ESG performance dynamic evaluation method based on configuration governance and large model

Abstract

The application provides an ESG performance dynamic evaluation method based on configuration governance and a large model, applied to the field of sustainable development management and information technology, and comprises the following steps: constructing a three-level ESG configuration governance index system of "dimension-element-configuration", establishing an element correlation rule and a weight dynamic allocation mechanism and an ESG configuration library; building a multi-source data acquisition and processing network to form a structured ESG data pool; taking an open source large language model as a base to construct an ESG special large model with professional ability; combining the heterogeneity characteristics of enterprises and the industry characteristics to generate personalized and industrialized basic configuration baseline and establish a double auditing and dynamic updating mechanism; performing dynamic evaluation by the ESG special large model fusion configuration governance framework; and outputting personalized evaluation results based on the release baseline and user demand. Through the application, real-time, accurate and in-depth evaluation of ESG performance can be realized.

Description

Technical Field

[0001] This invention relates to the field of sustainable development management and information technology, and in particular to a dynamic evaluation method for ESG performance based on configurational governance and large models. Background Technology

[0002] With the widespread adoption of sustainable development concepts, ESG (Environmental, Social, and Governance) performance has become a core indicator for measuring a company's long-term value, the rationality of its investment decisions, and its social responsibility. Good ESG performance signals a company's strong risk management capabilities. The new generation of consumers and customers increasingly prefer brands with a strong sense of social and environmental responsibility. A company's ESG reputation directly impacts its market share and brand value. Currently, mainstream ESG assessment systems generally revolve around the three pillars of environment, society, and governance, based on international standards such as GRI and SASB. They rely on manually collected data from publicly reported corporate reports, constructing fixed indicator systems for information disclosure, and are showing a trend towards moving from voluntary disclosure to mandatory standardization, and from single scores to multi-dimensional perspectives.

[0003] Existing ESG assessment methods and systems mainly cover four core aspects: indicator system construction, data collection and integration, indicator quantification and calculation, and visualization report generation. However, existing ESG assessment methods generally rely on corporate disclosures and static indicator systems, making it difficult to meet the needs of continuous supervision, multi-source data fusion, auditability, and dynamic assessment. This is mainly reflected in the following aspects: Data fragmentation and static nature, low quality of disclosed data, and serious data gaps. ESG data is scattered across multiple departments, including environment, production, human resources, supply chain, and legal, with inconsistent formats, creating data silos. Current ESG assessments rely heavily on annual or quarterly reports, presenting isolated "static snapshots" that cannot effectively capture and reflect the continuous dynamic evolution of ESG performance over time. Corporate disclosures are incomplete and inconsistent, with a large amount of "self-reported data" lacking third-party verification, which can easily lead to subjective descriptions and omissions.

[0004] Lack of change management and traceability When a company adjusts its operational boundaries or external reporting standards are updated, the scope and methodology for calculating ESG data will change accordingly. Existing ESG assessment methods lack effective version control, change management, and auditing mechanisms, leading to distorted comparisons of historical data, unclear performance evolution paths, and opaque assessment processes. External investors, rating agencies, and regulators find it difficult to verify the data sources, calculation logic, and consistency with historical data, severely impacting the credibility of assessment results.

[0005] With increasingly stringent ESG regulatory policies and diversified market demands, traditional assessment methods can no longer meet the needs for dynamic, auditable, and systematic assessments. Therefore, developing a method that integrates constructive governance principles with advanced AI technology to capture the correlations between ESG elements, integrate multi-source dynamic data, and achieve personalized assessments has become an urgent need in the field of ESG assessment. Summary of the Invention

[0006] This invention provides a dynamic evaluation method for ESG performance based on configurational governance and large models, which addresses the shortcomings of existing ESG evaluation methods, such as insufficient dynamism, poor adaptability, low governance correlation, lack of unified governance and traceability mechanisms, and inability to meet the requirements of continuous supervision, multi-source data fusion and auditability, thereby achieving real-time, accurate and in-depth evaluation of ESG performance.

[0007] This invention provides a dynamic evaluation method for ESG performance based on configurational governance and large models, comprising the following steps.

[0008] Construct an ESG governance indicator system, which includes: breaking down core elements from three dimensions: environment, society, and governance, and establishing a three-level indicator system of dimensions, elements, and configuration; Based on the ESG configuration governance index system, multi-source ESG data is collected and preprocessed to form a structured ESG data pool. Based on an open-source large language model, an ESG-specific large model is constructed by fine-tuning the ESG domain corpus and instructions in the ESG data pool. The ESG-specific large model has the ability to interpret ESG text, perform data correlation analysis, identify risks, and generate assessment reports. Based on the ESG configuration governance indicator system, the ESG data pool, and the ESG-specific large model, the basic configuration baseline for enterprise ESG assessment and the basic configuration baseline for the industry are generated. Based on the aforementioned ESG-specific large model and configuration governance framework, the system calls upon the enterprise's basic ESG assessment configuration baseline and the industry's basic configuration baseline to generate a phased assessment baseline. It also automatically creates assessment tasks for related configuration items and assigns execution units. The system performs element matching, correlation analysis, risk level determination, and performance scoring on the preprocessed data to achieve ESG assessment and obtain the published baseline. Based on the aforementioned release baseline and user needs, the evaluation results are output in the form of real-time dashboards, dynamic scoring reports, element association maps, risk warning lists, and optimization suggestions.

[0009] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and large model is provided. The ESG configurational governance indicator system includes a primary dimension, secondary elements and a tertiary configuration, wherein the secondary elements are refined based on the primary dimension, and the tertiary configuration is a quantifiable indicator of the secondary elements. The method further includes: Construct an element association rule model, wherein the element association rule model is used to quantify the synergistic effects and constraints among the secondary elements; Based on the target company's industry type, company size, and development stage, set the weight adjustment coefficients for the ESG governance indicator system. Based on the element association rule model and the weight adjustment coefficient, an ESG configuration library is constructed. The ESG configuration library includes: the first-level dimension, the second-level element, the third-level configuration, and the four-level mapping relationship between association rules.

[0010] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and a large model is provided. The method involves collecting multi-source ESG data based on the ESG configurational governance indicator system and preprocessing the multi-source ESG data to form a structured ESG data pool, including: A microservice architecture is used to build a flexible and scalable data network, wherein the different layers of the data network communicate with each other through standardized API interfaces; Data is collected through preset multi-source data collection nodes to obtain publicly available automated data, privately authorized access data, and supplementary customized data. The collected publicly available automated data, privately authorized access data, and supplementary customized data are cleaned, transformed, and fused for verification to obtain standardized ESG data, forming a structured ESG data pool.

[0011] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and a large model is provided. The method uses an open-source large language model as a foundation and constructs a dedicated ESG large model through fine-tuning of ESG domain corpora and instructions in the ESG data pool. The method includes: Select a general pre-trained large model with language understanding capabilities as the base model, and determine the parameter scale, architecture type, and input / output format of the base model; An ESG domain training dataset is constructed based on the ESG data pool, and the ESG domain training dataset is subjected to secondary semantic cleaning and assigned standardized domain labels. Based on the ESG domain training dataset, the base model is fine-tuned in three stages: domain knowledge injection, task capability enhancement, and human feedback optimization, to obtain the initial ESG-specific large model. An ESG-specific evaluation set is constructed to conduct multi-dimensional tests on the initial ESG-specific large model. After continuous optimization based on the test results, a well-trained ESG-specific large model is obtained. Deploy the trained ESG-specific large model using API, microservices, or localization.

[0012] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and a large-scale model is provided. The method generates a basic configurational baseline for enterprise ESG evaluation and an industry-specific basic configurational baseline based on the ESG configurational governance indicator system, the ESG data pool, and the ESG-specific large-scale model, including: Based on the ESG governance index system, the heterogeneity characteristics of the target enterprise are collected, and enterprise feature vectors are formed through feature normalization, classification coding, and semantic modeling. Based on the enterprise feature vector, the ESG configuration library, and the ESG-specific large model, matching configuration elements and configuration items are selected to generate a basic configuration baseline for enterprise ESG assessment that includes indicator definitions, calculation rules, evidence requirements, and evaluation logic. Based on the enterprise feature vectors, the ESG configuration library, and the ESG-specific large model, industry corpus is analyzed to screen industry-adaptive configuration items, and industry-related data is combined to form an industry benchmark value system and generate industry basic configuration baselines.

[0013] According to the present invention, a dynamic ESG performance evaluation method based on configurational governance and a large model is provided. Based on the ESG-specific large model and configurational governance framework, it calls upon the enterprise's basic ESG evaluation configuration baseline and the industry's basic configuration baseline to generate a phased evaluation baseline. It automatically creates evaluation tasks and assigns execution units to related configuration items. The method performs element matching, correlation analysis, risk level determination, and performance scoring on the preprocessed data to achieve ESG evaluation and obtain a published baseline, including: Based on the enterprise ESG assessment baseline and industry baseline, a phased assessment baseline is formed, and a baseline snapshot is taken of the phased assessment baseline to form an initial version record. The assessment tasks corresponding to the phased assessment baseline are broken down and assigned to the execution units using the ESG-specific large model. The evaluation results completed by the execution unit are reviewed, and the review records are incorporated into the configuration documentation system; The evaluation results after review are subject to multi-dimensional cross-review, and a release baseline is generated after the review is passed. Based on the aforementioned release baseline, update the baseline version chain.

[0014] This invention provides a dynamic ESG performance evaluation method based on configurational governance and a large model. It constructs a three-tiered ESG configurational governance indicator system ("dimension-element-configuration"), establishes element association rules and a dynamic weight allocation mechanism, and an ESG configuration library. It builds a multi-source data collection and processing network, forming a structured ESG data pool through "intelligent cleaning + manual review." Based on an open-source large language model, it constructs a professional ESG-specific large model through fine-tuning with ESG domain corpora and instructions. Combining enterprise heterogeneity and industry characteristics, it generates personalized and industry-specific basic configuration baselines and establishes a dual review and dynamic update mechanism. Finally, it performs dynamic evaluation by integrating the configurational governance framework with the ESG-specific large model, achieving full traceability of the evaluation process. Based on the published baselines and user needs, it outputs personalized evaluation results in various forms, such as real-time dashboards and dynamic scoring reports. This effectively overcomes the limitations of traditional evaluation methods—static, singular, and lacking specificity—significantly improving the traceability, verifiability, and multi-scenario adaptability of evaluation results. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is an overall schematic diagram of the dynamic evaluation method for ESG performance based on configurational governance and large models provided by this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. 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.

[0018] This invention describes a dynamic evaluation method for ESG performance based on configurational governance and a large-scale model. Using configurational governance theory as a framework and a large-scale model as support, it constructs an ESG configurational governance indicator system, establishes a multi-source data acquisition and processing network with a dual mechanism of "intelligent cleaning + manual review," trains a large-scale model in the ESG vertical domain as an intelligent ESG data processing and analysis engine, and achieves dynamic evaluation of ESG performance and data quality control and traceability based on the ESG-specific large-scale model integrated with the configurational governance framework. Finally, it outputs evaluation results and optimization suggestions through multi-dimensional correlation analysis.

[0019] Figure 1 This is an overall schematic diagram of the dynamic evaluation method for ESG performance based on configurational governance and large models provided by this invention.

[0020] The present invention provides a dynamic evaluation method for ESG performance based on configurational governance and large-scale models, comprising the following steps: Construct an ESG governance indicator system, which includes: breaking down core elements from three dimensions: environment, society, and governance, and establishing a three-level indicator system of dimensions, elements, and configuration; Based on the ESG configuration governance indicator system, multi-source ESG data is collected and preprocessed to form a structured ESG data pool. Based on an open-source large language model, and through fine-tuning of ESG domain corpus and instructions in the ESG data pool, a dedicated ESG large model is constructed. The dedicated ESG large model has the ability to interpret ESG text, perform data correlation analysis, identify risks, and generate assessment reports. Based on the ESG configuration governance indicator system, ESG data pool and ESG-specific large model, generate basic configuration baselines for enterprise ESG assessment and basic configuration baselines for the industry. Based on the ESG-specific large model and configuration governance framework, the system calls the enterprise's basic ESG assessment configuration baseline and the industry's basic configuration baseline to generate a phased assessment baseline. It also automatically creates assessment tasks and assigns execution units to related configuration items. The system performs element matching, correlation analysis, risk level determination and performance score calculation on the preprocessed data to achieve ESG assessment and obtain the published baseline. Based on the release baseline and user needs, the evaluation results are output in the form of real-time dashboards, dynamic scoring reports, element relationship maps, risk warning lists, and optimization suggestions.

[0021] S1. Construct an ESG configurational governance indicator system. Based on configurational governance theory, the core elements are broken down into three dimensions: environment (E), society (S), and governance (G), and a three-level indicator system of "dimension-element-configuration" is established, along with the correlation rules and dynamic weight allocation mechanism among the elements. S2 establishes an ESG data collection and processing network. Through multi-source data collection and a dual mechanism of "intelligent cleaning + manual review," it integrates internal enterprise data, public data, third-party monitoring data, public opinion data, and customized survey data. Through data cleaning, format standardization, and deduplication and noise reduction, it forms a structured ESG data pool. S3 builds a large-scale ESG-specific model. Based on an open-source large language model, it is fine-tuned through ESG domain corpus and instruction fine-tuning, and has the ability to interpret ESG texts, perform data correlation analysis, identify risks, and generate assessment reports. S4. Establish personalized and industry-specific benchmarks. Based on a dedicated ESG model, combined with the heterogeneous characteristics of enterprises, the ESG governance indicator system, and the ESG data pool, and after dual manual review by ESG experts and enterprise management, a basic configuration baseline for enterprise ESG assessment and an industry-specific basic configuration baseline are formed.

[0022] S5 performs dynamic ESG assessments. It integrates a dedicated ESG model with a configurational governance framework, calls upon the basic configuration baseline to generate phased assessment baselines, automatically creates assessment tasks for related configuration items and assigns execution units, performs element matching, correlation analysis, risk level determination, and performance scoring calculation on preprocessed data, and combines business needs with configurational control theory to achieve traceability and change analysis capabilities for ESG assessment data, processes, and change trajectories. S6 generates personalized assessment results. Based on the published baseline and user needs, the assessment results are output in formats including real-time dashboards, dynamic scoring reports, element relationship maps, risk warning lists, and optimization suggestions, enabling real-time updates and multi-scenario applications of the assessment results.

[0023] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and large model is provided. The ESG configurational governance indicator system includes a primary dimension, secondary elements and a tertiary configuration, wherein the secondary elements are refined based on the primary dimension, and the tertiary configuration is a quantifiable indicator of the secondary elements. The method further includes: Construct an element association rule model, wherein the element association rule model is used to quantify the synergistic effects and constraints among the secondary elements; Based on the target company's industry type, company size, and development stage, set the weight adjustment coefficients for the ESG governance indicator system. Based on the element association rule model and the weight adjustment coefficient, an ESG configuration library is constructed. The ESG configuration library includes: the first-level dimension, the second-level element, the third-level configuration, and the four-level mapping relationship between association rules.

[0024] Furthermore, the method for constructing the ESG configuration governance indicator system described in step S1 includes the following steps: S11 clarifies the hierarchical structure of the indicator system. The first-level dimensions include Environment (E), Society (S), and Governance (G). Second-level elements are detailed based on the first-level dimensions; under the E dimension, these include carbon emission management and resource utilization efficiency; under the S dimension, these include employee rights protection and supply chain responsibility; and under the G dimension, these include corporate governance structure and internal control. The third-level configurations are the specific implementation indicators of the second-level elements. These indicators are configuration items within the configuration governance framework. For example, under carbon emission management, quantifiable indicators such as carbon emission intensity and carbon emission reduction target completion rate are further set. S12, construct an element association rule model, quantitatively analyze the synergistic effects (such as the positive correlation between green energy input and carbon emission reduction effectiveness) and constraint relationships (such as the short-term negative correlation between environmental protection input and short-term profitability) among elements, and form a dynamic association matrix; S13, Establish a dynamic weight allocation mechanism. Set weight adjustment coefficients based on key characteristics of the target entity, such as industry type, company size, and development stage. S14. Establish an ESG configuration library. Based on the principles of "full coverage, strong correlation, and traceability," deeply integrate mainstream domestic and international ESG standards and policy requirements, establish a four-level mapping relationship of "dimension-configuration element-configuration item-association rule," and build a structured data hub based on this relationship to establish an ESG configuration library; Through the embodiments of the present invention, by constructing an ESG configuration library containing four-level mapping relationships, and combining an element association rule model and a dynamic weight adjustment mechanism, accurate and dynamic evaluation of the ESG performance of enterprises in different industries, sizes and development stages is achieved, thereby improving the scientificity, adaptability and quantifiability of the evaluation.

[0025] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and a large model is provided. The method involves collecting multi-source ESG data based on the ESG configurational governance indicator system and preprocessing the multi-source ESG data to form a structured ESG data pool, including: A microservice architecture is used to build a flexible and scalable data network, wherein the different layers of the data network communicate with each other through standardized API interfaces; Data is collected through preset multi-source data collection nodes to obtain publicly available automated data, privately authorized access data, and supplementary customized data. The collected publicly available automated data, privately authorized access data, and supplementary customized data are cleaned, transformed, and fused for verification to obtain standardized ESG data, forming a structured ESG data pool.

[0026] Furthermore, the construction of the ESG data acquisition and processing network described in step S2 includes the following steps: S21, Build a data network architecture. Adopt a microservice architecture to construct a flexible and scalable data network. Each layer achieves data interoperability through standardized API interfaces, supporting on-demand expansion and feature iteration. S22, deploy multi-source data collection nodes. Following the logic of "automatic collection of public data + authorized access to private data + customized collection of supplementary data," deploy three types of collection nodes to achieve comprehensive coverage of data sources; S23, Construction of a full-process data processing system. Establish a full-process data processing mechanism of "cleaning-transformation-fusion-verification" to transform raw data into standardized and usable ESG data assets, forming an ESG data resource pool; Through the embodiments of the present invention, an elastic data network is built using a microservice architecture and standardized APIs to efficiently integrate multi-source heterogeneous ESG data. After cleaning, transformation, and fusion verification, a structured ESG data pool is formed, which significantly improves the coverage, consistency, and processing efficiency of data collection, and provides a high-quality and scalable data foundation for subsequent dynamic evaluation.

[0027] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and a large model is provided. The method uses an open-source large language model as a foundation and constructs a dedicated ESG large model through fine-tuning of ESG domain corpora and instructions in the ESG data pool. The method includes: Select a general pre-trained large model with language understanding capabilities as the base model, and determine the parameter scale, architecture type, and input / output format of the base model; An ESG domain training dataset is constructed based on the ESG data pool, and the ESG domain training dataset is subjected to secondary semantic cleaning and assigned standardized domain labels. Based on the ESG domain training dataset, the base model is fine-tuned in three stages: domain knowledge injection, task capability enhancement, and human feedback optimization, to obtain the initial ESG-specific large model. An ESG-specific evaluation set is constructed to conduct multi-dimensional tests on the initial ESG-specific large model. After continuous optimization based on the test results, a well-trained ESG-specific large model is obtained. Deploy the trained ESG-specific large model using API, microservices, or localization.

[0028] Furthermore, the construction of the ESG-specific large model described in step S3 includes the following steps: S31, Base Model Selection and Initialization. Select a general-purpose pre-trained large model with strong language understanding capabilities as the base model, and determine its basic capabilities such as parameter size, architecture type, and input / output format to ensure that it has the potential to support complex semantic reasoning and cross-modal information processing in the ESG field; S32, Construction of ESG Domain Training Dataset. Based on the ESG resource pool constructed in step S2, the corpus undergoes secondary semantic cleaning, including noise reduction, format unification, label supplementation, and syntactic correction. According to the ESG standard system and the configuration library structure constructed in step S1, standardized domain labels are assigned to the corpus for use in the design of supervision signals in subsequent training phases. S33, Multi-stage model fine-tuning. A three-stage fine-tuning strategy of "domain knowledge injection - task capability enhancement - human feedback optimization" is adopted to gradually improve the model's ESG professional capabilities. Based on the four-level mapping structure of "dimension - configuration element - configuration item - association rule" in the ESG configuration library, the model's understanding of structured configuration knowledge and its ability to make associative inferences are strengthened. S34, Model Testing, Optimization, and Deployment. Build a dedicated ESG evaluation suite to perform multi-dimensional testing of the model, including accuracy, robustness, interpretability, and coverage. Perform continuous optimization based on the test results, and finally deploy the model to the ESG assessment system via API, microservices, or localization.

[0029] Through the embodiments of the present invention, a highly adaptable ESG-specific large model is constructed through three-stage fine-tuning and multi-dimensional evaluation. This significantly improves the model's semantic understanding, task execution, and reasoning capabilities in the ESG field, and supports flexible deployment, providing intelligent, professional, and practical technical support for dynamic evaluation of ESG performance.

[0030] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and a large-scale model is provided. The method generates a basic configurational baseline for enterprise ESG evaluation and an industry-specific basic configurational baseline based on the ESG configurational governance indicator system, the ESG data pool, and the ESG-specific large-scale model, including: Based on the ESG governance index system, the heterogeneity characteristics of the target enterprise are collected, and enterprise feature vectors are formed through feature normalization, classification coding, and semantic modeling. Based on the enterprise feature vector, the ESG configuration library, and the ESG-specific large model, matching configuration elements and configuration items are selected to generate a basic configuration baseline for enterprise ESG assessment that includes indicator definitions, calculation rules, evidence requirements, and evaluation logic. Based on the enterprise feature vectors, the ESG configuration library, and the ESG-specific large model, industry corpus is analyzed to screen industry-adaptive configuration items, and industry-related data is combined to form an industry benchmark value system and generate industry basic configuration baselines.

[0031] Furthermore, the establishment of personalized and industry-specific benchmarks in step S4 includes the following steps: S41 collects heterogeneous features such as basic information, industry attributes, size type, organizational governance structure, and development stage of enterprises, and forms enterprise feature vectors through feature normalization, classification coding, semantic modeling and other methods; S42, based on the enterprise feature vector, ESG configuration library and the ESG-specific large model constructed in step S3, select the configuration elements and configuration items that best match the enterprise characteristics from the ESG configuration library, and generate the enterprise ESG assessment basic configuration baseline that includes indicator definitions, calculation rules, evidence requirements, evaluation logic and other contents according to the enterprise basic configuration item set. S43 identifies industry-specific disclosure standards, risk exposure characteristics, and operating models. It analyzes industry corpus through a dedicated ESG model, selects industry-appropriate configuration items from the ESG configuration library, and combines industry reports, industry statistics, quantitative indicators released by authoritative institutions, or industry average levels to form an industry benchmark system. It combines industry configuration items with industry benchmarks to generate industry basic configuration baselines, providing reference standards for enterprise benchmarking analysis. S44 establishes a dual review and dynamic update mechanism involving experts and managers. Through dynamic calibration of "data feedback - policy tracking - expert review", it ensures that the benchmark is updated in sync with the company's current situation, industry practices, regulatory requirements, and technological development. At the same time, through dual review by ESG experts and managers, it ensures that the indicators are reasonable, the configuration is complete, the baseline is scientific, compliant, and executable.

[0032] Through the embodiments of the present invention, by integrating the heterogeneous characteristics of enterprises with industry corpora, and combining the ESG configuration library and dedicated large models, personalized enterprise ESG assessment basic configuration baselines and industry benchmark industry basic configuration baselines are automatically generated, achieving accurate adaptation and dynamic benchmarking of assessment standards, and improving the pertinence, comparability and intelligence of ESG performance assessment.

[0033] According to the present invention, a dynamic ESG performance evaluation method based on configurational governance and a large model is provided. Based on a dedicated ESG large model and configurational governance framework, it calls upon the enterprise's basic ESG assessment configuration baseline and the industry's basic configuration baseline to generate a phased assessment baseline. It automatically creates assessment tasks and assigns execution units to related configuration items. The method performs element matching, correlation analysis, risk level determination, and performance scoring on the preprocessed data to achieve ESG assessment and obtain a published baseline, including: Based on the enterprise ESG assessment baseline and industry baseline, a phased assessment baseline is formed, and a baseline snapshot is taken of the phased assessment baseline to form an initial version record. The assessment tasks corresponding to the phased assessment baseline are broken down and assigned to the execution units using the ESG-specific large model. The evaluation results completed by the execution unit are reviewed, and the review records are incorporated into the configuration documentation system; The evaluation results after review are subject to multi-dimensional cross-review, and a release baseline is generated after the review is passed. Based on the aforementioned release baseline, update the baseline version chain.

[0034] Furthermore, the ESG dynamic assessment described in step S5 includes the following steps: S51, From Basic Configuration to Phase Assessment Baseline Generation. Using the personalized, industry-specific benchmarks established in step S4 as the core benchmark blueprint for the assessment work, and based on the specific assessment needs of the enterprise at each stage, targeted selection and combination are performed on the basic configuration baseline to generate phase assessment baselines. After the phase assessment baseline is generated, a baseline snapshot mechanism is automatically triggered to record the initial state of the baseline for that stage (including core information such as configuration item composition, indicator weights, scoring rules, and data collection sources), forming an initial version record, providing a basis for subsequent change tracking and version comparison; S52, Automatic Creation and Execution Unit Allocation of Assessment Tasks. Based on the generated phased assessment baseline, the task scheduling capabilities of the ESG-specific large model built in step S3 are used to automatically decompose assessment tasks. Each task must clearly define core elements including task name, corresponding configuration items, assessment criteria, data requirements, completion deadline, and quality requirements. The model combines the enterprise's organizational structure, the responsibilities and permissions of each person, and historical assessment execution efficiency data to accurately allocate the decomposed assessment tasks to the corresponding execution units. A task management dashboard is built to track the execution status of each assessment task in real time. S53, Task Implementation and Preliminary Result Review. The task execution unit carries out its work according to the assessment task requirements, completing data collection, indicator evaluation, and result reporting, including specific indicator values, scoring results, completion status descriptions, and relevant supporting materials. Simultaneously, an internal review process is executed, comparing the assessment results against the requirements of the interim assessment baseline to verify the accuracy of the assessment results and the completeness and validity of the supporting materials, laying the foundation for subsequent review stages. If any issues are found, the task is returned to the task execution unit for correction, and then resubmitted for review until it meets the requirements. Internal review records are simultaneously incorporated into the configuration documentation system. S54, Multi-dimensional Review and Phase Baseline Release. Relying on the configuration review mechanism, the review team conducts multi-dimensional cross-reviews on the consistency between the assessment results and the phase baseline requirements, the compliance and credibility of the data sources, and the authenticity and sufficiency of the supporting materials. After the review is passed, the final status of the phase assessment baseline is confirmed, and the phase baseline release and phase transition process is completed. At the same time, based on the configuration record and configuration review record, an ESG assessment version chain is automatically constructed. S55, Change Iteration, Phase Baseline Adjustment and Release Baseline Generation. Based on changes in business requirements and configuration control theory, the released phase baselines are adaptively adjusted, changed, and reviewed. The change process must strictly follow the configuration control process, and all adjustment actions are recorded in the configuration documentation system to ensure the traceability of the change trajectory. After the phase baseline is adaptively adjusted according to the change plan, the adjustment review process is initiated. After the review is approved, an enterprise ESG release baseline at a specific time point is generated, the version chain is updated synchronously, and a snapshot of the baseline after this change is recorded to achieve closed-loop management of the entire chain of "baseline - task - execution - result - change".

[0035] Through the embodiments of the present invention, this method intelligently decomposes evaluation tasks using a large model, automatically allocates execution units, and combines baseline snapshots, review, and multi-dimensional cross-audit mechanisms to achieve traceability of the evaluation process, reliability of results, and controllability of versions, significantly improving the automation, standardization, and dynamic updating capabilities of ESG performance evaluation.

[0036] According to the present invention, a dynamic evaluation method for ESG performance based on configurational governance and large-scale models is provided, wherein the evaluation results are output based on the release baseline and user requirements, including: The assessment data of each of the aforementioned phased assessment baselines are integrated, and the data integration, verification and calibration are completed by combining the configuration documentation system. The correlation between configuration items in the assessment data is supplemented by the ESG-specific large model to obtain the integrated assessment data. Based on the integrated assessment data, the enterprise ESG assessment baseline and the industry baseline, the results are synthesized from multiple dimensions to form a hierarchical, multi-dimensional core assessment outcome. Based on the user's needs, we will customize the output format and content focus.

[0037] Furthermore, the generation of personalized evaluation results in step S6 includes the following steps: S61, Assessment Data Integration and Calibration. Automatically summarize the baseline assessment data from each stage that has passed configuration review in step S5, and combine it with the process records and version chain information in the configuration documentation system to complete the integration, verification and calibration of the data, providing high-quality data support for the generation of results. At the same time, through the correlation analysis capabilities of the ESG-specific large model, it supplements the synergistic / constraint relationships between configuration items and improves the data dimensions. S62, Multi-dimensional assessment result synthesis. Based on the integrated and calibrated assessment data, and combined with the personalized and industry-specific benchmarks established in step S4, the results are synthesized from multiple dimensions, including "dimension-element-configuration" and "compliance-effectiveness-advancedness," to form hierarchical and multi-dimensional core assessment results; S63, Personalized Output Adapted to Required Needs. Based on the needs of different stakeholders within the enterprise, customized assessment results output formats and content priorities are provided. Output formats include real-time dashboards, dynamic scoring reports, element correlation diagrams, risk warning lists, and optimization suggestions, achieving personalized adaptation of "one assessment, multiple results."

[0038] Through the embodiments of the present invention, by integrating multi-stage assessment data and using a dedicated ESG model to complete the correlation of configuration items, and combining enterprise and industry baselines, multi-dimensional and hierarchical assessment results are synthesized, and customized output according to user needs is supported, which significantly improves the completeness, interpretability and application adaptability of assessment results.

[0039] This invention describes a dynamic evaluation method for ESG performance based on configurational governance and a large-scale model. Using configurational governance theory as a framework and a large-scale model as support, it constructs an ESG configurational governance indicator system, establishes a multi-source data acquisition and processing network with a dual mechanism of "intelligent cleaning + manual review," trains a large-scale model in the ESG vertical domain as an intelligent ESG data processing and analysis engine, and achieves dynamic evaluation of ESG performance and data quality control and traceability based on the ESG-specific large-scale model integrated with the configurational governance framework. Finally, it outputs evaluation results and optimization suggestions through multi-dimensional correlation analysis.

[0040] The following describes an example of the practical application of the ESG performance dynamic evaluation method based on configurational governance and large models provided by this invention. Specifically, it includes the following steps.

[0041] S1. Construct an ESG governance indicator system. This embodiment is based on the core logic of "system collaboration and element linkage" in governance, and constructs an ESG governance indicator system suitable for steel enterprises in the growth stage of a high-energy-consuming industry through a three-step method of "hierarchical decomposition - correlation modeling - dynamic weighting". The specific steps are as follows: S11, based on the new environmental regulations for the steel industry and combined with the governance needs of steel enterprises for "basic compliance + green transformation start-up," constructs a three-level hierarchical structure of "dimension-elements-configuration." The dimensions are three core dimensions: Environment (E), Society (S), and Governance (G). The E dimension focuses on configuration elements such as "energy management," "emission control," and "water resource utilization." The S dimension focuses on configuration elements such as "employee health and safety," "labor rights," and "supply chain responsibility." The G dimension focuses on configuration elements such as "governance structure," "internal control compliance," and "information disclosure." Elements are further concretized through configuration items. For example, "emission control," a core configuration element in the energy-intensive steel industry, is broken down into indicators including: carbon emission intensity (tons / 10,000 yuan of output value), green electricity usage ratio (%), and carbon emission reduction target completion rate (%). For instance, under carbon emission management, quantifiable indicators such as carbon emission intensity and carbon emission reduction target completion rate are further set, with each indicator constituting a configuration item. The dimensions, elements, and configuration items together form the configuration of the enterprise's ESG assessment. S12 introduces a Bayesian network model to quantify the synergistic and restrictive relationships among ESG elements, constructs an indicator correlation matrix, combines a review of ESG reports from leading steel companies over the past three years, identifies core related elements, and quantifies the correlation strength between elements through expert evaluation or statistical analysis. S13 employs a dynamic weighting system consisting of "basic weights + scenario adjustment coefficients" to establish a dynamic weight allocation mechanism. A judgment matrix is ​​constructed using the analytic hierarchy process (AHP) to set basic weights, and adjustment coefficients are set based on industry characteristics. S14, Establish an ESG Configuration Library. ESG configuration items, as the basic units for performance evaluation, combine configuration management theory with ESG assessment practice. Each configuration item possesses four core attributes across four dimensions: "identification, content, management, and application." The identification dimension ensures the unique positioning of the configuration item, including its code, name, level, and dimension. The content dimension clarifies the core meaning and calculation rules of the configuration item, including data type, calculation rules, meaning description, and industry benchmark values. The management dimension supports the full lifecycle management of configuration items, including data source, update frequency, status, and iteration cycle. The application dimension relates the assessment value and scenario adaptability of the configuration item, including applicable industries, applicable enterprises, and policy basis. The ESG configuration library integrates a four-level structure of "dimension—element—configuration item—association rules," serving as a unified knowledge foundation for subsequent assessment task generation, benchmark construction, and model training.

[0042] S2, Establish an ESG data acquisition and processing network. This embodiment constructs an ESG data network architecture of "distributed acquisition - centralized processing - distributed invocation" to realize the full-process processing of multi-source data. The specific steps are as follows: S21, the data network architecture adopts a microservice architecture, which includes a four-layer structure of acquisition layer, transmission layer, processing layer and storage layer, and the layers communicate with each other through standardized API interfaces. S22: Deploy data collection nodes based on three sources: "public data + private data + supplementary data". Public data collection involves developing targeted web crawlers and API integration modules to connect with the Shanghai Stock Exchange API, the Ministry of Ecology and Environment's public platform, news aggregation platforms, etc., to crawl enterprise ESG reports, public opinion information, etc. Private data collection involves connecting with enterprise ERP systems, energy management systems, human resource management systems, etc., to extract structured and unstructured data such as production energy consumption data, employee training hours, work injury rates, environmental management systems, and employee rights protection manuals. Supplementary data collection can be achieved by establishing data cooperation with carbon trading markets, ESG data service providers, and third-party environmental certification agencies, and by using standardized APIs to achieve real-time synchronization of authorized data or by building online targeted questionnaire survey platforms.

[0043] S23 establishes a full-process processing mechanism of "cleaning-transformation-fusion-verification". Data cleaning adopts a dual strategy of "rule engine + machine learning". The rule engine removes duplicate data and corrects format errors, while machine learning identifies abnormal data through algorithms such as random forest and clustering, and triggers manual review after marking. Data transformation uses models such as BERT to perform semantic analysis on unstructured text and performs unit conversion and indicator mapping on heterogeneous data. Data fusion uses "unique indicator identifier + timestamp" as the fusion primary key and compares data from multiple sources. Data verification verifies the rationality through logical consistency verification and rule construction. Finally, the raw data is transformed into standardized and usable ESG data assets to form an ESG data resource pool. S3. Construct a large-scale ESG-specific model. Based on a general pre-trained large-scale model, domain reinforcement and structured capability injection are performed by incorporating ESG governance knowledge. The specific steps are as follows: S31. We selected a general-purpose pre-trained large model with strong language understanding capabilities, such as LLaMA-3 70B, as the base model, built a GPU cluster training environment, and constructed a dedicated evaluation benchmark for ESG models that includes four dimensions: "text interpretation, association analysis, risk identification, and report generation". The quantitative indicators include text interpretation accuracy, association analysis accuracy, risk identification recall, and report generation satisfaction. S32 constructs a "structured, multi-type complementary" dataset for model training. This involves collecting industry reports (MSCI Steel Industry ESG Rating Methodology) and ESG reports from leading steel companies over the past five years. Core corpus collection and processing are completed through algorithmic deduplication, low-quality content filtering, and data annotation.

[0044] S33 completes multi-stage fine-tuning training of the model through domain knowledge injection, enhancement of structured task capabilities (such as configuration reasoning, element matching, and decoding of index scoring rules), and optimization based on human feedback (RLHF). S34 completes model testing and verification through evaluations of accuracy, robustness, and interpretability; implements containerized model management based on Kubernetes, supporting automatic scaling; and designs RESTful API interfaces to implement model calls in a microservice manner. S4. Establish personalized and industry-specific benchmarks. Using a dedicated ESG model as the core, and linking enterprise heterogeneity characteristics, ESG governance indicator systems, and ESG data pools, and after dual manual review by ESG experts and enterprise management, construct enterprise ESG assessment baselines and industry baselines. The specific steps are as follows: S41, Enterprise Heterogeneity Feature Collection and Feature Vector Construction. A multi-dimensional feature collection list is designed, covering core heterogeneous attributes of enterprises: basic information including enterprise name, unified social credit code, registered address, and establishment date; industry attributes, pre-defined economic industry classification to determine sub-sectors (in this example, the steel industry - ferrous metal smelting and rolling processing); size type, classified as medium-sized enterprises; organizational governance structure, including equity nature (private), board size, and independence; the collected heterogeneous features are processed hierarchically and semantically, extracting semantic feature vectors to achieve quantitative representation of textual information and form enterprise feature vectors; S42, based on enterprise feature vectors, an ESG configuration library, and a dedicated ESG model, completes the selection of configuration elements and the generation of basic configuration baselines for suitable enterprises. The enterprise feature vectors generated in S41 are input into the dedicated ESG model. The model uses a cosine similarity algorithm to calculate the matching degree between the enterprise feature vectors and the scenario feature vectors of each configuration item in the configuration library. Based on a set threshold, configuration items with satisfactory matching degrees are selected to form an initial configuration set. For growth-stage enterprises in the steel industry, the focus is on matching core configuration-related elements such as "carbon footprint management," "pollution prevention and control," and "improvement of governance structure." Based on the initial configuration set and combined with the enterprise's basic configuration item requirements, the final basic configuration baseline for enterprise ESG assessment is generated, covering all levels of elements in the three dimensions of E, S, and G, along with supporting definitions, rules, evidence, and evaluation content.

[0045] S43 focuses on the characteristics of the steel industry, relying on a dedicated ESG model to analyze relevant corpus data. This corpus covers steel industry disclosure standards, regulatory policies, and operational cases of leading companies (annual ESG reports). It identifies industry-specific disclosure requirements (such as mandatory disclosure of blast furnace gas recovery and utilization rates, and solid waste comprehensive utilization rates) and core risk exposure characteristics (such as carbon emission exceeding standards and pollutant emission compliance risks). Based on these core industry characteristics, it selects industry-appropriate configuration items from an ESG configuration library to form an industry configuration set. It also collects industry report data, industry statistics, and authoritative institutional quantitative indicators. Through the dedicated ESG model, it cleans, deduplicates, and statistically analyzes the collected raw data, calculating the industry average for each configuration item to form an industry benchmark system of "configuration item - compliance value - average value - excellent value." S44. Establish a dual review mechanism involving ESG experts and corporate management, as well as a dynamic calibration mechanism of "data feedback - policy tracking - expert review" to review the rationality of indicators, the completeness of the configuration, and the scientific nature of the baseline; build a dynamic update module on the ESG data management platform to realize data feedback, policy tracking, and expert review, and optimize and adjust the corporate ESG assessment basic configuration baseline and the industry basic configuration baseline; assign a version number to each update of the benchmark (e.g., V1.0 is the initial version, V1.1 is the first policy adjustment version), record the update time, update content, triggering reason, reviewer, and other information to form a complete version update log, ensuring that the benchmark adjustment process is traceable and auditable.

[0046] S5, based on configuration control theory, achieves dynamic ESG assessment and personalized result output for steel enterprises through a full-link mechanism of "baseline construction - task allocation - execution and reporting - review and release - change iteration - result generation". The specific steps are as follows: S51, using the steel enterprise's personalized basic configuration baseline and the steel industry's basic configuration baseline established in step S4 as the core benchmark blueprint, and combining the specific stage requirements of the enterprise's ESG performance assessment for a certain year, targeted screening and combination are carried out on the configuration items of the basic configuration baseline. All core configuration items related to the annual performance assessment are retained, while temporary and special configuration items are eliminated, and finally, a stage assessment baseline for a certain year is generated. After the stage assessment baseline is generated, the baseline snapshot mechanism of the ESG data management platform is automatically triggered to record the core information of the initial state of the baseline for that stage: including a complete list of configuration items, scoring rules for quantitative / qualitative indicators, and data collection sources corresponding to each indicator. The snapshot information is stored in the configuration record system, providing the original data foundation for subsequent baseline change tracing and comparison of different stage versions.

[0047] S52, based on the phased assessment baseline generated in step S51, calls the task scheduling module of the ESG-specific large model constructed in step S3. Relying on the model's natural language processing and task decomposition capabilities, the assessment tasks are automatically broken down. Each task has clearly defined core elements, including: task name (e.g., "2024 Annual Carbon Emission Intensity Data Collection and Scoring"), corresponding configuration items (carbon footprint management - carbon emission intensity), assessment criteria (industry compliance value, industry average value, industry excellent value), data requirements (2024 raw coal / electricity consumption data for each production line, annual operating revenue, third-party carbon emission monitoring data), and completion deadline (January 20, 2025). The ESG-specific large model simultaneously retrieves the list of responsibilities and permissions of personnel in various positions within the enterprise and assessment execution efficiency data from the past three years, achieving precise task allocation through multi-dimensional feature matching. S53, the execution unit carries out implementation work according to the assigned assessment task requirements, such as extracting raw coal and electricity consumption data of each production line in 2024 from the enterprise's energy management system, calculating the total carbon dioxide emissions of each production line in conjunction with the carbon emission coefficient of the steel industry published by the state, linking it with annual operating revenue, and finally completing the numerical calculation of the "carbon emission intensity" indicator and uploading supporting materials; at the same time, an internal audit process is executed, with each execution unit designating a specific person as the auditor to conduct a preliminary verification against the requirements of the phased assessment baseline; if problems are found during the audit, the auditor immediately returns the task to the execution unit and notes the reason for the return, and the execution unit supplements the relevant materials and resubmits for audit until it meets the requirements; all internal audit processes are synchronously entered into the configuration recording system to form a complete internal audit trajectory record; S54, after each execution unit completes task implementation and internal preliminary review, enters the multi-dimensional review and phase baseline release stage. The core is to verify the effectiveness of the assessment results through cross-review, confirm the final status of the baseline, and complete the transition process. Based on the pre-set configuration review mechanism, a review process is formulated. Reviewers verify the consistency between the assessment results and the phase baseline requirements, including whether the scores of each indicator strictly match the pre-set interval rules or level standards, whether the indicator calculation methods are completely consistent with the baseline definition, and the compliance and credibility of the data sources. Reviewers complete the review and issue review opinions (such as "passed" or "returned"). The review opinions must clearly indicate the review basis and the problems found. After the execution unit completes the problem correction, it resubmits for review until all reviews are passed. Then, the phase baseline release operation is carried out, a release announcement is generated and synchronized to all stakeholders, and the transition process is initiated. The configuration record system stores the entire process record, automatically builds the ESG assessment version chain, supports retrospective, comparative analysis and audit traceability of any version, and provides a clear historical basis for subsequent baseline change iterations. S55, based on changes in business needs and configuration control theory, conducts adaptive adjustments, change management, and review and confirmation of published phased baselines. The change process strictly follows the configuration control workflow of "application-review-implementation-review-release," and all adjustment actions are fully recorded in the configuration documentation system to ensure traceability of the change trajectory. First, the entire process begins with the enterprise's ESG-related personnel submitting a baseline change application, clearly including the reason for the change, the content of the change, the scope of impact, and the expected completion time. After review and confirmation, the change application is included in the configuration documentation system. Second, adaptive adjustments to the phased baseline are carried out according to the change plan. After the adjustment is completed, the adjustment review process is initiated, and a new release baseline is generated after the review is approved. Finally, the ESG assessment version chain is updated synchronously, and a snapshot of the baseline after this change is recorded. S6, Generate Personalized Assessment Results. Based on the release baseline generated in step S5 and the needs of different stakeholders, the assessment results are updated in real time and applied in multiple scenarios through a complete process of "assessment data integration and calibration - multi-dimensional result synthesis - demand-adapted personalized output". The specific steps are as follows: In step S61, the data integration module of the ESG data management platform automatically retrieves the baseline assessment data from each stage that has passed the configuration review in step S5. Simultaneously, it extracts process records and version chain information from the configuration documentation system, summarizing them to form an initial assessment dataset, which is stored in a Redis hot database to ensure subsequent processing efficiency. The association analysis capabilities of the ESG-specific large model built in step S3 are invoked, and the indicator association rule library built in step S1 is accessed to supplement the collaborative / constraint relationship data between configuration items. Through model calculation, four-dimensional association data of "dimension-element-configuration-association strength" is generated, improving the comparative dimensions and analytical value of the data.

[0048] S62, based on the integrated and calibrated high-quality assessment data, and combined with the personalized and industry-specific benchmarks established in step S4, conducts result synthesis to form hierarchical and multi-dimensional core assessment deliverables. Result synthesis is carried out hierarchically according to "dimension-element-configuration," while simultaneously generating a hierarchical score traceability link through a dedicated ESG model, supporting reverse penetration from the overall score to the original data and supporting materials of specific indicators. Multi-attribute synthesis is also conducted according to "compliance-effectiveness-advancedness," combining the industry compliance value, average value, and excellent value benchmark system established in step S4. Finally, a core deliverable package containing "hierarchical score details + industry benchmarking results + multi-attribute assessment conclusions" is generated and stored in the deliverables repository of the ESG data management platform.

[0049] S63 offers customized output formats and content focuses based on the core needs of different stakeholders; it uses visual charts (radar charts, line charts, heatmaps) to display core indicator scores, industry benchmarking, etc.; it supports automatic generation of Word / PDF reports based on preset templates, with automatic data filling, version tracking, and electronic signature functions; it displays the synergistic / constraint relationships between various ESG configuration items through visual graphs, supporting correlation path queries and impact degree analysis; and it leverages the real-time synchronization mechanism of the ESG data management platform to achieve dynamic updates of assessment results.

[0050] This invention proposes a dual-core framework of "configurational governance and large-scale model" for ESG assessment. For the first time, the core logic of "system synergy and element linkage" in configurational governance is introduced into the field of ESG assessment, breaking through the limitations of the traditional "linear listing" of ESG indicators. Furthermore, it deeply integrates with the structured semantic understanding capabilities of the large-scale model, achieving overall structuring and intelligentization from indicator system design and assessment process to result generation.

[0051] A dual-baseline collaborative assessment framework of personalized and industry-specific baselines is proposed. Through a dynamic matching mechanism driven by enterprise feature vectors and industry feature vectors, enterprise configuration baselines and industry configuration baselines are automatically generated, making ESG assessments both "enterprise-specific" and "industry-comparable," significantly overcoming the structural defects of the traditional "one-size-fits-all" ESG evaluation system.

[0052] It innovatively introduces the "configuration baseline" and "baseline version chain" mechanisms. It proposes a structured management approach that involves phased assessment of baselines, release of baselines, and baseline snapshots, enabling full lifecycle management of ESG assessment rules, weights, and configuration items at different stages and versions, thus solving the problems of untraceability and lack of transparency in traditional ESG assessments.

[0053] Automatic generation of configuration tasks and intelligent allocation of execution units based on large models. By using large models to perform semantic parsing and task decomposition of configuration items, evaluation tasks are automatically generated and intelligently allocated in conjunction with the enterprise's organizational structure. This achieves an efficient, low-cost, and controllable evaluation execution process, providing a new technical approach for the automation of ESG evaluation.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic evaluation method for ESG performance based on configurational governance and large-scale models, characterized in that, include: Construct an ESG governance indicator system, which includes: breaking down core elements from three dimensions: environment, society, and governance, and establishing a three-level indicator system of dimensions, elements, and configuration; Based on the ESG configuration governance index system, multi-source ESG data is collected and preprocessed to form a structured ESG data pool. Based on an open-source large language model, an ESG-specific large model is constructed by fine-tuning the ESG domain corpus and instructions in the ESG data pool. The ESG-specific large model has the ability to interpret ESG text, perform data correlation analysis, identify risks, and generate assessment reports. Based on the ESG configuration governance indicator system, the ESG data pool, and the ESG-specific large model, the basic configuration baseline for enterprise ESG assessment and the basic configuration baseline for the industry are generated. Based on the ESG-specific large model and configuration governance framework, the system calls the enterprise ESG assessment basic configuration baseline and the industry basic configuration baseline to generate a phased assessment baseline, automatically creates assessment tasks for related configuration items and assigns execution units, performs element matching, correlation analysis, risk level determination and performance score calculation on the preprocessed data, realizes ESG assessment, and obtains the published baseline. Based on the aforementioned release baseline and user needs, the evaluation results are output in the form of real-time dashboards, dynamic scoring reports, element association maps, risk warning lists, and optimization suggestions. The ESG configuration governance indicator system includes a primary dimension, secondary elements, and a tertiary configuration. The secondary elements are detailed based on the primary dimension, and the tertiary configuration is a quantifiable indicator of the secondary elements. The primary dimensions include the environmental dimension, the social dimension, and the governance dimension; The secondary elements include carbon emission management and resource utilization efficiency under the environmental dimension, employee rights protection and supply chain responsibility under the social dimension, and corporate governance structure and internal control under the governance dimension. The method further includes: Construct an element association rule model, wherein the element association rule model is used to quantify the synergistic effects and constraints among the secondary elements; Based on the target company's industry type, company size, and development stage, set the weight adjustment coefficients for the ESG governance indicator system. Based on the element association rule model and the weight adjustment coefficient, an ESG configuration library is constructed. The ESG configuration library includes: the first-level dimension, the second-level element, the third-level configuration, and the four-level mapping relationship between association rules.

2. The dynamic evaluation method for ESG performance based on configurational governance and large-scale models according to claim 1, characterized in that, The governance indicator system based on the ESG configuration involves collecting multi-source ESG data and preprocessing the multi-source ESG data to form a structured ESG data pool, including: A microservice architecture is used to build a flexible and scalable data network, wherein the different layers of the data network communicate with each other through standardized API interfaces; Data is collected through preset multi-source data collection nodes to obtain publicly available automated data, privately authorized access data, and supplementary customized data. The collected publicly available automated data, privately authorized access data, and supplementary customized data are cleaned, transformed, and fused for verification to obtain standardized ESG data, forming a structured ESG data pool.

3. The dynamic evaluation method for ESG performance based on configurational governance and large-scale models according to claim 1, characterized in that, The aforementioned method, based on an open-source large language model, constructs a dedicated ESG large model through fine-tuning of ESG domain corpora and instructions in the ESG data pool, including: Select a general pre-trained large model with language understanding capabilities as the base model, and determine the parameter scale, architecture type, and input / output format of the base model; An ESG domain training dataset is constructed based on the ESG data pool, and the ESG domain training dataset is subjected to secondary semantic cleaning and assigned standardized domain labels. Based on the ESG domain training dataset, the base model is fine-tuned in three stages: domain knowledge injection, task capability enhancement, and human feedback optimization, to obtain the initial ESG-specific large model. An ESG-specific evaluation set is constructed to conduct multi-dimensional tests on the initial ESG-specific large model. After continuous optimization based on the test results, a well-trained ESG-specific large model is obtained. Deploy the trained ESG-specific large model using API, microservices, or localization.

4. The dynamic evaluation method for ESG performance based on configurational governance and large-scale models according to claim 1, characterized in that, Based on the ESG governance indicator system, the ESG data pool, and the dedicated ESG model, the system generates basic baselines for enterprise ESG assessment and industry-specific baselines, including: Based on the ESG governance index system, the heterogeneity characteristics of the target enterprise are collected, and enterprise feature vectors are formed through feature normalization, classification coding, and semantic modeling. Based on the enterprise feature vector, the ESG configuration library, and the ESG-specific large model, matching configuration elements and configuration items are selected to generate a basic configuration baseline for enterprise ESG assessment that includes indicator definitions, calculation rules, evidence requirements, and evaluation logic. Based on the enterprise feature vectors, the ESG configuration library, and the ESG-specific large model, industry corpus is analyzed to screen industry-adaptive configuration items, and industry-related data is combined to form an industry benchmark value system and generate industry basic configuration baselines.

5. The dynamic evaluation method for ESG performance based on configurational governance and large-scale models according to claim 1, characterized in that, Based on the aforementioned ESG-specific large-scale model and configuration governance framework, the system calls upon the enterprise's basic ESG assessment configuration baseline and the industry's basic configuration baseline to generate a phased assessment baseline. It automatically creates assessment tasks and assigns execution units to related configuration items. The preprocessed data undergoes element matching, correlation analysis, risk level determination, and performance scoring calculation to achieve ESG assessment and obtain the published baseline, including: Based on the enterprise ESG assessment baseline and industry baseline, a phased assessment baseline is formed, and a baseline snapshot is taken of the phased assessment baseline to form an initial version record. The assessment tasks corresponding to the phased assessment baseline are broken down and assigned to the execution units using the ESG-specific large model. The evaluation results completed by the execution unit are reviewed, and the review records are incorporated into the configuration documentation system; The evaluation results after review are subject to multi-dimensional cross-review, and a release baseline is generated after the review is passed. Based on the aforementioned release baseline, update the baseline version chain.

6. The method for dynamic evaluation of ESG performance based on configurational governance and large-scale models according to claim 1, characterized in that, The process of outputting evaluation results based on the aforementioned release baseline and user needs includes: The assessment data of each of the aforementioned phased assessment baselines are integrated, and the data integration, verification and calibration are completed by combining the configuration documentation system. The correlation between configuration items in the assessment data is supplemented by the ESG-specific large model to obtain the integrated assessment data. Based on the integrated assessment data, the enterprise ESG assessment baseline and the industry baseline, the results are synthesized from multiple dimensions to form a hierarchical, multi-dimensional core assessment outcome. Based on the user's needs, we will customize the output format and content focus.

Images