Enterprise value symbiosis report intelligent agent evaluation method based on artificial intelligence large model

Abstract

This invention provides a method for evaluating enterprise value symbiosis reports based on an artificial intelligence big data model. The evaluation target is general enterprises with a unified social credit code starting with 91 and possessing complete operational data. The method combines internal historical transaction data, stakeholder supplementary data, and publicly available external big data. Through the deep integration of large language models, knowledge graphs, multi-agent collaboration, and machine learning algorithms, it effectively evaluates the quality of the symbiotic relationship and the total ecological value between the enterprise and its core stakeholders. This provides comprehensive information support for enterprises to optimize customer cooperation, supply chain management, employee motivation, environmental governance, and social responsibility fulfillment. Simultaneously, the system can automatically generate multiple versions of value symbiosis reports, output risk items and collaborative suggestions, effectively supporting enterprise strategic decision-making and sustainable development management. It also supports scenario simulation and natural language interaction, significantly improving the intelligence and automation level of enterprise value assessment.

Description

Technical Field

[0001] The invention relates to the fields of artificial intelligence and data mining technology, specifically to a method for evaluating intelligent agents in enterprise value symbiosis reporting based on a large artificial intelligence model. Background Technology

[0002] Currently, enterprises face the following prominent issues when conducting stakeholder management and sustainability assessments: First, data sources are scattered, and it is difficult to effectively integrate internal financial data, business process records, and external public information, leading to biased assessments; second, assessment indicators are mostly based on static financial ratios, lacking dynamic quantification of diverse symbiotic relationships such as customers, suppliers, employees, and the environment; third, traditional methods rely on human experience, which is inefficient, highly subjective, and difficult to handle massive amounts of unstructured text (such as ESG reports and litigation documents); and fourth, there is a lack of a unified framework for calculating total ecosystem value, which cannot support strategic decision-making and scenario simulation.

[0003] To deeply explore the value of enterprises' multi-source data assets and fully leverage the important role of artificial intelligence models in empowering the management and evaluation of enterprise interest symbiosis, this invention designs an intelligent agent system for enterprise interest symbiosis reporting that integrates large language models, multi-agent collaborative mechanisms, and expert knowledge and experience. Specifically, this invention starts with historical transaction data between the enterprise and its core stakeholders (including customers, suppliers, shareholders, employees, creditors, society, and the environment), internally accumulated supplementary data (such as litigation, non-compliance lists, and internal performance records), and publicly available external data (business registration, financial information, ESG reports, legal proceedings, environmental penalties, etc.). It constructs a value assessment index system oriented towards multiple stakeholders. After integrating, cleaning, and processing the data, it uses large language models, knowledge graphs, multi-agent collaboration, and machine learning algorithms to quantitatively assess the symbiotic relationship between the enterprise and various stakeholders, and automatically generates multiple versions of value symbiosis reports. This method not only provides the enterprise's comprehensive rating and total ecosystem value across various stakeholder dimensions but also outputs risk assessments and collaborative suggestions, effectively supporting enterprise strategic decision-making and sustainable development management. Summary of the Invention

[0004] This invention provides a method for evaluating enterprise value symbiotic reporting intelligent agents based on an artificial intelligence large-scale model. The evaluation object of this method is general enterprises with a unified social credit code starting with 91 and possessing complete operating data. The specific evaluation process relies on three parts of data: first, historical transaction data between the enterprise and its stakeholders, including accounts receivable, advances from customers, main business revenue, and other business revenue related to customers; accounts payable, prepayments, and procurement costs related to suppliers; paid-in capital, capital reserves, and dividends payable related to shareholders; employee compensation payable and labor costs in R&D expenses related to employees; short-term loans, long-term loans, and interest expenses in financial expenses related to creditors; and environmental depreciation of environmental protection equipment and environmental protection investment related to the environment. The data includes: 1) ending balances or amounts of accounting items related to social welfare, such as taxes and surcharges, and charitable donations; 2) supplementary data accumulated during business operations related to stakeholders, including litigation and non-compliance lists with customers, contract disputes and non-compliance lists with suppliers, internal records of shareholder equity pledge defaults and related-party transaction violations, labor disputes and disciplinary infraction lists with employees, internal records of debt defaults with creditors, internal environmental monitoring records of exceeding standards and self-inspections of environmental accidents, and community complaints and internal reports of work safety accidents; 3) publicly available external data, including business registration, financial information, tax arrears information, judicial litigation, administrative penalties, ESG reports and ratings, environmental penalties, and public opinion information. By integrating, cleaning, and processing these three types of data, leveraging large-scale artificial intelligence models and machine learning algorithms, and integrating expert knowledge and experience to establish rules, the system evaluates the symbiotic value between the enterprise and its stakeholders. While outputting risk levels and comprehensive ecological value across various dimensions, it automatically generates a value symbiotic report with strategic insights.

[0005] To achieve the objective of this invention, the technical solution adopted is as follows: A method for evaluating enterprise value symbiotic reporting intelligent agents based on a large-scale artificial intelligence model includes the following steps: S1. Multi-source data integration and preprocessing: Extract historical transaction data accumulated by the enterprise from the NC database, cooperate with other platforms to obtain and update external data, manually supplement relevant data with various stakeholders on a monthly basis, integrate historical transaction data, external data, and supplemented data to form a multi-theme data warehouse for stakeholder analysis, and write SQL scripts to obtain years of accounting subject data from the multi-theme data warehouse for historical transaction data modeling and obtain publicly available information of the enterprise for external data modeling. S2. Risk Rating and Value Quantification Indicator Construction: For customer, supplier, shareholder, employee, creditor, environmental, and social dimensions, set historical transaction indicators corresponding to historical transaction data, internal supplementary data indicators corresponding to supplementary data, and external indicators corresponding to external data to form a multi-dimensional, multi-stakeholder risk rating. S3. Actual Value Calculation and Assignment of Indicators: For the financial ratio indicators in the historical trading indicators corresponding to the historical trading data, use the pandas quantile function to calculate the quartiles of each indicator, divide the intervals and assign values. For positive indicators in the historical trading indicators corresponding to the historical trading data, the interval order is reversed. For the absolute value indicators in the historical trading indicators corresponding to the historical trading data and the internal supplementary indicators corresponding to the supplementary data, directly sort them, divide the indicator quartiles and assign values. For ratio-type supplementary indicators among the internal supplementary indicators corresponding to supplementary data and ratio-type external indicators among the external indicators corresponding to external data, calculate the actual values ​​of the indicators, sort them, determine the quartiles of the indicators, and assign values; for no-type indicators among the external indicators corresponding to external data, assign values ​​based on whether they are triggered. S4. Model Construction and Indicator Weight Generation: For historical transaction data, a model learning dataset is constructed using inference of qualified and unqualified lists. The model is trained using the random forest algorithm and optimized through grid search. The feature-importance method is used to obtain the weight vector of each indicator, and the overall risk score of the historical transaction data is calculated. For supplementary data, the indicator weights are directly set by expert knowledge, such as using the analytic hierarchy process or equal weighting plus key indicator weighting. These weights do not participate in machine learning modeling. The risk score of the supplementary data is calculated by multiplying the indicator risk score by the internal supplementary indicator weights. For external data, a combination of expert knowledge weighting and machine learning is used to set indicator weights. Risk assessment indicators are constructed and assigned values ​​using Python's pandas and numpy. The risk score of the external data is calculated by multiplying the indicator risk score by the external indicator weights. S5. Risk Score Calculation and Dimension Rating: For each stakeholder such as customers, suppliers, shareholders, employees, creditors, environment, and society, the internal risk score is calculated by multiplying the internal supplementary indicator weight by the indicator risk score, and the external risk score is calculated by multiplying the external indicator weight by the indicator risk score. The internal and external risk scores are sorted separately to obtain quartiles. The quartiles are then used to divide the intervals and assign values ​​to obtain the internal and external risk levels of each stakeholder, such as 0-normal, 1-vigilant, and 2-cautious risk levels. S6. Risk Level Integration and Comprehensive Rating and Enterprise Ecosystem Total Value Acquisition: For each stakeholder, their internal and external risk levels are integrated on the horizontal and vertical axes of a nine-square heatmap to generate a comprehensive rating. The total value of the enterprise ecosystem is calculated based on the value contribution of each stakeholder and the weights set according to industry characteristics and strategic importance. S7. Setting Strong Rules and Revising Comprehensive Rating: Based on the comprehensive rating fusion rules in step S6, the comprehensive rating is revised according to the strong rule indicators in the relevant supplementary data of each stakeholder: If any strong rule is triggered, the comprehensive rating of that stakeholder is directly adjusted to 2-Cautious; at the same time, when generating the value symbiosis report, the triggered strong rule matters are listed separately as key risk matters. S8. Collaborative Generation of Value Symbiosis Report: Employs a large language model to perform entity recognition, relation extraction, and event extraction on unstructured text, outputting structured information. Based on the Neo4j graph database, it constructs a dynamic knowledge graph of "enterprise-stakeholder-value stream," calls the enterprise ecosystem total value valuation model in step S6, calculates the value of each stakeholder and the total value of the enterprise ecosystem, and automatically generates a value symbiosis report with charts, attribution analysis, and strategic recommendations by using a built-in report template. S9. Visualized Output and Decision Support: Provides senior management with a dashboard interface that displays the overall value trend of the ecosystem, risk heatmaps, and comparisons of collaborative solutions; provides customized dashboards for business departments such as supply chain, human resources, sustainable development, finance, legal affairs, and risk management; and provides dedicated portals for stakeholders such as suppliers, employees, communities, customers, shareholders, and creditors to showcase collaborative value, individual contributions, and collaborative progress.

[0006] As a preferred embodiment, the specific operation of step S1 in the agent evaluation method is as follows: S1.1. Use ETL tools to extract historical transaction data related to customers, suppliers, shareholders, employees, creditors, environment, and society from the NC database on a monthly basis. The historical transaction data includes the ending balances and amounts of the following eighteen categories of accounting items: accounts receivable, accounts payable, employee compensation payable, environmental protection investment, donation expenditure, advances from customers, main business revenue, other business revenue, procurement costs, paid-in capital, capital reserve, dividends payable, labor costs in R&D expenses, short-term loans, long-term loans, interest expenses in financial expenses, depreciation of environmental protection equipment, and taxes and surcharges. External data is obtained through cooperation with other platforms and updated monthly. S1.2 Manually supplement the data monthly with twelve categories related to various stakeholders, including litigation, non-compliance lists, internal performance data, environmental accidents, community complaints, contract disputes, equity pledge defaults, related-party transaction violations, labor disputes, debt defaults, internal environmental monitoring exceedance records, and internal reports of work safety accidents. Integrate historical transaction data, external data, and supplementary data to form a multi-theme data warehouse for stakeholder analysis. The multi-theme data warehouse is a theme-oriented, integrated, relatively stable data set that reflects historical changes, used to support symbiotic value analysis and risk modeling. S1.3. Obtain the required multi-year accounting subject data from the multi-theme data warehouse for historical transaction data modeling by writing SQL scripts, and obtain publicly available enterprise information from the multi-theme data warehouse for external data modeling.

[0007] Furthermore, in step S1.1 of the intelligent agent evaluation method, other platforms are entities that focus on collecting information about social entities, with Tianyancha, Qichacha, and Qixinbao being the preferred choices.

[0008] Furthermore, the publicly available enterprise information in step S1.3 of the intelligent agent evaluation method includes the enterprise's business registration information, financial information, credit information, operational information, tax information, and judicial litigation information.

[0009] As a preferred embodiment, step S2 of the agent evaluation method is specifically performed as follows: S2.1. For the customer dimension, set historical transaction indicators corresponding to historical transaction data: accounts receivable turnover days, sales percentage of the top five customers, net profit, operating cash ratio, operating cash collection rate, and customer volatility; set internal supplementary data indicators corresponding to supplementary data: customer service system construction score, customer satisfaction survey results (NPS), and client litigation and complaint records; set external indicators corresponding to external data: ESG rating customer relationship specific score, customer service chapter text refinement and relationship evaluation. S2.2. For the supplier dimension, set historical transaction indicators corresponding to historical transaction data: accounts payable turnover days, accounts payable as a percentage of total liabilities, prepayment turnover rate, supply chain efficiency, and supplier volatility; set internal supplementary data indicators corresponding to supplementary data: supplier access assessment standard score, supplier support plan implementation status, and supplier litigation records; set external indicators corresponding to external data: ESG rating supply chain specific score, supply chain chapter text refinement and relationship evaluation. S2.3. For the shareholder dimension, set historical transaction indicators corresponding to historical transaction data: shareholder income (dividends), shareholding ratio of the top ten shareholders, shareholder nature, equity concentration, shareholding ratio of the largest shareholder, and details of related-party transactions; set internal supplementary data indicators corresponding to supplementary data: equity structure and actual controller information, implementation status of incentive and constraint mechanisms, relevant documents on shareholder rights protection and governance participation, and internal records of shareholder-related litigation or arbitration; set external indicators corresponding to external data: consecutive annual scores / ratings of ESG governance dimension, and information on shareholder-related litigation, arbitration, or major penalties. S2.4. For the employee dimension, set historical transaction indicators corresponding to historical transaction data: employee income (value-added caliber), total executive compensation, average compensation per ordinary employee, internal and external compensation gap, employee training investment or welfare expenditure or severance pay amount; set internal supplementary data indicators corresponding to supplementary data: employee rights protection information, employee development and incentive details, employment and personnel adjustment records, internal records of labor disputes, employee care and employer branding texts; set external indicators corresponding to external data: ESG social responsibility dimension annual scores or rating levels (employee sub-indicators), labor dispute related litigation or arbitration or major penalty information; S2.5. For the creditor dimension, set historical transaction indicators corresponding to historical transaction data: creditor income (interest expense), debt-to-equity ratio, current ratio, quick ratio, interest coverage ratio, debt cost, interest-bearing debt structure, and cash and cash equivalents to short-term debt ratio; set internal supplementary data indicators corresponding to supplementary data: corporate entity or bond credit rating, bond and debt repayment arrangements details, creditor protection mechanism, financial risk management system construction status, and internal records of debt contract disputes; set external indicators corresponding to external data: ESG governance dimension (risk management or compliance operation) score or rating level, corporate entity or bond credit rating (external), information on debt contract disputes related to litigation or arbitration or major penalties, and external evaluation of corporate debt repayment ability and credit status. S2.6. For the environmental dimension, set historical transaction indicators corresponding to historical transaction data: environmental protection investment (financial indicator), environmental penalty amount (if already recorded); set internal supplementary data indicators corresponding to supplementary data: total greenhouse gas emissions, greenhouse gas emissions per unit of output, total comprehensive energy consumption, comprehensive energy consumption per unit of output, total water consumption, compliant disposal rate of hazardous waste, proportion of green products, renewable energy usage, internal records of environmental violations, and internal records of environmental litigation; set external indicators corresponding to external data: qualitative description of ESG reports (environmental performance section), environmental administrative penalty decisions, environmental litigation judgments, records of violations of laws and regulations and litigation (environmental category). S2.7. For the social dimension, set historical transaction indicators corresponding to historical transaction data: donation expenditure, government revenue (various taxes and fees); set internal supplementary data indicators corresponding to supplementary data: cumulative number of public welfare projects (including school construction and rural revitalization projects), rural revitalization investment amount, number of community volunteer service visits and duration, emergency public welfare assistance records, non-financial indicator evaluation (internal self-evaluation); set external indicators corresponding to external data: non-financial indicator evaluation (external third-party scoring), tax level, work safety accident reports, negative reports from mainstream media (social reputation).

[0010] As a preferred embodiment, step S3 of the agent evaluation method is specifically performed as follows: S3.1 For the financial ratio indicators in the historical transaction indicators corresponding to historical transaction data, including accounts receivable turnover days, accounts payable turnover days, net profit, operating cash ratio, operating cash collection ratio, debt-to-equity ratio, current ratio, quick ratio, and interest coverage ratio, use the pandas quantile function to calculate the quartiles Q1 and Q3 of each indicator, dividing them into three intervals: [0, Q1], (Q1, Q3), and [Q3, +∞), corresponding to the three risk intervals of normal, vigilance, and caution, respectively. Assign three risk scores to the indicators: 0-normal, 1-vigilance, and 2-caution. For the positive indicators in the historical transaction indicators corresponding to historical transaction data, including net profit margin, operating cash ratio, accounts receivable turnover ratio, and current ratio, since the larger the value of the positive indicator, the lower the risk, the interval order is reversed. S3.2 For the absolute value indicators in the historical transaction indicators corresponding to the historical transaction data and the internal supplementary indicators corresponding to the supplementary data, including registered capital, environmental protection investment amount, donation expenditure, government revenue, and employee training investment amount, directly sort and define the quartiles of the indicators, and assign three risk scores of 0-normal, 1-warning, and 2-cautious. S3.3. The ratio-type supplementary indicators in the internal supplementary indicators corresponding to the supplementary data and the ratio-type external indicators in the external indicators corresponding to the external data include the ratio of the guarantee amount to the registered capital, the ratio of the cumulative amount involved in the case to the registered capital, the greenhouse gas emissions per unit of output value, the ratio of total comprehensive energy consumption to output value, and the compliance disposal rate of hazardous waste. After calculating the actual values ​​of the indicators, they are sorted and quartiles are determined, and three risk scores of 0-normal, 1-vigilance, and 2-caution are assigned. S3.4. Regarding "whether" type indicators: General "whether" indicators in external indicators corresponding to external data include whether there are administrative penalties, whether there are environmental penalties, whether there are tax arrears, and whether there are lawsuits. They are assigned two risk scores: 0-not triggered and 1-triggered, depending on whether they are triggered. Important risk "whether" indicators in external indicators corresponding to external data include whether there are serious violations of trust, whether there are dishonest entities subject to enforcement, whether there are major tax violations, and whether there is an entity listed in the list of abnormal business operations. They are assigned two risk scores: 0-not triggered and 2-triggered, depending on whether they are triggered.

[0011] As a preferred embodiment, step S4 of the agent evaluation method is specifically performed as follows: S4.1. Based on historical transaction data, including risk score information for each indicator, label information for qualified and unqualified lists, multi-year time series data information, and industry control variable information, construct a model learning dataset from qualified and unqualified list samples, train the model using the random forest algorithm, optimize it through grid search, obtain the weight vector ω of each indicator using the feature-importance method, and calculate the overall risk score of the internal data. S4.2 For the supplementary data, the weights of each indicator are directly set using expert knowledge. Specifically, equal weights are assigned to general internal supplementary indicators, and double weights are assigned to key risk indicators such as litigation records, unqualified lists, and major internal performance violations. The sum of the weights of all internal supplementary indicators is 1. No machine learning model training is performed. After the risk scores of each indicator are sorted, the quartiles of the indicators are determined and assigned values. The internal risk score is calculated by multiplying the internal supplementary indicator weight by the indicator risk score. S4.3 For external data, including business registration information, financial information, credit information, operating information, tax information, judicial litigation information, ESG reports and ratings, environmental penalties, and public opinion information, a combination of expert knowledge empowerment and machine learning is adopted. Indicator weights are set and the sum of the weights of all external indicators is 1. Risk assessment indicators are constructed and assigned values ​​using Python's pandas and numpy. The risk score of external data is calculated by summing the risk scores of each indicator of stakeholders multiplied by the indicator weights.

[0012] Furthermore, the specific operation of step S4.1 in the agent evaluation method is as follows: S4.1.1 Sample of qualified and unqualified reasoning lists: Unqualified list: For customers, those with accounts receivable outstanding for two consecutive years (beginning balance = ending balance > 0, cumulative debit = cumulative credit = 0) are added to the unqualified list; for suppliers, those with accounts payable outstanding for two consecutive years without any transactions are added to the unqualified list; for employees, those with unsatisfactory performance appraisals for two consecutive years and who have committed serious misconduct are added to the unqualified list. Qualified list: For customers, customers whose sum of the risk scores of the four indicators of accounts receivable turnover rate and collection rate in the previous year and the previous two years is less than or equal to 2 are included in the qualified list; for suppliers, suppliers whose comprehensive score of indicators such as accounts payable turnover rate and on-time delivery rate is less than or equal to 2 is included in the qualified list; for employees, suppliers whose comprehensive score of training investment, promotion rate, satisfaction rate, etc. is less than or equal to 2 is included in the qualified list. S4.1.2 Constructing the model learning dataset: Add the risk score data of the corresponding indicators for the three and four years prior to the unqualified list and qualified list samples to the industry control variables (one-hot encoding), and label them as unqualified list and qualified list, such as unqualified list = 1, qualified list = 0, to obtain the model learning data; S4.1.3. The random forest algorithm is used to train the model, and the grid search is used for optimization. The weight vector ω of each index is obtained using the feature-importance method. S4.1.4 Calculate the overall risk score of historical transaction data: Using the previous year as a benchmark, take the risk scores of the four core indicators of the sample, add industry control variables, and form an indicator matrix A. The risk score S = A × ω T Based on the scores, quartiles are calculated, and three risk levels are divided: 0-normal, 1-alert, and 2-cautious.

[0013] Furthermore, the specific operation of step S4.2 in the agent evaluation method is as follows: S4.2.1 The basic indicators are set to equal weight, but for the five key risk indicators, namely the cumulative amount involved in cases as a percentage of registered capital, the amount of environmental penalties, the amount involved in major litigation, the amount of environmental penalties, and the amount of tax arrears, the weight is set to twice that of the other three general indicators, such as registered capital, environmental protection investment, and donation expenditure. The sum of the weights of all external indicators is 1. S4.2.2 Use Python's pandas and numpy to construct risk assessment indicators. Assign three risk levels (0-normal, 1-alert, 2-cautious) to absolute value indicators and ratio indicators based on quartiles. Assign two risk levels (0-no, 1-yes) or two risk levels (0-no, 2-yes) to yes / no indicators based on whether they are general or important. S4.2.3. Based on the risk score of each stakeholder's indicators multiplied by the indicator weight, the external data risk score is obtained. The threshold is set according to the distribution of the score data clustering, and the three levels are divided into normal, vigilance and caution, and assigned three risk levels respectively: 0-normal, 1-vigilance and 2-caution.

[0014] As a preferred embodiment, step S6 of the agent evaluation method is specifically performed as follows: S6.1 For each stakeholder, its internal risk level and external risk level are respectively placed on the horizontal and vertical axes of a nine-square grid heatmap. The horizontal axis represents internal risk levels from left to right: Normal, Cautious, and Safe; the vertical axis represents external risk levels from bottom to top: Normal, Cautious, and Safe. The comprehensive rating integration rules are as follows: The overall rating is normal: the combination is (normal, normal), (normal, caution), (caution, normal). The overall rating is "Caution": the combination is (Caution, Caution), (Normal, Cautious), (Cautious, Normal). The overall rating is cautious: the combination is (cautious, cautious), (cautious, cautious), (cautious, cautious); S6.2, Total Enterprise Ecosystem Value V-total = Σ(Vi × wi), where Vi represents the value contribution of each stakeholder, such as customer lifetime value (LTV), employee value creation, supplier synergy value, shareholder returns, creditor interest, positive environmental externalities, and social contribution value; wi represents the weights assigned based on industry characteristics and strategic importance; Vi is calculated using the following model: Customer Lifetime Value: The Customer Lifetime Value model LTV = Σ(M×R / (1+i)^t) is used to sum up all customers; Supplier Collaboration Value: Employing a supply chain collaboration benefit model, we quantify the cost savings and revenue growth resulting from on-time supplier delivery and joint R&D. Employee value creation: Using the human capital value model, employee income (total compensation) + excess output per employee; Shareholder income: Shareholder income (dividends + share buybacks). Creditor interest: Interest expense; Environmental positive externality discounting: potential environmental cost savings from environmental protection investments (discounted using the cost avoidance method or carbon trading price method); Social contribution value: donation expenditure + government revenue (taxes) + community benefit calculation.

[0015] As a preferred embodiment, the specific operation of step S8 in the agent evaluation method is as follows: S8.1 Information Extraction Intelligent Agent: Using a large language model, it performs entity recognition, relation extraction, and event extraction on unstructured texts such as ESG reports, litigation documents, public opinion comments, annual reports, social responsibility reports, environmental reports, and board reports, and outputs structured information; S8.2 Relationship Mapping and Knowledge Graph Construction of Intelligent Agents: Based on the Neo4j graph database, a dynamic knowledge graph of "enterprise-stakeholder-value stream" is constructed. Nodes include enterprises, customers, suppliers, shareholders, employees, creditors, environmental indicators, social events, products, projects, technologies, and events. Relationships include employment, service, cooperation, investment, lending, emissions, donation, procurement, sales, governance, and litigation. The centrality, community structure, and value flow path of each stakeholder node in the knowledge graph are calculated periodically. S8.3 Value Quantification and Scenario Simulation Agent: Calls the enterprise ecosystem total value valuation model in step S6.2 to calculate the value of each stakeholder and the total value of the enterprise ecosystem; supports natural language definition of scenarios (such as "if the employee turnover rate decreases by 5%), drives Monte Carlo simulation, and outputs the impact on cash flow, discount rate, and overall valuation. S8.4 Report Generation and Insight Agent: Built-in report templates automatically generate value symbiotic reports with charts, attribution analysis and strategic recommendations, and support natural language question-and-answer interaction.

[0016] Furthermore, the large language model in step S8.1 of the agent evaluation method is a DeepSeek+Qwen3 dual-model collaborative architecture.

[0017] The beneficial effects of this invention are as follows: This invention's intelligent agent evaluation method combines internal historical transaction data, stakeholder supplementary data, and publicly available external big data. Through the deep integration of large language models, knowledge graphs, multi-agent collaboration, and machine learning algorithms, it effectively evaluates the quality of the symbiotic relationship and the total ecological value between an enterprise and its core stakeholders. This provides comprehensive information support for enterprises to optimize customer cooperation, supply chain management, employee motivation, environmental governance, and social responsibility fulfillment. Simultaneously, the system can automatically generate multiple versions of value symbiosis reports, supports scenario simulation and natural language interaction, and significantly improves the intelligence and automation level of enterprise value assessment. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall architecture of the enterprise value symbiosis report intelligent agent evaluation method based on the artificial intelligence big model of the present invention. Detailed Implementation

[0019] The present invention will be further described and illustrated below with reference to specific embodiments.

[0020] 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.

[0021] 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.

[0022] like Figure 1 As shown, a method for evaluating enterprise value symbiotic reporting intelligent agents based on a large-scale artificial intelligence model specifically includes the following steps: S1. Multi-source data integration and preprocessing: Extract historical transaction data accumulated by the enterprise from the NC database, cooperate with other platforms to obtain and update external data, manually supplement relevant data with various stakeholders on a monthly basis, integrate historical transaction data, external data, and supplemented data to form a multi-theme data warehouse for stakeholder analysis, and write SQL scripts to obtain years of accounting subject data from the multi-theme data warehouse for historical transaction data modeling and obtain publicly available information of the enterprise for external data modeling. S2. Risk Rating and Value Quantification Indicator Construction: For customer, supplier, shareholder, employee, creditor, environmental, and social dimensions, set historical transaction indicators corresponding to historical transaction data, internal supplementary data indicators corresponding to supplementary data, and external indicators corresponding to external data to form a multi-dimensional, multi-stakeholder risk rating. S3. Actual Value Calculation and Assignment of Indicators: For the financial ratio indicators in the historical trading indicators corresponding to the historical trading data, use the pandas quantile function to calculate the quartiles of each indicator, divide the intervals and assign values. For positive indicators in the historical trading indicators corresponding to the historical trading data, the interval order is reversed. For the absolute value indicators in the historical trading indicators corresponding to the historical trading data and the internal supplementary indicators corresponding to the supplementary data, directly sort them, divide the indicator quartiles and assign values. For ratio-type supplementary indicators among the internal supplementary indicators corresponding to supplementary data and ratio-type external indicators among the external indicators corresponding to external data, calculate the actual values ​​of the indicators, sort them, determine the quartiles of the indicators, and assign values; for no-type indicators among the external indicators corresponding to external data, assign values ​​based on whether they are triggered. S4. Model Construction and Indicator Weight Generation: For historical transaction data, a model learning dataset is constructed using inference of qualified and unqualified lists. The model is trained using the random forest algorithm and optimized through grid search. The feature-importance method is used to obtain the weight vector of each indicator, and the overall risk score of the historical transaction data is calculated. For supplementary data, the indicator weights are directly set by expert knowledge, such as using the analytic hierarchy process or equal weighting plus key indicator weighting. These weights do not participate in machine learning modeling. The risk score of the supplementary data is calculated by multiplying the indicator risk score by the internal supplementary indicator weights. For external data, a combination of expert knowledge weighting and machine learning is used to set indicator weights. Risk assessment indicators are constructed and assigned values ​​using Python's pandas and numpy. The risk score of the external data is calculated by multiplying the indicator risk score by the external indicator weights. S5. Risk Score Calculation and Dimension Rating: For each stakeholder such as customers, suppliers, shareholders, employees, creditors, environment, and society, the internal risk score is calculated by multiplying the internal supplementary indicator weight by the indicator risk score, and the external risk score is calculated by multiplying the external indicator weight by the indicator risk score. The internal and external risk scores are sorted separately to obtain quartiles. The quartiles are then used to divide the intervals and assign values ​​to obtain the internal and external risk levels of each stakeholder, such as 0-normal, 1-vigilant, and 2-cautious risk levels. S6. Risk Level Integration and Comprehensive Rating and Enterprise Ecosystem Total Value Acquisition: For each stakeholder, their internal and external risk levels are integrated on the horizontal and vertical axes of a nine-square heatmap to generate a comprehensive rating. The total value of the enterprise ecosystem is calculated based on the value contribution of each stakeholder and the weights set according to industry characteristics and strategic importance. S7. Setting Strong Rules and Revising Comprehensive Rating: Based on the comprehensive rating fusion rules in step S6, the comprehensive rating is revised according to the strong rule indicators in the relevant supplementary data of each stakeholder: If any strong rule is triggered, the comprehensive rating of that stakeholder is directly adjusted to 2-Cautious; at the same time, when generating the value symbiosis report, the triggered strong rule matters are listed separately as key risk matters. S8. Collaborative Generation of Value Symbiosis Report: Employs a large language model to perform entity recognition, relation extraction, and event extraction on unstructured text, outputting structured information. Based on the Neo4j graph database, it constructs a dynamic knowledge graph of "enterprise-stakeholder-value stream," calls the enterprise ecosystem total value valuation model in step S6, calculates the value of each stakeholder and the total value of the enterprise ecosystem, and automatically generates a value symbiosis report with charts, attribution analysis, and strategic recommendations by using a built-in report template. S9. Visualized Output and Decision Support: Provides senior management with a dashboard interface that displays the overall value trend of the ecosystem, risk heatmaps, and comparisons of collaborative solutions; provides customized dashboards for business departments such as supply chain, human resources, sustainable development, finance, legal affairs, and risk management; and provides dedicated portals for stakeholders such as suppliers, employees, communities, customers, shareholders, and creditors to showcase collaborative value, individual contributions, and collaborative progress.

[0023] like Figure 1 As shown, a method for evaluating intelligent agents in enterprise value symbiosis reporting based on a large artificial intelligence model is described. The specific operation of step S1 in the intelligent agent evaluation method is as follows: S1.1. Use ETL tools to extract historical transaction data related to customers, suppliers, shareholders, employees, creditors, environment, and society from the NC database on a monthly basis. The historical transaction data includes the ending balances and amounts of the following eighteen categories of accounting items: accounts receivable, accounts payable, employee compensation payable, environmental protection investment, donation expenditure, advances from customers, main business revenue, other business revenue, procurement costs, paid-in capital, capital reserve, dividends payable, labor costs in R&D expenses, short-term loans, long-term loans, interest expenses in financial expenses, depreciation of environmental protection equipment, and taxes and surcharges. External data is obtained through cooperation with other platforms and updated monthly. S1.2 Manually supplement the data monthly with twelve categories of data related to various stakeholders, including litigation, qualified and unqualified lists, internal performance, environmental accidents, community complaints, contract disputes, equity pledge defaults, related-party transaction violations, labor disputes, debt defaults, internal environmental monitoring exceedance records, and internal reports of work safety accidents. Integrate historical transaction data, external data, and supplementary data to form a multi-theme data warehouse for stakeholder analysis. This multi-theme data warehouse is a theme-oriented, integrated, relatively stable data set reflecting historical changes, used to support symbiotic value analysis and risk modeling. S1.3. Obtain the required multi-year accounting subject data from the multi-theme data warehouse for historical transaction data modeling by writing SQL scripts, and obtain publicly available enterprise information from the multi-theme data warehouse for external data modeling.

[0024] Furthermore, in step S1.1 of the intelligent agent evaluation method, other platforms are those that focus on collecting information about social entities, with Tianyancha, Qichacha, and Qixinbao being the preferred choices.

[0025] Furthermore, the publicly available enterprise information in step S1.3 of the intelligent agent evaluation method includes the enterprise's business registration information, financial information, credit information, operational information, tax information, and judicial litigation information.

[0026] like Figure 1As shown, a method for evaluating intelligent agents in enterprise value symbiosis reporting based on a large artificial intelligence model is described. The specific operation of step S2 in the intelligent agent evaluation method is as follows: S2.1. For the customer dimension, set historical transaction indicators corresponding to historical transaction data: accounts receivable turnover days, sales percentage of the top five customers, net profit, operating cash ratio, operating cash collection rate, and customer volatility; set internal supplementary data indicators corresponding to supplementary data: customer service system construction score, customer satisfaction survey results (NPS), and client litigation and complaint records; set external indicators corresponding to external data: ESG rating customer relationship specific score, customer service chapter text refinement and relationship evaluation. S2.2. For the supplier dimension, set historical transaction indicators corresponding to historical transaction data: accounts payable turnover days, accounts payable as a percentage of total liabilities, prepayment turnover rate, supply chain efficiency, and supplier volatility; set internal supplementary data indicators corresponding to supplementary data: supplier access assessment standard score, supplier support plan implementation status, and supplier litigation records; set external indicators corresponding to external data: ESG rating supply chain specific score, supply chain chapter text refinement and relationship evaluation. S2.3. For the shareholder dimension, set historical transaction indicators corresponding to historical transaction data: shareholder income (dividends), shareholding ratio of the top ten shareholders, shareholder nature, equity concentration, shareholding ratio of the largest shareholder, and details of related-party transactions; set internal supplementary data indicators corresponding to supplementary data: equity structure and actual controller information, implementation status of incentive and constraint mechanisms, relevant documents on shareholder rights protection and governance participation, and internal records of shareholder-related litigation or arbitration; set external indicators corresponding to external data: consecutive annual scores / ratings of ESG governance dimension, and information on shareholder-related litigation, arbitration, or major penalties. S2.4. For the employee dimension, set historical transaction indicators corresponding to historical transaction data: employee income (value-added caliber), total executive compensation, average compensation per ordinary employee, internal and external compensation gap, employee training investment or welfare expenditure or severance pay amount; set internal supplementary data indicators corresponding to supplementary data: employee rights protection information, employee development and incentive details, employment and personnel adjustment records, internal records of labor disputes, employee care and employer branding texts; set external indicators corresponding to external data: ESG social responsibility dimension annual scores or rating levels (employee sub-indicators), labor dispute related litigation or arbitration or major penalty information; S2.5. For the creditor dimension, set historical transaction indicators corresponding to historical transaction data: creditor income (interest expense), debt-to-equity ratio, current ratio, quick ratio, interest coverage ratio, debt cost, interest-bearing debt structure, and cash and cash equivalents to short-term debt ratio; set internal supplementary data indicators corresponding to supplementary data: corporate entity or bond credit rating, bond and debt repayment arrangements details, creditor protection mechanism, financial risk management system construction status, and internal records of debt contract disputes; set external indicators corresponding to external data: ESG governance dimension (risk management or compliance operation) score or rating level, corporate entity or bond credit rating (external), information on debt contract disputes related to litigation or arbitration or major penalties, and external evaluation of corporate debt repayment ability and credit status. S2.6. For the environmental dimension, set historical transaction indicators corresponding to historical transaction data: environmental protection investment (financial indicator), environmental penalty amount (if already recorded); set internal supplementary data indicators corresponding to supplementary data: total greenhouse gas emissions, greenhouse gas emissions per unit of output, total comprehensive energy consumption, comprehensive energy consumption per unit of output, total water consumption, compliant disposal rate of hazardous waste, proportion of green products, renewable energy usage, internal records of environmental violations, and internal records of environmental litigation; set external indicators corresponding to external data: qualitative description of ESG reports (environmental performance section), environmental administrative penalty decisions, environmental litigation judgments, records of violations of laws and regulations and litigation (environmental category). S2.7. For the social dimension, set historical transaction indicators corresponding to historical transaction data: donation expenditure, government revenue (various taxes and fees); set internal supplementary data indicators corresponding to supplementary data: cumulative number of public welfare projects (including school construction and rural revitalization projects), rural revitalization investment amount, number of community volunteer service visits and duration, emergency public welfare assistance records, non-financial indicator evaluation (internal self-evaluation); set external indicators corresponding to external data: non-financial indicator evaluation (external third-party scoring), tax level, work safety accident reports, negative reports from mainstream media (social reputation).

[0027] like Figure 1 As shown, a method for evaluating intelligent agents in enterprise value symbiosis reporting based on a large artificial intelligence model is described. The specific operation of step S3 in the intelligent agent evaluation method is as follows: S3.1 For the financial ratio indicators in the historical transaction indicators corresponding to historical transaction data, including accounts receivable turnover days, accounts payable turnover days, net profit, operating cash ratio, operating cash collection ratio, debt-to-equity ratio, current ratio, quick ratio, and interest coverage ratio, use the pandas quantile function to calculate the quartiles Q1 and Q3 of each indicator, dividing them into three intervals: [0, Q1], (Q1, Q3), and [Q3, +∞), corresponding to the three risk intervals of normal, vigilance, and caution, respectively. Assign three risk scores to the indicators: 0-normal, 1-vigilance, and 2-caution. For the positive indicators in the historical transaction indicators corresponding to historical transaction data, including net profit margin, operating cash ratio, accounts receivable turnover ratio, and current ratio, since the larger the value of the positive indicator, the lower the risk, the interval order is reversed. S3.2 For the absolute value indicators in the historical transaction indicators corresponding to the historical transaction data and the internal supplementary indicators corresponding to the supplementary data, including registered capital, environmental protection investment amount, donation expenditure, government revenue, and employee training investment amount, directly sort and define the quartiles of the indicators, and assign three risk scores of 0-normal, 1-warning, and 2-cautious. S3.3. The ratio-type supplementary indicators in the internal supplementary indicators corresponding to the supplementary data and the ratio-type external indicators in the external indicators corresponding to the external data include the ratio of the guarantee amount to the registered capital, the ratio of the cumulative amount involved in the case to the registered capital, the greenhouse gas emissions per unit of output value, the ratio of total comprehensive energy consumption to output value, and the compliance disposal rate of hazardous waste. After calculating the actual values ​​of the indicators, they are sorted and quartiles are determined, and three risk scores of 0-normal, 1-vigilance, and 2-caution are assigned. S3.4. Regarding "whether" type indicators: General "whether" indicators in external indicators corresponding to external data include whether there are administrative penalties, whether there are environmental penalties, whether there are tax arrears, and whether there are lawsuits. They are assigned two risk scores: 0-not triggered and 1-triggered, depending on whether they are triggered. Important risk "whether" indicators in external indicators corresponding to external data include whether there are serious violations of trust, whether there are dishonest entities subject to enforcement, whether there are major tax violations, and whether there is an entity listed in the list of abnormal business operations. They are assigned two risk scores: 0-not triggered and 2-triggered, depending on whether they are triggered.

[0028] like Figure 1 As shown, a method for evaluating intelligent agents in enterprise value symbiosis reporting based on a large artificial intelligence model is described. The specific operation of step S4 in the intelligent agent evaluation method is as follows: S4.1. Based on historical transaction data, including risk score information for each indicator, label information for qualified and unqualified samples, multi-year time series data, and industry control variable information, construct a model learning dataset from qualified and unqualified samples, train the model using the random forest algorithm, optimize it through grid search, obtain the weight vector ω of each indicator using the feature-importance method, and calculate the overall risk score of the internal data. S4.2 For the supplementary data, the weights of each indicator are directly set using expert knowledge. Specifically, equal weights are assigned to general internal supplementary indicators, and double weights are assigned to key risk indicators such as litigation records, unqualified lists, and major internal performance violations. The sum of the weights of all internal supplementary indicators is 1. No machine learning model training is performed. After the risk scores of each indicator are sorted, the quartiles of the indicators are determined and assigned values. The internal risk score is calculated by multiplying the internal supplementary indicator weight by the indicator risk score. S4.3 For external data, including business registration information, financial information, credit information, operating information, tax information, judicial litigation information, ESG reports and ratings, environmental penalties, and public opinion information, a combination of expert knowledge empowerment and machine learning is adopted. Indicator weights are set and the sum of the weights of all external indicators is 1. Risk assessment indicators are constructed and assigned values ​​using Python's pandas and numpy. The risk score of external data is calculated by summing the risk scores of each indicator of stakeholders multiplied by the indicator weights.

[0029] Furthermore, the specific operation of step S4.1 in the agent evaluation method is as follows: S4.1.1 Sample of qualified and unqualified reasoning lists: Unqualified list: For customers, those with accounts receivable outstanding for two consecutive years (beginning balance = ending balance > 0, cumulative debit = cumulative credit = 0) are added to the unqualified list; for suppliers, those with accounts payable outstanding for two consecutive years without any transactions are added to the unqualified list; for employees, those with unsatisfactory performance appraisals for two consecutive years and who have committed serious misconduct are added to the unqualified list. Qualified list: For customers, customers whose sum of the risk scores of the four indicators of accounts receivable turnover rate and collection rate in the previous year and the previous two years is less than or equal to 2 are included in the qualified list; for suppliers, suppliers whose comprehensive score of indicators such as accounts payable turnover rate and on-time delivery rate is less than or equal to 2 is included in the qualified list; for employees, suppliers whose comprehensive score of training investment, promotion rate, satisfaction rate, etc. is less than or equal to 2 is included in the qualified list. S4.1.2 Constructing the model learning dataset: Add the risk score data of the corresponding indicators for the three and four years prior to the unqualified list and qualified list samples to the industry control variables (one-hot encoding), and label them as unqualified list and qualified list, such as unqualified list = 1, qualified list = 0, to obtain the model learning data; S4.1.3. The random forest algorithm is used to train the model, and the grid search is used for optimization. The weight vector ω of each index is obtained using the feature-importance method. S4.1.4 Calculate the overall risk score of historical transaction data: Using the previous year as a benchmark, take the risk scores of the four core indicators of the sample, add industry control variables, and form an indicator matrix A. The risk score S = A × ω T Based on the scores, quartiles are calculated, and three risk levels are divided: 0-normal, 1-alert, and 2-cautious.

[0030] Furthermore, the specific operation of step S4.2 in the agent evaluation method is as follows: S4.2.1 The basic indicators are set to equal weight, but for the five key risk indicators, namely the cumulative amount involved in cases as a percentage of registered capital, the amount of environmental penalties, the amount involved in major litigation, the amount of environmental penalties, and the amount of tax arrears, the weight is set to twice that of the other three general indicators, such as registered capital, environmental protection investment, and donation expenditure. The sum of the weights of all external indicators is 1. S4.2.2 Use Python's pandas and numpy to construct risk assessment indicators. Assign three risk levels (0-normal, 1-alert, 2-cautious) to absolute value indicators and ratio indicators based on quartiles. Assign two risk levels (0-no, 1-yes) or two risk levels (0-no, 2-yes) to yes / no indicators based on whether they are general or important. S4.2.3. Based on the risk score of each stakeholder's indicators multiplied by the indicator weight, the external data risk score is obtained. The threshold is set according to the distribution of the score data clustering, and the three levels are divided into normal, vigilance and caution, and assigned three risk levels respectively: 0-normal, 1-vigilance and 2-caution.

[0031] like Figure 1 As shown, a method for evaluating intelligent agents in enterprise value symbiosis reporting based on a large artificial intelligence model is described. The specific operation of step S6 in the intelligent agent evaluation method is as follows: S6.1 For each stakeholder, its internal risk level and external risk level are respectively placed on the horizontal and vertical axes of a nine-square grid heatmap. The horizontal axis represents internal risk levels from left to right: Normal, Cautious, and Safe; the vertical axis represents external risk levels from bottom to top: Normal, Cautious, and Safe. The comprehensive rating integration rules are as follows: The overall rating is normal: the combination is (normal, normal), (normal, caution), (caution, normal). The overall rating is "Caution": the combination is (Caution, Caution), (Normal, Cautious), (Cautious, Normal). The overall rating is cautious: the combination is (cautious, cautious), (cautious, cautious), (cautious, cautious); S6.2, Total Enterprise Ecosystem Value V-total = Σ(Vi × wi), where Vi represents the value contribution of each stakeholder, such as customer lifetime value (LTV), employee value creation, supplier synergy value, shareholder returns, creditor interest, positive environmental externalities, and social contribution value; wi represents the weights assigned based on industry characteristics and strategic importance; Vi is calculated using the following model: Customer Lifetime Value: The Customer Lifetime Value model LTV = Σ(M×R / (1+i)^t) is used to sum up all customers; Supplier Collaboration Value: Employing a supply chain collaboration benefit model, we quantify the cost savings and revenue growth resulting from on-time supplier delivery and joint R&D. Employee value creation: Using the human capital value model, employee income (total compensation) + excess output per employee; Shareholder income: Shareholder income (dividends + share buybacks). Creditor interest: Interest expense; Environmental positive externality discounting: potential environmental cost savings from environmental protection investments (discounted using the cost avoidance method or carbon trading price method); Social contribution value: donation expenditure + government revenue (taxes) + community benefit calculation.

[0032] like Figure 1 As shown, a method for evaluating intelligent agents in enterprise value symbiosis reporting based on a large artificial intelligence model is described. The specific operation of step S8 in the intelligent agent evaluation method is as follows: S8.1 Information Extraction Intelligent Agent: Using a large language model, it performs entity recognition, relation extraction, and event extraction on unstructured texts such as ESG reports, litigation documents, public opinion comments, annual reports, social responsibility reports, environmental reports, and board reports, and outputs structured information; S8.2 Relationship Mapping and Knowledge Graph Construction of Intelligent Agents: Based on the Neo4j graph database, a dynamic knowledge graph of "enterprise-stakeholder-value stream" is constructed. Nodes include enterprises, customers, suppliers, shareholders, employees, creditors, environmental indicators, social events, products, projects, technologies, and events. Relationships include employment, service, cooperation, investment, lending, emissions, donation, procurement, sales, governance, and litigation. The centrality, community structure, and value flow path of each stakeholder node in the knowledge graph are calculated periodically. S8.3 Value Quantification and Scenario Simulation Agent: Calls the enterprise ecosystem total value valuation model in step S6.2 to calculate the value of each stakeholder and the total value of the enterprise ecosystem; supports natural language definition of scenarios (such as "if the employee turnover rate decreases by 5%), drives Monte Carlo simulation, and outputs the impact on cash flow, discount rate, and overall valuation. S8.4 Report Generation and Insight Agent: Built-in report templates automatically generate value symbiotic reports with charts, attribution analysis and strategic recommendations, and support natural language question-and-answer interaction.

[0033] Furthermore, the large language model in step S8.1 of the agent evaluation method is a DeepSeek+Qwen3 dual-model collaborative architecture.

[0034] Example 1 like Figure 1 As shown, a method for evaluating an intelligent agent in a corporate value symbiotic reporting model based on artificial intelligence includes the following steps: I. Data Preparation S1: Use ETL tools to extract historical transaction data from the NC database and set up scheduled tasks to update the data of seven stakeholders, namely customers, suppliers, shareholders, employees, creditors, environment, and society, on a monthly basis; cooperate with organizations that focus on collecting social entity information, such as Tianyancha, Qichacha, and Qixinbao, to obtain external data from various stakeholders; and build data warehouses for internal and external data respectively. S2: By writing SQL scripts, the required accounting subject data for many years (6 years in this embodiment, i.e. 2018-2023) is obtained from the internal historical transaction data warehouse for internal historical transaction data modeling, and publicly available corporate information such as business registration information, financial information, credit information, operating information, tax information, judicial litigation information, and ESG rating reports are obtained from the external data warehouse for external data modeling. II. Customer Dimension Modeling 1. Construct risk rating indicators S3: For the customer dimension, set historical transaction indicators: accounts receivable turnover days, sales percentage of the top five customers, net profit margin, operating cash ratio, operating cash collection rate, and customer volatility; set internal supplementary indicators: customer service system construction score, customer satisfaction survey results (NPS), and client litigation and complaint records; set external indicators: ESG rating customer relationship specific score, customer service chapter text refinement and relationship evaluation; 2. Sample lists of qualified and unqualified candidates for reasoning. S4: Based on customers' historical transaction data, customers with accounts receivable outstanding for two consecutive years (beginning balance = ending balance > 0, cumulative debit = cumulative credit = 0) are set as unqualified sample; according to supplier and customer document data, there were 11 lawsuits filed by customers from 2018 to 2023, and the number of complaints peaked at 70 in 2021 and then declined. Customers with such triggering lawsuits or high complaint volume can be considered as candidates for the unqualified list; S5: Customers whose sum of the risk scores of the four indicators—accounts receivable turnover rate, accounts receivable collection rate, and accounts receivable recovery rate of the previous year and the previous two years—is less than or equal to 2 are set as qualified sample list; taking the top five customers as an example, if the sales share of the top five customers is stable at 7.50% in 2023 and the customer volatility drops to 46.67%, such customers can be regarded as qualified sample list. 3. Calculate and assign values ​​to the actual indicators. S6: Taking customer data from 2023 as an example, calculate the actual values ​​of each indicator; accounts receivable turnover days are 22.6 days, sales to the top five customers account for 7.50%, net profit margin is 6.64%, operating cash ratio is 14.07%, operating cash collection rate is 110.00%, and customer volatility is 46.67%; use pandas' quantile function to calculate the quartiles of each indicator, divide the risk into three risk ranges: normal, vigilant, and cautious, and assign them 0, 1, and 2 respectively; S7: Regarding external indicators, the MSCI ESG rating for customer relations achieved an A grade in 2023, assigned a value of 0 (normal); the text evaluation of the customer service section is "driving the digital and intelligent transformation of services, covering the entire lifecycle", and the text sentiment analysis is positive, assigned a value of 0; 4. Construct a learning model and generate indicator weights. S8: Use the Python third-party library sklearn to split the training and test sets and build a random forest model. Add the risk score data of the corresponding indicators for the three and four years prior to the qualified and unqualified lists to the industry control variables (Shenwan Class IV home appliance industry, unique thermal coding), and label them as qualified and unqualified lists (unqualified list = 1, qualified list = 0) to obtain the model learning data; S9: Use grid search to obtain the optimal model, and use the feature-importance method to obtain the weight vector ω of each indicator in the customer dimension; the example weight allocation is as follows: accounts receivable turnover days weight 0.25, customer volatility weight 0.30, sales share of the top five customers weight 0.15, net profit margin weight 0.10, operating cash ratio weight 0.10, operating cash collection rate weight 0.10; S10: Using the previous year as a baseline, calculate the overall risk score for the client's internal data: S = A × ω T Matrix multiplication is calculated using the Python third-party library NumPy; quartiles are calculated based on risk scores, and three levels are divided: normal (0), alert (1), and cautious (2); for example, a core customer with a risk score of 0.15 is classified as normal (0). 5. External Data Modeling S11: External risk assessment indicators were constructed using Python's third-party tools pandas and numpy; for the MSCIIESG rating, BB level was assigned a value of 2 in 2020, BBB level was assigned a value of 1 in 2021-2022, and A level was assigned a value of 0 in 2023; for client-side litigation, there were 2 cases in 2018, assigned a value of 1, and 1 case in 2023, assigned a value of 0; for Black Cat Complaints, there were 45 cases in 2023, a decrease from the peak, assigned a value of 0. S12: Calculate the external data risk score by summing the risk scores of each customer's external indicators multiplied by their respective weights; set thresholds based on the distribution of the score data to divide the risk into three levels: normal, alert, and cautious, assigned values ​​of 0, 1, and 2 respectively; example data is shown in Table 1 below: Table 1. Example of Customer External Data Risk Rating Customer Code MSCI rating risk Litigation risks Complaint risk Comprehensive evaluation of external data CORE-001 0 0 0 0 (Normal) NORMAL-002 1 0 1 1 (Beware) RISK-003 2 1 2 2 (Caution) 6. Integration of internal and external data ratings S13: After obtaining the rating results of the customer's internal and external data respectively, place the internal and external data rating results on the horizontal and vertical axes of the nine-square grid heat map respectively; the horizontal axis is the internal data rating result, from left to right: normal, vigilant, cautious; the vertical axis is the external data rating result, from bottom to top: normal, vigilant, cautious, forming a customer rating nine-square grid. S14: The rules for integrating the comprehensive rating are as follows: The overall rating is normal: the combination is (normal, normal), (normal, caution), (caution, normal). The overall rating is "Caution": the combination is (Caution, Caution), (Normal, Cautious), (Cautious, Normal). The overall rating is cautious: the combination is (cautious, cautious), (cautious, cautious), (cautious, cautious); 7. Revise the overall rating based on the supplementary data. S15: If the customer's overall rating has been obtained in S14, the customer rating will be revised based on the supplementary data accumulated by the enterprise's operations; if the customer has a litigation record with Haier Smart Home (such as 11 customer lawsuits from 2018 to 2023) or appears on the unqualified list, the overall rating will be directly adjusted to cautious (2); the revised data sample table is shown in Table 2 below: Table 2. Example of revised customer comprehensive rating (based on supplementary data) Customer Code Original comprehensive evaluation Whether or not there is litigation with this unit Are you on the list of unqualified personnel in this unit? Revised Comprehensive CORE-001 0 no no 0 NOMRMAL-002 1 no no 1 RISK-003 1 yes no 2 BLACK-004 0 no yes 2 III. Modeling Dimensions such as Suppliers, Shareholders, and Employees S16: Following the modeling process of the customer dimension, complete the risk rating and value quantification of the other six stakeholders, namely suppliers, shareholders, employees, creditors, environment, and society. 8. Supplier Dimension S17: Set historical transaction metrics: accounts payable turnover days, accounts payable as a percentage of total liabilities, prepayments turnover rate, supply chain efficiency (SCE), and supplier volatility (SCS-S); taking 2023 data as an example, accounts payable turnover days were 21.9 days, and supplier volatility was 29.81%, which continued to decline from 47.25% in 2018; S18: Set external indicators: MSCI ESG rating supply chain specific score, reaching A level in 2023; set supplementary indicators: supplier-side litigation, reducing to 0 cases in 2023; after the integration of internal and external data, the overall rating of the supplier dimension is normal (0). 9. Shareholder Dimension S19: According to shareholder, employee and creditor documents, there is a core data gap at the shareholder level; set historical transaction indicators: shareholder income (dividends), shareholding ratio of the top ten shareholders, and equity concentration; since the dividend data is not disclosed in public information, this indicator is missing and must be noted in the report; S20: Set external indicators: Huazheng ESG rating improved from BB in 2009 to A in 2022-2023, with a comprehensive G score of 88.90 (A level) in 2021; Set supplementary recording indicators: internal records of equity pledge defaults and related-party transaction violations; Due to the lack of disclosure of reasons for rating fluctuations (such as downgrading to CCC in 2010), this item is a risk point; 10. Employee Dimension S21: Set historical transaction indicators: employee income (total compensation), total executive compensation, and employee training investment; set internal supplementation indicators: zero employee safety disputes from 2009 to 2020, which is an important positive indicator; set external indicators: MSCI ESG social responsibility dimension score, reaching BBB level (90.14 points) in 2017. S22: Regarding the layoffs from 2011 to 2015, since the compensation plan and reasons were not disclosed, they are listed as risk items requiring attention in accordance with the strict rules, but not directly downgraded; they are only mentioned in the report. 11. Creditor Dimension S23: Based on shareholder, employee, and creditor documents, core indicators for the creditor dimension are missing; historical transaction indicators should be set: debt-to-equity ratio, current ratio, and interest expense. External indicators should be set: corporate credit rating and external evaluation of solvency; due to missing data, a comprehensive rating cannot be calculated, and "data gap" must be clearly marked in the report; 12. Environmental Dimension S24: Based on the social and environmental documents, set historical transaction indicators: environmental protection investment amount (increased from 180 million yuan in 2018 to 550 million yuan in 2024), and green product sales revenue ratio (over 60% in 2024); set internal supplementary indicators: total greenhouse gas emissions (650,677.17 tons in 2023), greenhouse gas emissions per unit of output value (24.89 kg / 10,000 yuan in 2023), and waste household appliance recycling volume (7.69 million units in 2024). S25: Set external indicators: MSCI ESG rating environment dimension, grade C in 2020, upgraded to grade A in 2023; set supplementary strong rule indicators: no environmental violations or penalties from 2018 to 2024; after the integration of internal and external data, the comprehensive rating of the environment dimension is normal (0). 13. Social Dimension S26: Historical transaction indicators are set: donation expenditure (4.98 million yuan in 2024), government revenue (approximately 3.56 billion yuan in 2024); internal supplementary indicators are set: a cumulative total of 405 Hope Schools built, and employee volunteer service hours exceeding 50,000 hours. External indicators are set: tax rating of A for several consecutive years, and no administrative penalties or records of illegal or dishonest conduct. The overall rating is normal (0); IV. Calculation of Total Enterprise Ecosystem Value S27: Total Enterprise Ecosystem Value V-total = Σ(Vi × wi), where Vi represents the value contribution of each stakeholder, such as customer lifetime value (LTV), employee value creation, supplier synergy value, shareholder returns, creditor interest, positive environmental externalities, and social contribution value; wi represents the weights assigned based on industry characteristics and strategic importance; the value contribution Vi of each stakeholder is calculated as follows: Customer Value: Estimated using the Customer Lifetime Value (LTV) model, based on 2023 revenue of RMB 274.198 billion and customer volatility of 46.67%. Supplier Value: Using a supply chain synergy benefit model, synergy cost savings are estimated based on a supplier volatility reduction to 29.81% and accounts payable turnover days of 21.9 days. Shareholder value: Due to the lack of dividend data, net profit of RMB18.202 billion is temporarily used as a substitute indicator; Employee value: Employee income (total compensation) + excess output per employee. Since the total compensation is not disclosed, it is temporarily estimated indirectly based on R&D investment, etc. Creditor value: Interest expense is not included in the calculation due to missing data; Environmental value: Environmental protection investment of 550 million yuan, plus carbon emission reduction benefits (greenhouse gas emissions reduced from 927,000 tons in 2020 to 651,000 tons in 2023), calculated using the carbon trading price method; Social value: Donation expenditure of 4.98 million yuan + government revenue of 3.56 billion yuan + community benefits (405 schools, 50,000 hours of volunteer service) are estimated to be approximately 200 million yuan. S28: Based on industry characteristics (Shenwan Class IV Home Appliance Industry), the weights of each stakeholder are set as follows: Customers 0.30, Suppliers 0.20, Shareholders 0.15, Employees 0.10, Creditors 0.05, Environment 0.10, Society 0.10; V. Generate a Value Co-creation Report S29: The information extraction agent adopts a DeepSeek+Qwen3 dual-model collaborative architecture to perform entity recognition and relationship extraction on unstructured texts such as corporate ESG reports, social responsibility reports, and annual reports from 2020 to 2023, and extract key indicator data of various stakeholders. S30: Relationship Mapping and Knowledge Graph Construction. The intelligent agent is based on the Neo4j graph database to construct a dynamic knowledge graph of "enterprise-stakeholder-value stream". The nodes include Haier Smart Home, the top five customers, core suppliers, shareholders, employees, etc., and the relationships include sales, procurement, investment, employment, etc. S31: The value quantification and scenario simulation agent calls the valuation model in S27 to calculate the value of each stakeholder and the total value of the enterprise ecosystem; it supports natural language definition of scenarios, such as "if the supplier volatility drops to 25%", driving Monte Carlo simulation and outputting the impact on the total value of the ecosystem. S32: Report Generation and Insights. The intelligent agent has built-in report templates that automatically generate a value symbiosis report with a nine-grid heatmap, risk radar chart, and trend line chart. The report includes the following core content: Customer Perspective: Overall rating: Normal (0); Core Strengths: Accounts receivable turnover days of 22.6 days, better than the industry average; Customer volatility of 46.67% continues to decline; MSCI rating: A; Risk Warning: 45 complaints, after-sales service needs continuous optimization; Supplier Dimension: Overall rating is normal (0); Core strengths: Supplier volatility of 29.81% is industry-leading, accounts payable turnover days of 21.9 days is excellent payment efficiency, MSCI rating is A; Room for improvement: Specific quantitative thresholds for access assessment can be further disclosed; Shareholder perspective: Overall rating caution (1); Core issues: Dividend data not disclosed, reason for rating downgrade to CCC in 2010 not explained; Recommendation: Supplement disclosure of shareholder return information; Employee Dimension: Overall rating is normal (0); Core strength: Employee safety disputes remain at 0, S dimension rating is BBB; Risk warning: The compensation plan for layoffs from 2011 to 2015 was not disclosed and needs to be supplemented; Creditor Perspective: Comprehensive rating cannot be evaluated; Core Issue: Key indicators such as debt-to-equity ratio, current ratio, and interest expense are completely missing; Recommendation: Supplement disclosure of debt repayment capacity-related data; Environmental dimension: Overall rating is normal (0); Core strengths: Environmental protection investment has increased by an average of 20.3% annually, 7.69 million waste household appliances have been recycled, and there are no violations; Room for improvement: Greenhouse gas emission reduction measures can be quantified and disclosed.

[0035] Social dimension: Overall rating is normal (0); Core strengths: A total of 405 Hope Schools have been built, tax payment level is A, and there are no administrative penalties; Area for improvement: Long-term effectiveness evaluation data of public welfare projects can be supplemented and disclosed; S33: Visualized Output and Decision Support. Provides senior management with an ecosystem value trend map, risk heatmaps for seven key stakeholders, and a dashboard interface comparing collaborative solutions; provides the supply chain department with a supplier risk dashboard; and provides the human resources department with employee satisfaction and safety metrics dashboards.

[0036] In this invention, the intelligent agent evaluation method combines internal historical transaction data, stakeholder supplementary data, and publicly available external big data. Through the deep integration of large language models, knowledge graphs, multi-agent collaboration, and machine learning algorithms, it effectively evaluates the quality of the symbiotic relationship and the total ecological value between an enterprise and its seven core stakeholders. This method can automatically generate multi-dimensional value symbiosis reports, supports scenario simulation and natural language interaction, and provides comprehensive information support for enterprises to optimize customer cooperation, supply chain management, employee motivation, environmental governance, and social responsibility fulfillment.

[0037] 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 evaluating enterprise value symbiotic reporting intelligent agents based on a large-scale artificial intelligence model, characterized in that, The agent evaluation method specifically includes the following steps: S1. Multi-source data integration and preprocessing: Extract historical transaction data accumulated by the enterprise from the NC database, cooperate with other platforms to obtain and update external data, manually supplement relevant data with various stakeholders on a monthly basis, integrate historical transaction data, external data, and supplemented data to form a multi-theme data warehouse for stakeholder analysis, and write SQL scripts to obtain years of accounting subject data from the multi-theme data warehouse for historical transaction data modeling and obtain publicly available information of the enterprise for external data modeling. S2. Risk Rating and Value Quantification Indicator Construction: For customer, supplier, shareholder, employee, creditor, environmental, and social dimensions, set historical transaction indicators corresponding to historical transaction data, internal supplementary data indicators corresponding to supplementary data, and external indicators corresponding to external data to form a multi-dimensional, multi-stakeholder risk rating. S3. Actual Value Calculation and Assignment of Indicators: For the financial ratio indicators in the historical trading indicators corresponding to the historical trading data, use the pandas quantile function to calculate the quartiles of each indicator, divide the intervals and assign values. For positive indicators in the historical trading indicators corresponding to the historical trading data, the interval order is reversed. For the absolute value indicators in the historical trading indicators corresponding to the historical trading data and the internal supplementary indicators corresponding to the supplementary data, directly sort them, divide the indicator quartiles and assign values. For ratio-type supplementary indicators among the internal supplementary indicators corresponding to supplementary data and ratio-type external indicators among the external indicators corresponding to external data, calculate the actual values ​​of the indicators, sort them, determine the quartiles of the indicators, and assign values; for no-type indicators among the external indicators corresponding to external data, assign values ​​based on whether they are triggered. S4. Model Construction and Indicator Weight Generation: For historical transaction data, a model learning dataset is constructed using inference of qualified and unqualified lists. The model is trained using the random forest algorithm and optimized through grid search. The feature-importance method is used to obtain the weight vector of each indicator, and the overall risk score of the historical transaction data is calculated. For supplementary data, the indicator weights are directly set by expert knowledge and do not participate in machine learning modeling. The risk score of the supplementary data is calculated by multiplying the indicator risk score by the internal supplementary indicator weights. For external data, a combination of expert knowledge weighting and machine learning is used to set indicator weights. Risk assessment indicators are constructed and assigned values ​​using Python's pandas and numpy. The risk score of the external data is calculated by multiplying the indicator risk score by the external indicator weights. S5. Risk Score Calculation and Dimension Rating: For each stakeholder such as customers, suppliers, shareholders, employees, creditors, environment, and society, the internal risk score is calculated by multiplying the internal supplementary indicator weight by the indicator risk score, and the external risk score is calculated by multiplying the external indicator weight by the indicator risk score. The internal and external risk scores are sorted separately to obtain quartiles. The quartiles are then used to divide the intervals and assign values ​​to obtain the internal and external risk levels of each stakeholder, such as 0-normal, 1-vigilant, and 2-cautious risk levels. S6. Risk Level Integration and Comprehensive Rating and Enterprise Ecosystem Total Value Acquisition: For each stakeholder, their internal and external risk levels are integrated on the horizontal and vertical axes of a nine-square heatmap to generate a comprehensive rating. The total value of the enterprise ecosystem is calculated based on the value contribution of each stakeholder and the weights set according to industry characteristics and strategic importance. S7. Setting Strong Rules and Revising Comprehensive Rating: Based on the comprehensive rating fusion rules in step S6, the comprehensive rating is revised according to the strong rule indicators in the relevant supplementary data of each stakeholder: If any strong rule is triggered, the comprehensive rating of that stakeholder is directly adjusted to 2-Cautious; at the same time, when generating the value symbiosis report, the triggered strong rule matters are listed separately as key risk matters. S8. Collaborative Generation of Value Symbiosis Report: Employs a large language model to perform entity recognition, relation extraction, and event extraction on unstructured text, outputting structured information. Based on the Neo4j graph database, it constructs a dynamic knowledge graph of "enterprise-stakeholder-value stream," calls the enterprise ecosystem total value valuation model in step S6, calculates the value of each stakeholder and the total value of the enterprise ecosystem, and automatically generates a value symbiosis report with charts, attribution analysis, and strategic recommendations by using a built-in report template. S9. Visualized Output and Decision Support: Provides senior management with a dashboard interface that displays the overall value trend of the ecosystem, risk heatmaps, and comparisons of collaborative solutions; provides customized dashboards for business departments such as supply chain, human resources, sustainable development, finance, legal affairs, and risk management; and provides dedicated portals for stakeholders such as suppliers, employees, communities, customers, shareholders, and creditors to showcase collaborative value, individual contributions, and collaborative progress.

2. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 1, characterized in that, The specific operation of step S1 in the agent evaluation method is as follows: S1.

1. Use ETL tools to extract historical transaction data related to customers, suppliers, shareholders, employees, creditors, environment, and society from the NC database on a monthly basis. The historical transaction data includes the ending balances and amounts of the following eighteen categories of accounting items: accounts receivable, accounts payable, employee compensation payable, environmental protection investment, donation expenditure, advances from customers, main business revenue, other business revenue, procurement costs, paid-in capital, capital reserve, dividends payable, labor costs in R&D expenses, short-term loans, long-term loans, interest expenses in financial expenses, depreciation of environmental protection equipment, and taxes and surcharges. External data is obtained through cooperation with other platforms and updated monthly. S1.2 Manually supplement the data monthly with twelve categories related to various stakeholders, including litigation, non-compliance lists, internal performance data, environmental accidents, community complaints, contract disputes, equity pledge defaults, related-party transaction violations, labor disputes, debt defaults, internal environmental monitoring exceedance records, and internal reports of work safety accidents. Integrate historical transaction data, external data, and supplementary data to form a multi-theme data warehouse for stakeholder analysis. The multi-theme data warehouse is a theme-oriented, integrated, relatively stable data set that reflects historical changes, used to support symbiotic value analysis and risk modeling. S1.

3. Obtain the required multi-year accounting subject data from the multi-theme data warehouse for historical transaction data modeling by writing SQL scripts, and obtain publicly available enterprise information from the multi-theme data warehouse for external data modeling.

3. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 2, characterized in that, The other platforms in step S1.1 of the intelligent agent evaluation method are units that focus on collecting information about social entities, with Tianyancha, Qichacha, and Qixinbao being preferred.

4. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 2, characterized in that, The publicly available enterprise information in step S1.3 of the intelligent agent evaluation method includes the enterprise's business registration information, financial information, credit information, operating information, tax information, and judicial litigation information.

5. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 1, characterized in that, The specific operation of step S2 in the agent evaluation method is as follows: S2.

1. For the customer dimension, set historical transaction indicators corresponding to historical transaction data: accounts receivable turnover days, sales percentage of the top five customers, net profit, operating cash ratio, operating cash collection rate, and customer volatility; set internal supplementary data indicators corresponding to supplementary data: customer service system construction score, customer satisfaction survey results, and client litigation and complaint records; set external indicators corresponding to external data: ESG rating customer relationship specific score, customer service chapter text refinement and relationship evaluation. S2.

2. For the supplier dimension, set historical transaction indicators corresponding to historical transaction data: accounts payable turnover days, accounts payable as a percentage of total liabilities, prepayment turnover rate, supply chain efficiency, and supplier volatility; set internal supplementary data indicators corresponding to supplementary data: supplier access assessment standard score, supplier support plan implementation status, and supplier litigation records; set external indicators corresponding to external data: ESG rating supply chain specific score, supply chain chapter text refinement and relationship evaluation. S2.

3. For the shareholder dimension, set historical transaction indicators corresponding to historical transaction data: shareholder income, shareholding ratio of the top ten shareholders, shareholder nature, equity concentration, shareholding ratio of the largest shareholder, and details of related party transactions; Set up internal supplementary data indicators corresponding to the supplementary data: equity structure and actual controller information, implementation status of incentive and restraint mechanisms, relevant documents on shareholder rights protection and governance participation, and internal records of shareholder-related litigation or arbitration; set up external indicators corresponding to external data: annual scores / ratings of ESG governance dimension, and information on shareholder-related litigation, arbitration, or major penalties. S2.

4. For the employee dimension, set historical transaction indicators corresponding to historical transaction data: employee income, total executive compensation, average salary of ordinary employees, internal and external salary gap, employee training investment or welfare expenditure or severance pay amount; set internal supplementary data indicators corresponding to supplementary data: employee rights protection information, employee development and incentive details, employment and personnel adjustment records, internal records of labor disputes, employee care and employer branding texts; set external indicators corresponding to external data: ESG social responsibility dimension annual scores or rating levels, labor dispute related litigation or arbitration or major penalty information; S2.

5. For the creditor dimension, set historical transaction indicators corresponding to historical transaction data: creditor proceeds, debt-to-equity ratio, current ratio, quick ratio, interest coverage ratio, debt cost, interest-bearing debt structure, and cash and cash equivalents to short-term debt ratio; set internal supplementary data indicators corresponding to supplementary data: corporate entity or bond credit rating, details of bonds and debt repayment arrangements, creditor protection mechanisms, financial risk management system construction status, and internal records of debt contract disputes; set external indicators corresponding to external data: ESG governance dimension score or rating level, corporate entity or bond credit rating, information on litigation, arbitration, or major penalties related to debt contract disputes, and external evaluation of corporate debt repayment ability and credit status. S2.

6. For the environmental dimension, set historical transaction indicators corresponding to historical transaction data: environmental protection investment and environmental penalty amount; set internal supplementary data indicators corresponding to supplementary data: total greenhouse gas emissions, greenhouse gas emissions per unit of output value, total comprehensive energy consumption, comprehensive energy consumption per unit of output value, total water consumption, compliance disposal rate of hazardous waste, proportion of green products, renewable energy usage, internal records of environmental violations, and internal records of environmental litigation; set external indicators corresponding to external data: qualitative description of ESG reports, environmental administrative penalty decisions, environmental litigation judgments, records of violations of laws and regulations and litigation. S2.

7. For the social dimension, set historical transaction indicators corresponding to historical transaction data: donation expenditure and government revenue; set internal supplementary data indicators corresponding to supplementary data: cumulative number of public welfare projects, investment amount in rural revitalization, number of community volunteer service visits and duration, emergency public welfare assistance records, and non-financial indicator evaluation; set external indicators corresponding to external data: non-financial indicator evaluation, tax level, work safety accident reports, and negative reports from mainstream media.

6. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 1, characterized in that, The specific operation of step S3 in the agent evaluation method is as follows: S3.1 For the financial ratio indicators in the historical transaction indicators corresponding to historical transaction data, including accounts receivable turnover days, accounts payable turnover days, net profit, operating cash ratio, operating cash collection ratio, debt-to-equity ratio, current ratio, quick ratio, and interest coverage ratio, use the pandas quantile function to calculate the quartiles Q1 and Q3 of each indicator, dividing them into three intervals: [0, Q1], (Q1, Q3), and [Q3, +∞), corresponding to the three risk intervals of normal, vigilance, and caution, respectively. Assign three risk scores to the indicators: 0-normal, 1-vigilance, and 2-caution. For the positive indicators in the historical transaction indicators corresponding to historical transaction data, including net profit margin, operating cash ratio, accounts receivable turnover ratio, and current ratio, since the larger the value of the positive indicator, the lower the risk, the interval order is reversed. S3.2 For the absolute value indicators in the historical transaction indicators corresponding to the historical transaction data and the internal supplementary indicators corresponding to the supplementary data, including registered capital, environmental protection investment amount, donation expenditure, government revenue, and employee training investment amount, directly sort and define the quartiles of the indicators, and assign three risk scores of 0-normal, 1-warning, and 2-cautious. S3.

3. The ratio-type supplementary indicators in the internal supplementary indicators corresponding to the supplementary data and the ratio-type external indicators in the external indicators corresponding to the external data include the ratio of the guarantee amount to the registered capital, the ratio of the cumulative amount involved in the case to the registered capital, the greenhouse gas emissions per unit of output value, the ratio of total comprehensive energy consumption to output value, and the compliance disposal rate of hazardous waste. After calculating the actual values ​​of the indicators, they are sorted and quartiles are determined, and three risk scores of 0-normal, 1-vigilance, and 2-caution are assigned. S3.

4. For yes / no indicators: General yes / no indicators in external indicators corresponding to external data, including whether there are administrative penalties, whether there are environmental penalties, whether there are tax arrears, and whether there are lawsuits. Two risk scores, 0-not triggered and 1-triggered, are assigned based on whether they are triggered. The key risks among the external indicators corresponding to external data include whether there are serious violations of laws and regulations, whether there are cases of dishonesty being enforced, whether there are major tax violations, and whether there is a business abnormality list. Based on whether these are triggered, two risk scores are assigned: 0-not triggered and 2-triggered.

7. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 1, characterized in that, The specific operation of step S4 in the agent evaluation method is as follows: S4.

1. Based on historical transaction data, including risk score information for each indicator, label information for qualified and unqualified samples, multi-year time series data, and industry control variable information, construct a model learning dataset from qualified and unqualified samples, train the model using the random forest algorithm, optimize it through grid search, obtain the weight vector ω of each indicator using the feature-importance method, and calculate the overall risk score of the internal data. S4.2 For the supplementary data, the weights of each indicator are directly set using expert knowledge. Specifically, equal weights are assigned to general internal supplementary indicators, and double weights are assigned to key risk indicators such as litigation records, unqualified lists, and major internal performance violations. The sum of the weights of all internal supplementary indicators is 1. No machine learning model training is performed. After the risk scores of each indicator are sorted, the quartiles of the indicators are determined and assigned values. The internal risk score is calculated by multiplying the internal supplementary indicator weight by the indicator risk score. S4.3 For external data, including business registration information, financial information, credit information, operating information, tax information, judicial litigation information, ESG reports and ratings, environmental penalties, and public opinion information, a combination of expert knowledge empowerment and machine learning is adopted. Indicator weights are set and the sum of the weights of all external indicators is 1. Risk assessment indicators are constructed and assigned values ​​using Python's pandas and numpy. The risk score of external data is calculated by summing the risk scores of each indicator of stakeholders multiplied by the indicator weights.

8. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 7, characterized in that, The specific operation of step S4.1 in the agent evaluation method is as follows: S4.1.1 Sample of qualified and unqualified reasoning lists: Unqualified list: For customers, those with outstanding accounts receivable for two consecutive years will be added to the unqualified list; for suppliers, those with outstanding accounts payable for two consecutive years without any transactions will be added to the unqualified list; for employees, those with unsatisfactory performance appraisals for two consecutive years and who have committed serious misconduct will be added to the unqualified list. Qualified list: For customers, customers whose sum of the risk scores of the four indicators of accounts receivable turnover rate and collection rate in the previous year and the previous two years is less than or equal to 2 are included in the qualified list; for suppliers, suppliers whose comprehensive score of indicators such as accounts payable turnover rate and on-time delivery rate is less than or equal to 2 is included in the qualified list; for employees, suppliers whose comprehensive score of training investment, promotion rate, satisfaction rate, etc. is less than or equal to 2 is included in the qualified list. S4.1.2 Constructing the model learning dataset: Add the risk score data of the corresponding indicators for the previous three and four years of the unqualified list and qualified list samples to the industry control variables, and label them as unqualified list and qualified list, such as unqualified list=1, qualified list=0, to obtain the model learning data; S4.1.

3. The random forest algorithm is used to train the model, and the grid search is used for optimization. The weight vector ω of each index is obtained using the feature-importance method. S4.1.4 Calculate the overall risk score of historical transaction data: Using the previous year as a benchmark, take the risk scores of the four core indicators of the sample, add industry control variables, and form an indicator matrix A. The risk score S = A × ω T ; Based on the scores, quartiles are calculated, and three risk levels are divided: 0-normal, 1-alert, and 2-cautious.

9. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 7, characterized in that, The specific operation of step S4.2 in the agent evaluation method is as follows: S4.2.1 The basic indicators are set to equal weight, but for the five key risk indicators, namely the cumulative amount involved in cases as a percentage of registered capital, the amount of environmental penalties, the amount involved in major litigation, the amount of environmental penalties, and the amount of tax arrears, the weight is set to twice that of the other three general indicators, such as registered capital, environmental protection investment, and donation expenditure. The sum of the weights of all external indicators is 1. S4.2.2 Use Python's pandas and numpy to construct risk assessment indicators. Assign three risk levels (0-normal, 1-alert, 2-cautious) to absolute value indicators and ratio indicators based on quartiles. Assign two risk levels (0-no, 1-yes) or two risk levels (0-no, 2-yes) to yes / no indicators based on whether they are general or important. S4.2.

3. The external data risk score is obtained by summing the risk scores of each stakeholder's indicators by multiplying the indicator weights. Thresholds are set based on the distribution of score data clusters to divide the data into three levels: normal, alert, and cautious, and assigned three risk levels: 0-normal, 1-alert, and 2-cautious, respectively.

10. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 1, characterized in that, The specific operation of step S6 in the agent evaluation method is as follows: S6.1 For each stakeholder, its internal risk level and external risk level are respectively placed on the horizontal and vertical axes of a nine-square grid heatmap. The horizontal axis represents internal risk levels from left to right: Normal, Cautious, and Safe; the vertical axis represents external risk levels from bottom to top: Normal, Cautious, and Safe. The comprehensive rating integration rules are as follows: The overall rating is normal: the combination is (normal, normal), (normal, caution), (caution, normal). The overall rating is "Caution": the combination is (Caution, Caution), (Normal, Cautious), (Cautious, Normal). The overall rating is cautious: the combination is (cautious, cautious), (cautious, cautious), (cautious, cautious); S6.2, Total Enterprise Ecosystem Value V-total = Σ(Vi × wi), where Vi represents the value contribution of each stakeholder, such as customer lifetime value (LTV), employee value creation, supplier synergy value, shareholder returns, creditor interest, positive environmental externalities, and social contribution value; wi represents the weights assigned based on industry characteristics and strategic importance; Vi is calculated using the following model: Customer Lifetime Value: The Customer Lifetime Value model LTV = Σ(M×R / (1+i)^t) is used to sum up all customers; Supplier Collaboration Value: Employing a supply chain collaboration benefit model, we quantify the cost savings and revenue growth resulting from on-time supplier delivery and joint R&D. Employee value creation: Using a human capital value model, employee income plus excess output per employee; Shareholder income: The amount of income earned by shareholders; Creditor interest: Interest expense; Environmental positive externality discounting: potential environmental cost savings resulting from environmental protection investments; Social contribution value: donation expenditure + government revenue + community benefit depreciation.

11. The method for evaluating enterprise value symbiotic reporting intelligent agents based on a large-scale artificial intelligence model according to claim 1 or 10, characterized in that, The specific operation of step S8 in the agent evaluation method is as follows: S8.1 Information Extraction Intelligent Agent: Using a large language model, it performs entity recognition, relation extraction, and event extraction on unstructured texts such as ESG reports, litigation documents, public opinion comments, annual reports, social responsibility reports, environmental reports, and board reports, and outputs structured information; S8.2 Relationship Mapping and Knowledge Graph Construction of Intelligent Agents: Based on the Neo4j graph database, a dynamic knowledge graph of "enterprise-stakeholder-value stream" is constructed. Nodes include enterprises, customers, suppliers, shareholders, employees, creditors, environmental indicators, social events, products, projects, technologies, and events. Relationships include employment, service, cooperation, investment, lending, emissions, donation, procurement, sales, governance, and litigation. The centrality, community structure, and value flow path of each stakeholder node in the knowledge graph are calculated periodically. S8.3 Value Quantification and Scenario Simulation Intelligent Agent: Call the enterprise ecosystem total value valuation model in step S6.2 to calculate the value of each stakeholder and the total value of the enterprise ecosystem; It supports natural language-defined scenarios, drives Monte Carlo simulations, and outputs the impact on cash flow, discount rate, and overall valuation. S8.4 Report Generation and Insight Agent: Built-in report templates automatically generate value symbiotic reports with charts, attribution analysis and strategic recommendations, and support natural language question-and-answer interaction.

12. The enterprise value symbiotic reporting intelligent agent evaluation method based on an artificial intelligence large model according to claim 11, characterized in that, The large language model in step S8.1 of the agent evaluation method is a DeepSeek+Qwen3 dual-model collaborative architecture.

Images