7 are the numbers of seven corporate governance dimensions: G1 corresponds to equity structure and actual controller; G2 corresponds to related-party transactions; G3 corresponds to external guarantees; G4 corresponds to litigation, arbitration and major penalties; G5 corresponds to incentive and restraint mechanisms; G6 corresponds to audit, information disclosure and compliance; and G7 corresponds to bonds and debt repayment arrangements.

[0041] The sixteen key governance issues include: whether control is excessively concentrated or insufficiently checked; whether the control structure is unbalanced; whether the actual controller has special governance risks; whether the scale of related-party transactions is abnormal; whether the structure of related-party transactions is abnormal; whether there are abnormal related-party transactions; whether the scale of guarantees is too large; whether there are procedural or substantive risks in guarantees; whether there are major lawsuits or arbitrations; whether regulatory penalties are significant; whether the incentive intensity is too high; whether the constraint mechanism is weak; whether audit supervision is effective; whether there are problems with internal control and disclosure compliance; whether debt pressure is concentrated; and whether financing and rating pressures are increasing.

[0042] Furthermore, the sixty-four atomic governance indicators in step S2 of the financial fraud identification method are as follows: 8 G1 items: G1-Q1-I1 Shareholding Ratio of the Largest Shareholder; G1-Q1-I2 Concentration of the Top Ten Shareholders; G1-Q1-I3 Herfindahl-Hirschman Index (HHI); G1-Q2-I1 Control-Cash Flow Dispersion; G1-Q2-I2 Dual-Class Shares / Special Voting Rights / Voting Rights Arrangements; G1-Q3-I1 Type of Actual Controller; G1-Q3-I2 Change of Actual Controller / Instable Control / No Actual Controller; G1-Q3-I3 Control Risks such as Share Pledge, Freezing, and Judicial Auction. 11 G2 items: G2-Q1-I1 Related Party Transaction Amount; G2-Q1-I2 Related Party Transaction Amount / Operating Revenue; G2-Q1-I3 Related Party Transaction Amount / Total Assets; G2-Q2-I1 Related Party Sales Ratio; G2-Q2-I2 Related Party Procurement Ratio; G2-Q2-I3 Related Party Fund Transactions Ratio; G2-Q2-I4 Other Receivables / Related Party Occupation; G2-Q2-I5 Related Party Guarantee Ratio; G2-Q3-I1 Abnormal Related Party Transactions; G2-Q3-I2 Unfair Pricing / Circular Transactions / Year-End Sudden Transactions; G2-Q3-I3 Abnormal Approval and Disclosure Procedures; 8 G3 items: G3-Q1-I1 Guarantee Balance; G3-Q1-I2 Guarantee Balance / Net Assets; G3-Q1-I3 Guarantee Balance / Total Assets; G3-Q1-I4 Number of Guarantees; Whether G3-Q2-I1 Has Any Irregular Guarantees; Whether G3-Q2-I2 Has Any Implicit Guarantees; Whether G3-Q2-I3 Lacks Counter-Guarantees; Whether G3-Q2-I4 Has Any Guarantees Involved in Compensation or Litigation; 9 G4 items: G4-Q1-I1: Are there any major lawsuits / arbitrations? G4-Q1-I2: Number of major lawsuits / arbitrations; G4-Q1-I3: Amount involved; G4-Q1-I4: Amount involved / Net assets; G4-Q1-I5: Amount involved / Total assets; G4-Q2-I1: Number of regulatory penalties; G4-Q2-I2: Severity of penalties; G4-Q2-I3: Is there any case under investigation? G4-Q2-I4: Are there any major compliance incidents? 9 G5 items: G5-Q1-I1 Sensitivity to Stock Option Incentives; G5-Q1-I2 Total Executive Compensation; G5-Q1-I3 Bonus Ratio; G5-Q1-I4 Existence of Equity Incentive Plan; G5-Q2-I1 Board Independence; G5-Q2-I2 Audit Committee Characteristics; G5-Q2-I3 Board Meeting Frequency; G5-Q2-I4 Supervisory Board Oversight; G5-Q2-I5 Internal Oversight. 11 items in G6: G6-Q1-I1 Auditor size; G6-Q1-I2 Audit opinion type; G6-Q1-I3 Exceptions in emphasis of matter / key audit matters; G6-Q1-I4 Auditor change; G6-Q1-I5 Audit fees; G6-Q1-I6 Exceptional audit fees; G6-Q2-I1 Whether there are material weaknesses in internal control; G6-Q2-I2 Internal control audit opinion type; G6-Q2-I3 Information disclosure evaluation; G6-Q2-I4 Frequency of regulatory inquiries; G6-Q2-I5 Whether there are supplementary disclosures / corrected disclosures / delayed disclosures? 8 G7 items: G7-Q1-I1 Debt-to-equity ratio; G7-Q1-I2 Short-term debt ratio; G7-Q1-I3 Interest coverage ratio; G7-Q1-I4 Bond maturity concentration; G7-Q2-I1 Credit rating; G7-Q2-I2 Rating outlook change; G7-Q2-I3 Bond spread / yield; G7-Q2-I4 Whether there is a put option / rollover / default / cross-default. In the atomic governance indicator names, the slash " / " indicates a ratio relationship or parallel items according to the indicator semantics: in ratio-type atomic governance indicators, it means "divided by", and in parallel item-type atomic governance indicators, it means parallel selection or parallel listing, without changing the number of individual atomic governance indicators.

[0043] Furthermore, the corporate governance thinking chain framework in step S2 of the financial fraud identification method consists of the following layers in sequence: governance facts layer, governance problem layer, governance defect judgment layer, fraud condition mapping layer, risk transmission reasoning layer, and identification output layer. The fixed mapping relationship between corporate governance dimensions and the hexagonal elements of fraud is as follows: equity structure and actual controller correspond to opportunity, ability, collusion, and arrogance; related-party transactions correspond to opportunity, collusion, and rationalization; external guarantees correspond to pressure, opportunity, and collusion; litigation, arbitration, and major penalties correspond to pressure, opportunity, and rationalization; incentive and constraint mechanisms correspond to pressure, rationalization, arrogance, and ability; auditing, information disclosure, and compliance correspond to opportunity, collusion, and ability; bonds and debt repayment arrangements correspond to pressure, rationalization, and ability. Among these, pressure, opportunity, rationalization, ability, arrogance, and collusion are the six types of behavioral conditions in the hexagonal theory of fraud. G1 to G7 are the numbers of the corporate governance dimensions. Each corporate governance dimension corresponds to one or more hexagonal elements of fraud through a fixed mapping relationship, which are used to explain the behavioral conditions that transform governance defects into financial fraud risk signals in indicator recording and risk transmission reasoning.

[0044] The aforementioned fixed mapping relationship is determined based on the role of each governance dimension in the formation of financial fraud: equity structure and actual controllers determine the strength of insider control and checks and balances, thus mainly corresponding to opportunity, capability, collusion, and arrogance; related-party transactions are characterized by non-marketization, concealment, and the ability to be manipulated by related parties, thus corresponding to opportunity, collusion, and rationalization; external guarantees may create contingent liabilities and risk transfer pressure, while exposing gaps in approval and disclosure opportunities, thus corresponding to pressure, opportunity, and collusion; litigation, arbitration, and major penalties reflect compliance failures, reputational damage, and increased financing constraints, thus corresponding to pressure, opportunity, and rationalization; incentive and constraint mechanisms affect management's short-term performance motivation, accountability constraints, and ability to manipulate financial statements, thus corresponding to pressure, rationalization, arrogance, and capability; auditing, information disclosure, and compliance constitute external supervision and internal control defenses, and their weakening can create opportunities to hide fraud and may lead to collusion to distort disclosures, thus corresponding to opportunity, collusion, and capability; bonds and debt repayment arrangements reflect debt contracts, rating maintenance, and refinancing pressures, and require management to have the ability to allocate funds and package financial statements, thus corresponding to pressure, rationalization, and capability.

[0045] Furthermore, the indicator status in step S4 of the financial fraud identification method is selected from the following seven categories: observed_value, explicit_none, explicit_not_applicable, partial_disclosure_but_not_computable, section_present_but_blank, not_found_in_input, and missing_required_disclosure (if the entire document should be disclosed). Among these, when the input scope is the entire document... When a full document of an indicator is not found, it is identified as a missing document that should be disclosed (missing_required_disclosure); when the input range is an excerpt or an unknown range of text and no relevant content is found, it is identified as not found in the current input (not_found_in_input); when only a title, table header, or chapter name exists but no substantive content exists, it is identified as a present section but blank content (section_present_but_blank); when relevant content is found but the indicator calculation cannot be completed, it is identified as partially disclosed but not computable (partial_disclosure_but_not_computable).

[0046] Specifically, when a clear indicator value or clear fact is found in the original text, it is identified as a clear observed value; when the original text clearly states "none," "did not occur," "does not exist," "does not involve," "none in this period," or "zero," it is identified as an explicit none; when the original text clearly states "not applicable," it is identified as explicit not applicable; when relevant content exists but is insufficient to complete the full indicator calculation, it is identified as partially disclosed but not computable; when only a relevant title, table header, or chapter name is found without substantive content, it is identified as a chapter present but blank; when the input range is an excerpt or an unknown text and no relevant content is found, it is identified as not found in the current input; when the input range is the full document and the full document is still not found, it is identified as a missing required disclosure. Regarding anomaly labeling, if the indicator's presence status is a clearly observed value and the content is normal, the abnormal feature flag is set to 0, and the abnormal feature type is set to none. If the indicator's presence status is a clearly observed value but the text displays governance defects, program abnormalities, or high-risk facts, the abnormal feature flag is set to 1, and the abnormal feature type is set to observed governance defects. If the indicator's presence status is partial disclosure but not computable, or the chapter... If a section exists but is blank (section_present_but_blank), is not found in the current input (not_found_in_input), or is missing from a complete document that should be disclosed (missing_required_disclosure), then the abnormal feature flag (abnormal_feature_flag) is set to 1, and the abnormal feature type (abnormal_feature_type) is set to missing_disclosure_or_not_mentioned. If the indicator presence status (indicator_presence_status) is explicit_none or explicit_not_applicable, then the abnormal feature flag (abnormal_feature_flag) is set to 0, and the abnormal feature type (abnormal_feature_type) is set to disclosed_none or_not_applicable. Regarding indicator value acquisition, if the original text directly provides the indicator value, the indicator value acquisition method (value_obtaining_method) is marked as direct extraction (direct_extract); if the indicator value can be calculated from the table or numerical value in the current text, the indicator value acquisition method (value_obtaining_method) is marked as computed from text (computed_from_text); if the current text information is insufficient, the indicator value (indicator_value) is set to empty, and the indicator value acquisition method (value_obtaining_method) is marked as not computable in current input (not_computable_in_current_input).

[0047] Furthermore, the abnormal feature flags and abnormal feature types in step S4 of the financial fraud identification method are determined according to the following rules: When the indicator's status is a clearly observed value and no clear governance defect is observed, the abnormal feature flag is 0, and the abnormal feature type is none; when the indicator's status is a clearly observed value and a governance defect, procedural anomaly, or high-risk fact is observed, the abnormal feature flag is 1, and the abnormal feature type is observed governance defect; when the indicator's status is partially disclosed but not computable, section exists but content is blank, or current input is not found... When a complete document should disclose something but is missing (missing_required_disclosure), the anomaly flag is 1, and the anomaly type is missing_disclosure_or_not_mentioned; when the indicator status is explicitly none (explicit_none) or explicitly not applicable (explicit_not_applicable), the anomaly flag is 0, and the anomaly type is disclosed_none_or_not_applicable; among them, the indicator value acquisition methods include three categories: direct extraction (direct_extract), computed from text (computed_from_text), and not_computable_in_current_input (not_computable_in_current_input).

[0048] Furthermore, in step S6 of the financial fraud identification method, the dimension risk scores for each corporate governance dimension are calculated based on the severity determined in step S4 and the dimension aggregation results in step S5, according to the following formula: Dimension risk score (dimension_score) = round(average severity of all indicators within the dimension × 100 / 3, 2); The overall risk score is calculated according to the following formula: Overall risk score (overall_risk_score) = round(average of the dimension risk scores of the seven corporate governance dimensions, 2); When the overall risk score (overall_risk_score) is greater than or equal to 45, the financial fraud identification result is output as "moderately high"; when the overall risk score (overall_risk_score) is less than or equal to 45, the financial fraud identification result is output as "moderately high"; when the overall risk score (overall_risk_score) is less than or equal to 45, the financial fraud identification result is output as "moderately high"; when the overall risk score (overall_risk_score) is less than or equal to 45, the financial fraud identification result is output as "moderately high". When sk_score is less than 45, the output financial fraud identification result is "moderate to low"; the structured result includes all the following fields: indicator record set, dimension summary, core analysis conclusions, overall risk score, final risk level, risk judgment basis, document summary, and completeness check result.

[0049] Furthermore, the integrity verification in step S7 of the financial fraud identification method includes all of the following verification items: verifying whether the indicator record set (indicator_records) covers all sixty-four atomic governance indicators and whether the indicator codes are consistent with the fixed indicator dictionary; verifying whether the dimension summary results (dimension_summary) cover all seven corporate governance dimensions; verifying the output of 8 items for the equity structure and actual controller dimension, 11 items for the related party transaction dimension, 8 items for the external guarantee dimension, 9 items for the litigation, arbitration and major penalty dimension, 9 items for the incentive and constraint mechanism dimension, 11 items for the audit, information disclosure and compliance dimension, and 8 items for the bond and debt repayment arrangement dimension; verifying whether each indicator record contains the indicator code (indicator_code), indicator name (indicator_name), the dimension (dimension), and key governance issues (governance_queries). The system outputs the following parameters: indicator presence status, abnormal feature flag, abnormal feature type, indicator value, value obtaining method, supporting source evidence, severity, confidence, and hexagonal elements. The final risk level is verified to be either "low to moderate" or "high to moderate". If any verification item fails, a verification failure flag and a list of missing indicator codes or fields are output. Upon successful verification, the final financial fraud identification result and structured results are output.

[0050] Furthermore, regarding the equity structure and actual controller, excessive concentration of control, insufficient checks and balances, deviation between control and cash flow rights, or unstable control can weaken the checks and balances of the board of directors, supervisory board, independent directors, and minority shareholders, and enhance the ability of insiders to control major transactions, fund allocation, and information disclosure. Regarding related-party transactions, unfair related-party purchases and sales, fund lending, abnormal related-party transactions, and abnormal approval and disclosure can easily transform related-party transactions into a means of transferring benefits, adjusting profits, and transferring assets. Regarding external guarantees, illegal guarantees, implicit guarantees, excessive guarantees, or litigation involving compensation can create potential contingent liabilities and liquidity risks, and amplify the motivation to expose hidden risks.

[0051] Furthermore, regarding litigation, arbitration, and major penalties, significant litigation compensation, administrative penalties, investigations, and major compliance events often indicate a weak compliance culture, pressure on profits and cash flow, and increased refinancing constraints. Regarding incentive and constraint mechanisms, aggressive performance commitments, short-term incentive arrangements, and weak constraint mechanisms will translate operational targets into pressure on financial statements and reinforce the tendency to rationalize "phased smoothing" and "temporary packaging." Regarding auditing, information disclosure, and compliance, weak audit quality, significant internal control deficiencies, frequent inquiries, and supplementary or corrective disclosures will amplify information asymmetry and prolong the time for fraud to be concealed. Regarding bonds and debt repayment arrangements, concentrated bond maturities, rating downgrades, put options, or default risks will increase the incentive to maintain financing capacity and market confidence through financial statement manipulation. Example 1

[0052] This embodiment uses the 2018 publicly disclosed texts of Kangmei Pharmaceutical Co., Ltd. as input. Document types include annual reports, audit reports, internal control reports, responses to inquiry letters, and administrative penalty decisions. The analysis period is 2018, and the input scope is set to full documents. The method first receives the company name, document type, analysis period, input scope, and publicly disclosed texts. It then performs text parsing, table recognition, and plain text cleaning on the PDF versions of the annual reports, audit reports, internal control reports, responses to inquiry letters, and penalty decisions, forming cleaned plain text that can be processed by the structured prompt word construction module as input text.

[0053] Subsequently, the method constructs seven corporate governance dimensions, sixteen key governance issues, sixty-four atomic governance indicators, a fixed hexagonal mapping relationship for fraud, and a risk transmission template, and generates indicator record sets (indicator_records) item by item according to the fixed indicator dictionary. Taking this example, G1-Q3-I3 can extract control risk signals of a high proportion of equity pledge by the controlling shareholder; G2-Q2-I4 and G2-Q3-I1 can extract related-party transaction risk signals such as related-party occupation and non-operating fund occupation in other receivables; G4-Q2-I3 can extract compliance risk signals of being investigated by regulatory agencies; G6-Q2-I1 and G6-Q2-I2 can extract material weaknesses in internal control and adverse opinions in internal control audits; G7-Q1-I2 and G7-Q1-I3 can extract debt risk signals such as short-term debt pressure and insufficient interest coverage.

[0054] During the indicator status identification phase, if the original text directly provides a numerical value or fact, it is identified as an explicit observation (observed_value); if the original text explicitly states that there are no external guarantees or defaults, it is identified as explicit none (explicit_none); if only partial information is available but a complete proportion or amount cannot be formed, it is identified as partial disclosure but not computable (partial_disclosure_but_not_computable). For indicators such as related-party fund misappropriation, abnormal approval and disclosure procedures, significant deficiencies in internal control, and investigations in the Kangmei Pharmaceutical example, the abnormality flag is set to 1, and the severity is set to level 2 or 3 based on the risk level; for indicators such as external guarantees and defaults that are explicitly disclosed as none, the abnormality flag is set to 0.

[0055] In the corporate governance thought chain reasoning stage, the method follows the sequence of "governance fact extraction → key governance issue determination → governance defect identification → fraud hexagonal element mapping → risk transmission reasoning → financial fraud identification conclusion generation". It summarizes facts such as equity pledge and control risk, related party fund occupation, internal control failure, false information disclosure, regulatory investigation and debt repayment pressure into governance defects such as insufficient control checks and balances, abnormal related party transactions and fund occupation, ineffective supervision and constraints, weakened audit and internal control defenses, and rising external debt pressure. It further maps these defects into fraud hexagonal elements such as opportunity, pressure, collusion, capability, arrogance and rationalization.

[0056] In the risk assessment and structured output stage, the method calculates the dimension risk score by multiplying the average severity of all indicators within a dimension by 100 / 3 and retaining two decimal places. The overall risk score is then obtained by averaging the scores of the seven dimensions. For example, dimensions such as related-party transactions, litigation penalties, audit information disclosure and compliance, and bonds and debt repayment arrangements have relatively high scores, reaching the upper-middle-high threshold for overall risk, resulting in a final risk level output of "middle-high". The structured results simultaneously include the indicator record set (indicator_records), dimension summary results (dimension_summary), core analysis conclusions (core_analysis_conclusions), overall risk score (overall_risk_score), final risk level (final_risk_level), risk judgment basis (risk_judgement_basis), document summary (document_summary), and completeness check result (completeness_check_result).

[0057] Finally, the method performs an integrity check: it verifies whether the indicator record set (indicator_records) covers all sixty-four atomic governance indicators, whether the actual output quantities of G1 to G7 are 8, 11, 8, 9, 9, 11, and 8 respectively, and whether each indicator record contains the indicator code, indicator name, dimension, key governance issue, indicator occurrence status, anomaly characteristic flag, anomaly characteristic type, indicator value, indicator value acquisition method, supporting original text, severity, confidence level, and fraud hexagonal elements. After all checks pass, the final identification result is output. Example 2

[0058] This embodiment uses excerpts of Kangmei Pharmaceutical Co., Ltd.'s 2018 inquiry responses or announcement summaries as input, with the input range set to excerpted text (excerpt). Since the input text is not a complete document, the method prioritizes determining that relevant indicators not found during the status identification stage are not found in the current input (not_found_in_input) or missing from the incomplete document that should be disclosed (missing_required_disclosure). For cases where there is only a partial description but it is insufficient to form a complete indicator value, it is determined to be partially disclosed but not computable (partial_disclosure_but_not_computable), and this is retained as an abnormal feature in the indicator record.

[0059] In this embodiment, the method still outputs a complete set of indicator records (indicator_records) according to the sixty-four atomic governance indicators, and combines the observed governance anomalies, missing disclosures, and partial disclosures in the governance thought chain reasoning stage to form a conclusion on financial fraud identification. By distinguishing between two types of input ranges, namely full documents and excerpts, this invention can maintain a unified framework, unified rules, and unified output structure in both full text and excerpt scenarios, thereby improving the consistency and verifiability of the identification results.

[0060] This invention takes seven corporate governance dimensions, sixteen key governance issues and sixty-four atomic governance indicators as its core, and establishes a fixed thinking chain of "governance facts - governance defects - fraud conditions - risk transmission - financial fraud identification conclusions", so that corporate governance texts can directly serve the identification of financial fraud, rather than just extracting local indicators or general risk warnings. This invention includes situations such as undisclosed, unmentioned, only title without content, and insufficient information to calculate within the scope of abnormal feature identification, so that the governance disclosure gap itself can be stably transformed into financial fraud identification features, thereby improving governance transparency and the ability to utilize abnormal information. This invention improves the interpretability of financial fraud identification conclusions by establishing a fixed mapping relationship between corporate governance dimensions and the hexagonal elements of fraud, as well as risk transmission templates corresponding to the seven dimensions. This enables the identification results to not only indicate the "level of risk" but also to explain "how governance deficiencies are transformed into risk signals along the thought chain." This invention improves the standardization, comparability, and verifiability of analysis results from different companies, different text types, and different periods by using an integrity verification mechanism, unified status classification rules, and unified severity and overall risk assessment rules.

[0061] The technical solutions disclosed in the embodiments of the present invention have been described in detail above. Specific embodiments have been used to illustrate the principles and implementation methods of the embodiments of the present invention. The description of the above embodiments is only for helping to understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying financial fraud by integrating corporate governance thinking, characterized in that, The financial fraud identification method includes the following steps: S1. Receive the name of the listed company to be analyzed and the publicly disclosed text of the listed company, document type, analysis period and input scope, perform layout parsing, paragraph segmentation, table recognition and plain text cleaning on the publicly disclosed text of the listed company, and obtain the cleaned plain text. S2. Construct a corporate governance thinking chain framework, which includes seven corporate governance dimensions, sixteen key governance issues, sixty-four atomic governance indicators, a fixed mapping relationship between corporate governance dimensions and the six-corner element of fraud, and a risk transmission template for each corporate governance dimension corresponding to governance defects to financial fraud risk signals. S3. Take the cleaned plain text from step S1 as the input text, extract all of the input text according to the sixty-four atomic governance indicators, generate indicator records for each atomic governance indicator, and retain indicator records that are not disclosed, not mentioned, have only a title, or have insufficient information to be calculated. S4. Identify the occurrence status of each indicator record, and determine the abnormal feature flag, abnormal feature type, indicator value, indicator value acquisition method, severity, confidence level, and corresponding fraud hexagonal elements for each indicator record based on the occurrence status. S5. Following the corporate governance thinking chain sequence of "governance fact extraction → key governance issue determination → governance defect identification → fraud hexagonal element mapping → risk transmission reasoning → financial fraud identification conclusion generation", reason about the indicator records to generate dimensional summary results and core analytical conclusions from governance defects to financial fraud risk signals. S6. Based on the corporate governance dimension to which each indicator belongs, the severity of each indicator recorded in step S4, and the dimension summary results generated in step S5, the sixty-four atomic governance indicators are aggregated to the corresponding corporate governance dimensions. The severity of all indicators recorded within each corporate governance dimension is averaged, and this average is multiplied by 100 / 3 to obtain the dimension risk score; the overall risk score is obtained by averaging the dimension risk scores of the seven corporate governance dimensions. Based on the overall risk score and preset threshold, the system outputs financial fraud identification results, and outputs structured results including indicator record set, dimension summary results, core analysis conclusions, overall risk score, final risk level, risk judgment basis, document summary and integrity verification results. S7. Based on step S6, perform an integrity check to verify whether all indicator records cover all sixty-four atomic governance indicators, and verify whether the actual output indicator quantity of each corporate governance dimension is consistent with the preset quantity. After the check passes, output the final financial fraud identification result and structured result.

2. The financial fraud identification method incorporating corporate governance thinking as described in claim 1, characterized in that, The publicly disclosed texts of listed companies in step S1 of the financial fraud identification method are input from external data sources, including: one or more of the following in the PDF version of the annual report: corporate governance section, notes to the annual report, audit report and audit opinion, internal control or compliance self-inspection report, exchange inquiry letter and inquiry response, announcement of major events, bond prospectus and credit rating documents; The document type in step S1 of the financial fraud identification method is used to identify the disclosure document category of the publicly disclosed text of the listed company; the analysis period is used to mark the fiscal year, semi-annual, quarterly, specific reporting period or announcement date range covered by the publicly disclosed text of the listed company; the input range is used to identify the coverage of the input text in step S2, including three categories: full_document, excerpt, and unknown.

3. The financial fraud identification method incorporating corporate governance thinking as described in claim 1, characterized in that, The seven corporate governance dimensions in step S2 of the financial fraud identification method include: equity structure and actual controller; related-party transactions; external guarantees; litigation, arbitration and major penalties; incentive and constraint mechanisms; audit, information disclosure and compliance; bonds and debt repayment arrangements. The seven corporate governance dimensions in step S2 of the financial fraud identification method cover, in turn, the basis of control, channels for transferring benefits, tools for transferring risks, signals of violations, mechanisms of behavioral motivation, mechanisms of supervision and checks and balances, and external contractual pressure.

4. The financial fraud identification method incorporating corporate governance thinking as described in claim 1, characterized in that, The sixteen key governance issues in step S2 of the financial fraud identification method include: whether control is excessively concentrated or insufficiently checked; whether the control structure is unbalanced; whether the actual controller has special governance risks; whether the scale of related-party transactions is abnormal; whether the structure of related-party transactions is abnormal; whether there are abnormal related-party transactions; whether the scale of guarantees is too large; whether there are procedural or substantive risks in guarantees; whether there are major lawsuits or arbitrations; whether regulatory penalties are significant; whether the incentive intensity is too high; whether the constraint mechanism is weak; whether audit supervision is effective; whether there are problems with internal control and disclosure compliance; whether debt pressure is concentrated; and whether financing and rating pressures are increasing. The sixty-four atomic governance indicators in step S2 of the financial fraud identification method are configured as follows: G1 = 8 indicators, G2 = 11 indicators, G3 = 8 indicators, G4 = 9 indicators, G5 = 9 indicators, G6 = 11 indicators, and G7 = 8 indicators. Among them, G1 to G7 are the numbers of seven corporate governance dimensions: G1 corresponds to equity structure and actual controller; G2 corresponds to related-party transactions; G3 corresponds to external guarantees; G4 corresponds to litigation, arbitration and major penalties; G5 corresponds to incentive and restraint mechanisms; G6 corresponds to audit, information disclosure and compliance; and G7 corresponds to bonds and debt repayment arrangements.

5. The financial fraud identification method integrating corporate governance thinking chain according to claim 4, characterized in that, The sixty-four atomic governance indicators in step S2 of the financial fraud identification method are as follows: 8 G1 items: G1-Q1-I1 Shareholding Ratio of the Largest Shareholder; G1-Q1-I2 Concentration of the Top Ten Shareholders; G1-Q1-I3 Herfindahl-Hirschman Index (HHI); G1-Q2-I1 Control-Cash Flow Dispersion; G1-Q2-I2 Dual-Class Shares / Special Voting Rights / Voting Rights Arrangements; G1-Q3-I1 Type of Actual Controller; G1-Q3-I2 Change of Actual Controller / Instable Control / No Actual Controller; G1-Q3-I3 Control Risks such as Share Pledge, Freezing, and Judicial Auction. 11 G2 items: G2-Q1-I1 Related Party Transaction Amount; G2-Q1-I2 Related Party Transaction Amount / Operating Revenue; G2-Q1-I3 Related Party Transaction Amount / Total Assets; G2-Q2-I1 Related Party Sales Ratio; G2-Q2-I2 Related Party Procurement Ratio; G2-Q2-I3 Related Party Fund Transactions Ratio; G2-Q2-I4 Other Receivables / Related Party Occupation; G2-Q2-I5 Related Party Guarantee Ratio; G2-Q3-I1 Abnormal Related Party Transactions; G2-Q3-I2 Unfair Pricing / Circular Transactions / Year-End Sudden Transactions; G2-Q3-I3 Abnormal Approval and Disclosure Procedures; 8 G3 items: G3-Q1-I1 Guarantee Balance; G3-Q1-I2 Guarantee Balance / Net Assets; G3-Q1-I3 Guarantee Balance / Total Assets; G3-Q1-I4 Number of Guarantees; Whether G3-Q2-I1 Has Any Irregular Guarantees; Whether G3-Q2-I2 Has Any Implicit Guarantees; Whether G3-Q2-I3 Lacks Counter-Guarantees; Whether G3-Q2-I4 Has Any Guarantees Involved in Compensation or Litigation; 9 G4 items: G4-Q1-I1: Are there any major lawsuits / arbitrations? G4-Q1-I2: Number of major lawsuits / arbitrations; G4-Q1-I3: Amount involved; G4-Q1-I4: Amount involved / Net assets; G4-Q1-I5: Amount involved / Total assets; G4-Q2-I1: Number of regulatory penalties; G4-Q2-I2: Severity of penalties; G4-Q2-I3: Is there any case under investigation? G4-Q2-I4: Are there any major compliance incidents? 9 G5 items: G5-Q1-I1 Sensitivity to Stock Option Incentives; G5-Q1-I2 Total Executive Compensation; G5-Q1-I3 Bonus Ratio; G5-Q1-I4 Existence of Equity Incentive Plan; G5-Q2-I1 Board Independence; G5-Q2-I2 Audit Committee Characteristics; G5-Q2-I3 Board Meeting Frequency; G5-Q2-I4 Supervisory Board Oversight; G5-Q2-I5 Internal Oversight. 11 items in G6: G6-Q1-I1 Auditor size; G6-Q1-I2 Audit opinion type; G6-Q1-I3 Exceptions in emphasis of matter / key audit matters; G6-Q1-I4 Auditor change; G6-Q1-I5 Audit fees; G6-Q1-I6 Exceptional audit fees; G6-Q2-I1 Whether there are material weaknesses in internal control; G6-Q2-I2 Internal control audit opinion type; G6-Q2-I3 Information disclosure evaluation; G6-Q2-I4 Frequency of regulatory inquiries; G6-Q2-I5 Whether there are supplementary disclosures / corrected disclosures / delayed disclosures? 8 G7 items: G7-Q1-I1 Debt-to-equity ratio; G7-Q1-I2 Short-term debt ratio; G7-Q1-I3 Interest coverage ratio; G7-Q1-I4 Bond maturity concentration; G7-Q2-I1 Credit rating; G7-Q2-I2 rating outlook changes; G7-Q2-I3 bond spreads / yields; G7-Q2-I4 potential for put options / rollovers / defaults / cross-defaults; In the atomic governance indicator names, the slash " / " indicates a ratio relationship or parallel items according to the indicator semantics: in ratio-type atomic governance indicators, it means "divided by", and in parallel item-type atomic governance indicators, it means parallel selection or parallel listing, without changing the number of individual atomic governance indicators.

6. The financial fraud identification method incorporating corporate governance thinking chain according to claim 1, characterized in that, The corporate governance thinking chain framework in step S2 of the financial fraud identification method consists of the following layers in sequence: governance facts layer, governance problem layer, governance defect judgment layer, fraud condition mapping layer, risk transmission reasoning layer, and identification output layer. The fixed mapping relationship between the corporate governance dimension and the fraud hexagonal elements in step S2 of the financial fraud identification method is as follows: equity structure and actual controller correspond to opportunity, ability, collusion, and arrogance; related-party transactions correspond to opportunity, collusion, and rationalization; external guarantees correspond to pressure, opportunity, and collusion; litigation, arbitration, and major penalties correspond to pressure, opportunity, and rationalization; incentive and constraint mechanisms correspond to pressure, rationalization, arrogance, and ability; auditing, information disclosure, and compliance correspond to opportunity, collusion, and ability; bonds and debt repayment arrangements correspond to pressure, rationalization, and ability. Among these, pressure, opportunity, rationalization, ability, arrogance, and collusion are the six types of behavioral conditions in the fraud hexagonal theory, and each corporate governance dimension corresponds to one or more fraud hexagonal elements through a fixed mapping relationship.

7. The financial fraud identification method incorporating corporate governance thinking as described in claim 1, characterized in that, The indicator status in step S4 of the financial fraud identification method is selected from the following seven categories: observed_value, explicit_none, partial_disclosure_but_not_computable, explicit_not_applicable, section_present_but_blank, not_found_in_input, and missing_required_disclosure (which should be disclosed in the complete document). Among these, when the input range is the complete document... When the full text of a certain indicator is not found, it is identified as a complete document that should be disclosed but is missing (missing_required_disclosure); when the input range is an excerpt of text (excerpt) or an unknown range of text (unknown) and no relevant content is found, it is identified as not found (not_found_in_input); when relevant content is found but the indicator calculation cannot be completed, it is identified as partially disclosed but not computable (partial_disclosure_but_not_computable); when only a title, table header, or chapter name exists but no substantive content exists, it is identified as a chapter that exists but the content is blank (section_present_but_blank).

8. The financial fraud identification method incorporating corporate governance thinking as described in claim 1, characterized in that, The abnormal feature flags and abnormal feature types in step S4 of the financial fraud identification method are determined according to the following rules: When the indicator status is a clearly observed value (observed_value) and no clear governance defect is observed, the abnormal feature flag is 0, and the abnormal feature type is none; when the indicator status is a clearly observed value (observed_value) and a governance defect, procedural anomaly, or high-risk fact is observed, the abnormal feature flag is 1, and the abnormal feature type is observed_governance_defect; when a section exists but its content is blank (section_present_but_blank), the indicator status is partially disclosed but not computable (partial_disclosure_but_not_computable), or the current input is not found (not_found_in_input), the abnormal feature flag is 0. When a document or complete document should disclose information but is missing (missing_required_disclosure), the anomaly flag is 1, and the anomaly type is missing_disclosure_or_not_mentioned. When the indicator status is explicitly none (explicit_none) or explicitly not applicable (explicit_not_applicable), the anomaly flag is 0, and the anomaly type is disclosed none (disclosed_none_or_not_applicable). The indicator value is obtained through three methods: direct_extract, computed_from_text, and not_computable_in_current_input.

9. The financial fraud identification method incorporating corporate governance thinking chain according to claim 1, characterized in that, The dimension risk score of each corporate governance dimension in step S6 of the financial fraud identification method is calculated according to the following formula: dimension_score = round(average severity of all indicators within the dimension × 100 / 3, 2). The overall risk score in step S6 of the financial fraud identification method is calculated according to the following formula: overall_risk_score = round(average of risk scores of the seven corporate governance dimensions, 2); when overall_risk_score is greater than or equal to 45, the financial fraud identification result is output as "moderately high"; when overall_risk_score is less than 45, the financial fraud identification result is output as "moderately low". The structured result in step S6 of the financial fraud identification method includes all of the following fields: indicator_records; dimension_summary; core_analysis_conclusions; overall_risk_score; final_risk_level; risk_judgement_basis; document_summary; and completeness_check_result.

10. The financial fraud identification method integrating corporate governance thinking chain according to claim 1, characterized in that, The integrity verification in step S7 of the financial fraud identification method includes all of the following verification items: verifying whether the indicator record set `indicator_records` covers all sixty-four atomic governance indicators and whether the indicator codes are consistent with the fixed indicator dictionary; verifying whether the dimension summary result `dimension_summary` covers all seven corporate governance dimensions; verifying the output of 8 items for the equity structure and actual controller dimension, 11 items for the related party transaction dimension, 8 items for the external guarantee dimension, 9 items for the litigation, arbitration and major penalty dimension, 9 items for the incentive and constraint mechanism dimension, 11 items for the audit, information disclosure and compliance dimension, and 8 items for the bond and debt repayment arrangement dimension; and verifying whether each indicator record contains the indicator code `indicator_code`. The indicator name, dimension, governance question, indicator presence status, abnormal feature flag, abnormal feature type, indicator value, supporting source evidence, severity, confidence, method of obtaining the indicator value, and hexagon elements. The final risk level (final_risk_level) is verified to be either "moderately low" or "moderately high". If any verification item fails, a verification failure flag and a list of missing indicator codes or missing fields are output. If the verification passes, the final financial fraud identification result and structured result are output.

Images