ESG rinsing behavior identification method and device and ESG evaluation system

Abstract

The invention provides an ESG rinsing behavior identification method and device and an ESG evaluation system, and the method comprises the steps: obtaining ESG evaluation data of a plurality of enterprises, the ESG evaluation data comprising enterprise annual report data and enterprise operation data; training the sparse representation model by adopting ESG evaluation data to obtain an ESG rinsing behavior prediction model; inputting ESG evaluation data of an enterprise to be predicted and identified into the ESG rinsing behavior prediction model to obtain an ESG rinsing behavior index; the ESG rinsing behavior prediction model is subjected to postmortem interpretability analysis through an SHAP algorithm, an SHAP analysis result is obtained, and the SHAP analysis result is the distribution situation of influences of all indexes of ESG evaluation data on ESG rinsing behavior indexes; and the ESG rinsing behavior index and the SHAP analysis result are displayed. The problem that in the prior art, enterprise ESG rinsing behavior recognition and interpretation are lack of effective technical means is solved.

Description

Technical Field

[0001] This application relates to the field of ESG rinsing identification technology, and more specifically, to a method, apparatus, computer-readable storage medium, and ESG evaluation system for identifying ESG rinsing behavior. Background Technology

[0002] Companies often exhibit discrepancies between their stated and actual ESG practices. This "ESG washing" behavior artificially creates information asymmetry, thereby impacting stakeholder interests. Leveraging artificial intelligence to curb corporate washing tendencies is a viable approach. Predicting and identifying ESG washing behaviors can enhance stakeholder understanding of corporate ESG practices, reduce the difficulty for regulators in identification, alleviate information asymmetry, and provide stronger support for investor decision-making.

[0003] The stringent regulatory and credibility requirements in the financial sector necessitate certain standards for the transparency, interpretability, and fairness of the models themselves. Secondly, the high dimensionality and tabular nature of financial data are sources of data redundancy. Therefore, addressing the issues of interpretability, credibility, and data redundancy is crucial for the regulation and identification of corporate ESG washing practices. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, computer-readable storage medium, and ESG evaluation system for identifying ESG rinsing behavior, so as to at least solve the problem of poor interpretability of ESG rinsing behavior identification results in the prior art.

[0005] To achieve the above objectives, according to one aspect of this application, a method for identifying ESG rinsing behavior is provided, comprising: acquiring ESG evaluation data from multiple enterprises, the ESG evaluation data including enterprise annual report data and enterprise operating data; training a sparse representation model using the ESG evaluation data to obtain an ESG rinsing behavior prediction model; inputting the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain an ESG rinsing behavior index; performing a post-hoc interpretability analysis on the ESG rinsing behavior prediction model using the SHAP algorithm to obtain SHAP analysis results, the SHAP analysis results being the distribution of the influence of each indicator of the ESG evaluation data on the ESG rinsing behavior index; and displaying the ESG rinsing behavior index and the SHAP analysis results.

[0006] Optionally, the sparse representation model is trained using the ESG evaluation data to obtain an ESG rinsing behavior prediction model, including: preprocessing the ESG evaluation data of multiple companies to obtain preprocessed ESG evaluation data, wherein the preprocessing includes handling missing values, logarithmic transformation, normalization, and removal of text symbols; performing sentiment analysis on the preprocessed ESG evaluation data of each company to obtain ESG evaluation data of multiple sentiment types for each company, wherein the sentiment types include positive, negative, neutral, and meaningless; and training the sparse representation model using the ESG evaluation data of multiple sentiment types from all the companies to obtain the ESG rinsing behavior prediction model.

[0007] Optionally, the sparse representation model is trained using ESG evaluation data of multiple sentiment types from all the enterprises to obtain the ESG rinsing behavior prediction model. This includes: inputting the ESG evaluation data of multiple sentiment types from all the enterprises into a training configuration page, where the training configuration page is used to select the indicators for training and configure the dataset partitioning ratio, where the dataset consists of a training set and a validation set; and training the sparse representation model through the training configuration page to obtain the ESG rinsing behavior prediction model.

[0008] Optionally, training the sparse representation model through the training configuration page to obtain the ESG rinsing behavior prediction model includes: training the sparse representation model through the training configuration page until the loss function value meets a first predetermined condition to obtain a predetermined ESG rinsing behavior prediction model; if the output result of the predetermined ESG rinsing behavior prediction model meets a second predetermined condition corresponding to the evaluation method, the predetermined ESG rinsing behavior prediction model is determined as the ESG rinsing behavior prediction model, wherein the evaluation method includes the MAE evaluation method and the R2 evaluation method.

[0009] Optionally, sentiment analysis is performed on the preprocessed ESG evaluation data of each enterprise to obtain ESG evaluation data of multiple sentiment types for each enterprise, including: obtaining a pre-trained large model; training the pre-trained large model using the ESG evaluation data of multiple enterprises and the corresponding manually labeled sentiment types to obtain a sentiment analysis model; and inputting the preprocessed ESG evaluation data of each enterprise into the sentiment analysis model to perform sentiment analysis to obtain ESG evaluation data of multiple sentiment types for each enterprise.

[0010] Optionally, the ESG evaluation data of the multiple enterprises are preprocessed to obtain preprocessed ESG evaluation data, including: performing missing value processing, logarithmic processing, and normalization processing on the numerical ESG evaluation data to obtain the preprocessed ESG evaluation data; and removing text symbols from the text-based ESG evaluation data to obtain the preprocessed ESG evaluation data.

[0011] Optionally, displaying the ESG rinsing behavior index and the SHAP analysis results includes: obtaining the minimum, maximum, and average values ​​of the ESG rinsing behavior index for all sample data in the industry where the enterprise to be identified is located, wherein the sample data is the ESG evaluation data of enterprises in the industry where the enterprise to be identified is located; inputting the ESG rinsing behavior index, the SHAP analysis results, and the minimum, maximum, and average values ​​of the ESG rinsing behavior index into a large model to obtain the large model interpretation results, wherein the large model interpretation results are an explanation of the ESG rinsing behavior index and the SHAP analysis results; and displaying the ESG rinsing behavior index, the SHAP analysis results, and the large model interpretation results.

[0012] According to another aspect of this application, an ESG rinsing behavior identification device is provided, comprising: an acquisition unit for acquiring ESG evaluation data of multiple enterprises, the ESG evaluation data including enterprise annual report data and enterprise operating data; a training unit for training a sparse representation model using the ESG evaluation data to obtain an ESG rinsing behavior prediction model; an input unit for inputting the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain an ESG rinsing behavior index; an analysis unit for performing a post-hoc interpretability analysis on the ESG rinsing behavior prediction model using the SHAP algorithm to obtain SHAP analysis results, the SHAP analysis results being the distribution of the influence of each indicator of the ESG evaluation data on the ESG rinsing behavior index; and a display unit for displaying the ESG rinsing behavior index and the SHAP analysis results.

[0013] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform any of the ESG rinsing behavior identification methods described above.

[0014] According to another aspect of this application, an ESG evaluation system is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including an identification method for performing any of the ESG rinsing behaviors described above.

[0015] By applying the technical solution of this application, the above-mentioned method for identifying ESG rinsing behavior obtains an ESG rinsing behavior prediction model by training a sparse representation model. This model can then predict the ESG rinsing behavior of the enterprise to be identified, yielding an ESG rinsing behavior index. Since the sparse representation model has good interpretability, the SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model, resulting in SHAP analysis results. The distribution of the influence of each indicator on the ESG rinsing behavior index not only characterizes the overall ESG rinsing level of the enterprise to be identified based on the ESG rinsing behavior index, but also displays the distribution of indicators affecting the ESG rinsing behavior index, improving the interpretability of the identification results and solving the problem of poor interpretability of existing ESG rinsing behavior identification results. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for identifying ESG rinsing behavior according to an embodiment of this application is shown.

[0018] Figure 2 A flowchart illustrating a method for identifying ESG rinsing behavior according to an embodiment of this application is shown.

[0019] Figure 3 A flowchart illustrating another method for identifying ESG rinsing behavior according to an embodiment of this application is shown.

[0020] Figure 4 A schematic diagram of a process for rapid import and processing of multi-source data according to an embodiment of this application is shown;

[0021] Figure 5 A flowchart illustrating a model training and prediction module according to an embodiment of this application is shown.

[0022] Figure 6 A flowchart illustrating an ex post-explanability module provided according to an embodiment of this application is shown.

[0023] Figure 7 A structural block diagram of an ESG rinsing behavior identification device provided according to an embodiment of this application is shown.

[0024] The above figures include the following reference numerals:

[0025] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0030] Machine learning: Machine learning is a branch of artificial intelligence that uses algorithms and models to enable computers to automatically learn from data and make predictions or decisions. Its core goal is to allow computers to analyze large amounts of data, identify patterns and rules, and build models that adapt to new data without explicit programming instructions. Machine learning includes different types such as supervised learning, unsupervised learning, and reinforcement learning, and is widely used in fields such as image recognition, natural language processing, recommendation systems, and autonomous driving. It possesses adaptive, automated, and generalization capabilities and represents data-driven technological innovation.

[0031] ESG: Environmental, Social and Governance (ESG) encompasses three aspects: a company's environmental impact, its social responsibility, and its internal governance. It is used by companies to regulate and monitor their own behavior, and is also a basis for investors to measure the sustainability of the companies or assets they invest in. It is also one of the important strategies used by institutional investors internationally for decision-making.

[0032] ESG Washing: When there is ambiguity in the evaluation standards for corporate ESG performance and uncertainty in returns, companies may disclose ESG information that is inconsistent with reality. This is called ESG washing, which generally includes ESG greening (reflecting "exaggeration") and ESG browning (reflecting "more talk than action").

[0033] SHAP (Shapley Value Analysis) is a widely used interpretability analysis method for machine learning models. Its core idea is to interpret the model's predictions by calculating the marginal contribution of features to the model's output. Based on Shapley value theory, SHAP can explain "black box models" from both global and local perspectives. The model-independent KernelSHAP approximation method is applicable to any type of machine learning model. It borrows ideas from LIME (Limited Information Modeling) and estimates the SHAP value by performing weighted linear regression on the local behavior of the original model.

[0034] Random Forest: An ensemble learning algorithm that makes a final decision by constructing multiple decision trees and aggregating their predictions. This approach not only increases the model's accuracy but also improves its robustness to data anomalies and overfitting.

[0035] XGBoost is an optimized version of the gradient boosting algorithm that uses an additive model, where each newly added model reduces the residual of the previous model. The core of XGBoost lies in constructing multiple decision trees and summing their predictions to obtain the final prediction.

[0036] Sparse representation refers to representing a signal or sample using linear combinations of elements from a dictionary. Its main purpose is to represent a signal using as few atoms as possible from a given overcomplete dictionary, thus achieving a more concise representation that facilitates signal processing and analysis. Mathematically, sparse representation can be achieved by finding a coefficient matrix and a dictionary matrix whose product best restores the original signal while preserving the sparsity of the coefficient matrix.

[0037] The LISTA algorithm, or Learned Iterative Shrinkage Thresholding Algorithm, is an optimization of the ISTA algorithm. ISTA is a greedy algorithm that minimizes the objective function iteratively to recover the original signal. LISTA, on the other hand, replaces the threshold function in ISTA with a learnable deep neural network, thereby improving recovery performance.

[0038] As described in the background section, the interpretability of existing ESG rinsing behavior identification results is poor. To address this issue, embodiments of this application provide an ESG rinsing behavior identification method, apparatus, computer-readable storage medium, and ESG evaluation system.

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0040] The methods and embodiments provided in this application can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for an ESG rinsing behavior identification method according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0041] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the ESG rinsing behavior identification method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0042] This embodiment provides a method for identifying ESG rinsing behavior running on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0043] Figure 2 This is a flowchart of a method for identifying ESG rinsing behavior according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0044] Step S201: Obtain ESG evaluation data from multiple companies, including company annual report data and company operating data;

[0045] Specifically, the measurement method for a company's ESG washing behavior is based on the difference between its environmental information performance and its actual environmental rating performance, and the formula is as follows: ERi,t represents the acquired indicator data. This represents the average value of the indicators reflecting the true performance of the environmental rating. This represents the average value of the indicators reflecting the true performance of the environmental rating. The mean squared error of the indicator data representing the true performance of the environmental rating. The mean squared error of the indicator data representing the true performance of the environmental rating.

[0046] Step S202: Train the sparse representation model using ESG evaluation data to obtain the ESG rinsing behavior prediction model.

[0047] Specifically, ESG evaluation data, including corporate annual reports and operational data, was acquired, and an ESG washing behavior prediction model was constructed through training a sparse representation model. Due to the special nature of financial data, the financial data used for ESG score prediction is typically prepared tabular data. This tabular data is quite large in both length and width dimensions, containing a large amount of redundant information. This information redundancy can lead to performance degradation, and it is also difficult to capture important features and correlations between tabular data. Lista's sparse representation-based algorithm can reduce the dimensionality of the data to sparsity, thereby capturing key features while maintaining good interpretability and generalization performance.

[0048] Step S203: Input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index.

[0049] Specifically, the predictive model was used to analyze the ESG evaluation data of a specific company, generating a quantitative index that reflects its ESG washing behavior tendencies.

[0050] Step S204: The SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model to obtain the SHAP analysis results. The SHAP analysis results show the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index.

[0051] Specifically, the SHAP algorithm was used to conduct a post-hoc interpretability analysis of the prediction model, revealing the specific contributions of each ESG indicator to the rinsing behavior index, i.e., the distribution of its influence.

[0052] Step S205: Display the ESG rinsing behavior index and SHAP analysis results.

[0053] Specifically, the ESG rinsing behavior index and SHAP analysis results are presented to users in a visual way, thus providing the possibility of a deeper understanding of the prediction results.

[0054] In this embodiment, the above-mentioned method for identifying ESG rinsing behavior obtains an ESG rinsing behavior prediction model by training a sparse representation model. This model can then predict the ESG rinsing behavior of the enterprise to be identified, resulting in an ESG rinsing behavior index. Since the sparse representation model has good interpretability, the SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model, yielding SHAP analysis results. The distribution of the influence of each indicator on the ESG rinsing behavior index not only characterizes the overall ESG rinsing level of the enterprise to be identified based on the ESG rinsing behavior index, but also displays the distribution of indicators affecting the ESG rinsing behavior index, improving the interpretability of the identification results and solving the problem of poor interpretability of ESG rinsing behavior identification results in existing technologies.

[0055] Further, in this embodiment, step S202 includes:

[0056] Step S2021: Preprocess the ESG evaluation data of multiple enterprises to obtain preprocessed ESG evaluation data. The preprocessing includes missing value handling, logarithmic processing, normalization processing, and text symbol removal.

[0057] Step S2022: Perform sentiment analysis on the preprocessed ESG evaluation data of each enterprise to obtain ESG evaluation data of multiple sentiment types for each enterprise. The sentiment types include positive, negative, neutral and meaningless.

[0058] Step S2023: Use ESG evaluation data of multiple sentiment types from all enterprises to train the sparse representation model and obtain the ESG rinsing behavior prediction model.

[0059] In this embodiment, ESG evaluation data from multiple enterprises are preprocessed, specifically including missing value handling, logarithmic transformation, normalization, and removal of text symbols to ensure data consistency and reduce noise interference. Next, sentiment analysis is performed on the preprocessed ESG evaluation data from each enterprise, categorizing the data into four sentiment types: positive, negative, neutral, and meaningless. This aims to comprehensively understand the attitudes and tendencies implied in the enterprises' ESG annual reports. Subsequently, a sparse representation model is trained using ESG evaluation data from all enterprises across multiple sentiment types to obtain an ESG rinsing behavior prediction model. This process not only considers the diversity of data but also fully explores the emotional value behind the text. Through the learning of the sparse representation model, key features can be identified in high-dimensional data environments, thereby effectively identifying and predicting potential ESG rinsing behaviors of enterprises. Furthermore, the preprocessing and sentiment analysis steps in this embodiment improve the model's training effect, enhance the accuracy and stability of the prediction model, and provide a solid data foundation and technical support for subsequent ESG rinsing behavior identification. In other embodiments, the classification of sentiment types can also be fine-tuned using different deep learning models, such as the BERT model, which will further improve the accuracy of sentiment analysis and better serve the task of identifying ESG rinsing behaviors.

[0060] Further, in this embodiment, step S2023 includes:

[0061] Step S20231: Input the ESG evaluation data of multiple sentiment types of all enterprises into the training configuration page. The training configuration page is used to select indicators for training and configure the division ratio of the dataset. The dataset consists of a training set and a validation set.

[0062] Step S20232: Train the sparse representation model through the training configuration page to obtain the ESG rinsing behavior prediction model.

[0063] In this embodiment, a sparse representation model is trained using ESG evaluation data of multiple sentiment types from all enterprises to obtain an ESG washout behavior prediction model. The ESG evaluation data of all sentiment types are configured through the training configuration page, allowing for indicator selection and dataset partitioning. The dataset is divided into training and validation sets. During this process, the administrator can flexibly adjust the training parameters according to actual conditions to optimize model performance. By training the sparse representation model through the training configuration page, the model can learn the characteristic patterns of enterprise ESG washout behavior from multi-source data, thereby constructing a prediction model. This design effectively handles high-dimensional data in the financial field, reduces data redundancy, and improves the model's generalization ability and interpretability. When new data is introduced, the prediction model can quickly adapt and accurately predict the degree of ESG washout tendency of enterprises, providing timely and reliable decision support for regulatory agencies and investors. In other embodiments, different machine learning or deep learning models can also be used for training, but the sparse representation model is the preferred solution of this invention due to its advantages in handling high-dimensional data and maintaining model interpretability. In this way, the present invention can meet the specific needs of the financial sector, improve the accuracy and credibility of predictive models, and, through the combination of SHAP and large-scale models, provide in-depth interpretation of the results, enhancing the persuasiveness and practicality of the predictions. Based on this, enterprises, regulators, and investors can more accurately understand and respond to ESG washout practices, promoting sustainable development practices.

[0064] Further, in this embodiment, step S20232 includes:

[0065] Step S202321: Train the sparse representation model through the training configuration page until the loss function value meets the first predetermined condition, and obtain the predetermined ESG rinsing behavior prediction model.

[0066] Step S202322: If the output of the predetermined ESG rinsing behavior prediction model meets the second predetermined condition corresponding to the evaluation method, the predetermined ESG rinsing behavior prediction model is determined as the ESG rinsing behavior prediction model. The evaluation method includes the MAE evaluation method and the R2 evaluation method.

[0067] In this embodiment, the sparse representation model is trained through the training configuration page until the loss function value meets the first predetermined condition, resulting in a predetermined ESG washing behavior prediction model 2. During this process, by adjusting the model parameters and training dataset, the sparse representation model can effectively learn and extract key features from high-dimensional multi-source data to predict corporate ESG washing behavior tendencies. If the output of the predetermined ESG washing behavior prediction model 2 meets the second predetermined conditions corresponding to the MAE evaluation method and the R2 evaluation method, this model is determined as the final ESG washing behavior prediction model. This process ensures the model's accuracy and generalization ability, enabling it to provide reliable results when predicting ESG washing behavior. By setting specific evaluation methods, such as the MAE evaluation method to assess the mean absolute error between the model's prediction results and the actual values, and the R2 evaluation method to measure the model's ability to explain data variations, model performance can be effectively detected, thereby ensuring the credibility of the prediction results. As model training progresses, the loss function value gradually decreases, reaching the first predetermined condition, indicating that the model's prediction accuracy has been optimized, enabling better identification of corporate ESG washing behavior. Of course, in other embodiments not shown, other machine learning models or deep learning models may also be selected for training and prediction, but the high efficiency and interpretability advantages of sparse representation models in processing financial data will be used as evaluation criteria for other models to ensure that the final ESG washing behavior prediction model is not only accurate, but also interpretable and understandable.

[0068] Further, in this embodiment, step S2022 includes:

[0069] Step S20221: Obtain the pre-trained large model;

[0070] Step S20222: Use ESG evaluation data from multiple enterprises and corresponding manually labeled sentiment types to train the pre-trained large model to obtain the sentiment analysis model;

[0071] Step S20223: Input the preprocessed ESG evaluation data of each enterprise into the sentiment analysis model for sentiment analysis to obtain ESG evaluation data of multiple sentiment types for each enterprise.

[0072] In this embodiment, sentiment analysis is performed on the preprocessed ESG evaluation data of various enterprises to obtain ESG evaluation data of multiple sentiment types for each enterprise. The specific implementation process is as follows: First, a pre-trained large model is obtained, which has basic semantic understanding and text classification capabilities. Second, the pre-trained large model is trained using ESG evaluation data from multiple enterprises and corresponding manually labeled sentiment types to obtain a sentiment analysis model. This process involves fine-tuning the model parameters to more accurately identify the sentiment tendencies contained in the ESG evaluation data. Finally, the preprocessed ESG evaluation data of each enterprise is input into the sentiment analysis model for sentiment analysis to obtain ESG evaluation data of multiple sentiment types for each enterprise. This large-model-based sentiment analysis method can effectively process text data such as corporate annual reports and announcements, accurately identify positive, negative, or neutral sentiment tendencies, and provide more comprehensive and in-depth data support for subsequent ESG washing behavior identification, improving the accuracy and reliability of identification. At the same time, through fine-tuning and manually labeled sentiment types, this sentiment analysis model can adapt to the characteristics of different industries and enterprises, enhancing its applicability and interpretability in the financial field. In other embodiments not shown in the figure, the sentiment analysis model can be fine-tuned using different pre-trained large models or by combining industry-specific corpora to further enhance the model's generalization ability and industry adaptability.

[0073] Further, in this embodiment, step S2021 includes:

[0074] Step S20211: Perform missing value processing, logarithmic processing, and normalization processing on the numerical ESG evaluation data to obtain preprocessed ESG evaluation data.

[0075] Step S20212: Remove text symbols from the text-based ESG evaluation data to obtain preprocessed ESG evaluation data.

[0076] In this embodiment, ESG evaluation data from multiple enterprises are preprocessed to obtain preprocessed ESG evaluation data. Specifically, this includes handling missing values, logarithmic transformation, and normalization for numerical ESG evaluation data to ensure data consistency and model training stability. For textual ESG evaluation data, text symbols are removed to eliminate interference from non-semantic information and improve the quality of the textual data. Preprocessing of numerical data, by filling in missing values ​​and transforming the data scale, makes the data more suitable for subsequent model training and prediction. Simultaneously, logarithmic transformation and normalization effectively avoid the impact of extreme values ​​in the data on model performance. For textual data, the removal of non-text symbols improves the accuracy of text analysis, ensuring the precision of sentiment analysis and semantic understanding. The entire preprocessing process aims to provide a clean and standardized dataset for the identification and analysis of enterprise ESG washing behaviors, which is beneficial for improving model training efficiency and prediction accuracy, while also laying a solid foundation for model interpretability and result reliability. In this embodiment, numerical ESG evaluation data undergoes preprocessing, including missing value handling, logarithmic transformation, and normalization, making the data more suitable for model training and prediction needs, reducing noise and bias in the data, and improving the model's generalization ability and prediction accuracy. Textual ESG evaluation data is processed by removing text symbols, eliminating redundant information and making the text data cleaner, which is beneficial for the accuracy of sentiment analysis and semantic understanding, further enhancing the model's recognition performance.

[0077] Further, in this embodiment, step S205 includes:

[0078] Step S2051: Obtain the minimum, maximum, and average values ​​of the ESG rinsing behavior index for all sample data of the industry in which the enterprise to be identified is located. The sample data consists of the ESG evaluation data of enterprises in the industry in which the enterprise to be identified is located.

[0079] Step S2052: Input the ESG rinsing behavior index, SHAP analysis results, and the minimum, maximum, and average values ​​of the ESG rinsing behavior index into the large model to obtain the large model interpretation results. The large model interpretation results are the interpretation of the ESG rinsing behavior index and SHAP analysis results.

[0080] Step S2053 presents the ESG rinsing behavior index, SHAP analysis results, and large model interpretation results.

[0081] In this embodiment, when displaying the enterprise's ESG washing behavior index and SHAP analysis results, we obtained the minimum, maximum, and average values ​​of the ESG washing behavior index from all sample data within the industry of the enterprise to be identified. The sample data includes ESG evaluation data from enterprises in that industry. By inputting the ESG washing behavior index, SHAP analysis results, and these statistical values ​​into a large model based on the Prompt project, we can obtain the large model interpretation results, i.e., an in-depth analysis of the ESG washing behavior index and SHAP analysis results. Visualizing the ESG washing behavior index, SHAP analysis results, and large model interpretation results provides users with intuitive understanding and decision-making support, realizing a full-process service from data input and model prediction to result interpretation. This process not only improves the transparency of model predictions but also enhances the interpretability of results through large model analysis, enabling non-professional users to understand the logic behind model predictions, thereby improving the accuracy and fairness of decision-making. In other embodiments not shown in the figure, different visualization techniques can be employed to further enhance the user's understanding of the data. For example, a heatmap can be used to display SHAP values, with the intensity of color visually reflecting the degree of influence of each feature on the prediction results. Alternatively, a bar chart can be used to compare the ESG rinsing behavior index with the industry average, helping users quickly pinpoint a company's relative performance. These alternatives can also achieve the invention's objectives, providing diverse and personalized results presentation methods.

[0082] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the ESG rinsing behavior identification method of this application will be described in detail below with reference to specific embodiments.

[0083] This embodiment relates to a specific method for identifying ESG rinsing behavior, such as... Figure 3 As shown, it includes the following steps:

[0084] The main processes involved in the rapid import and processing of multi-source data are as follows: Figure 4 As shown, the detailed steps are as follows:

[0085] Step 1: The measurement method for an enterprise's ESG washing behavior is to use the difference between environmental information performance and actual environmental rating performance.

[0086] Step 2: Administrators can import indicators and data into the data access module. The information includes stock code, company name, industry code, indicator name, indicator meaning, indicator calculation method, etc.

[0087] Step 3: For numerical data, use methods such as missing value handling, logarithmic transformation, and normalization to ensure data consistency.

[0088] Step 4: Perform data preprocessing on text data, removing special characters such as spaces and newlines.

[0089] Step 5: Organize the existing annual report data, divide it into four categories: positive, negative, neutral, and meaningless, collect and organize the dataset for fine-tuning the large model, and carry out fine-tuning training on the DeepSeek 32B large model.

[0090] Step Six: Following the dataset organization method described in Step Five, collect the ESG annual report sentiment analysis dataset, use K-means clustering to obtain cluster centers, and use the corresponding data entries as the data required for the few-sample engineering.

[0091] Step 7: For the company's ESG annual report data, use the DeepSeek 32B large model based on fine-tuning, combined with few-sample engineering, to conduct sentiment analysis on the ESG annual report data. Sentences are divided by period, etc., and divided into four categories: positive, negative, neutral, and meaningless. The frequency of each type of sentence is counted, and the score is calculated according to the proportion of the frequency of each sentiment evaluation to the total number of sentences.

[0092] The main processes involved in the model training and prediction modules are as follows: Figure 5 As shown, the detailed steps are as follows:

[0093] Step 1: Once new data is introduced, it will be integrated into the dataset. Administrators can select which metrics to use for training and configure the dataset splitting ratio, experimental parameters, etc., on the training configuration page.

[0094] Step 2: After the administrator selects a suitable model, training can begin. The system can provide appropriate loss functions, such as RMSE, and use evaluation methods such as MAE and R2 to evaluate the training effect of the model.

[0095] Step 3: For the trained model, the administrator can choose to keep the trained model as a usable version and save the training configuration, model and other information.

[0096] Step 4: Administrators can import new models, such as other sparse representation models, import the model configuration information on the page, and then upload the model file that meets the input and output data format requirements in the backend.

[0097] For ex post-hoc interpretability models, the main processes involved are as follows: Figure 6 As shown, the detailed steps are as follows:

[0098] Step 1: Import the relevant data of the company to be predicted and identified into the system, select the desired model version, and import the corresponding data file.

[0099] Step 2: Begin prediction and identification; after prediction, obtain the enterprise's ESG rinsing behavior index.

[0100] Step 3: Calculate the minimum, maximum, and average values ​​of the ESG behavior index for all sample data in the current company's industry, as the data required for presentation.

[0101] Step 4: Use the SHAP algorithm to conduct post-hoc interpretability analysis on the prediction model, and obtain the distribution of the influence of each indicator on the model output results for visualization.

[0102] Step 5: Input the SHAP analysis results, prediction results, and current industry analysis status into the DeepSeek 32B large model based on the Prompt project. The large model will interpret the prediction and analysis results through preset prompts.

[0103] Step Six: Present the prediction results, SHAP analysis results, and large model interpretation results to users in a visual manner, providing easy-to-understand explanations of the results.

[0104] This application also provides an ESG rinsing behavior identification device. It should be noted that the ESG rinsing behavior identification device of this application can be used to execute the ESG rinsing behavior identification method provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0105] The following describes the ESG rinsing behavior identification device provided in the embodiments of this application.

[0106] Figure 7 This is a schematic diagram of an ESG rinsing behavior identification device according to an embodiment of this application. Figure 7 As shown, the device includes:

[0107] Acquisition unit 10 is used to acquire ESG evaluation data from multiple companies, including company annual report data and company operating data.

[0108] Specifically, the measurement method for a company's ESG washing behavior is based on the difference between its environmental information performance and its actual environmental rating performance, and the formula is as follows: ERi,t represents the acquired indicator data. This represents the average value of the indicators reflecting the true performance of the environmental rating. This represents the average value of the indicators reflecting the true performance of the environmental rating. The mean squared error of the indicator data representing the true performance of the environmental rating. The mean squared error of the indicator data representing the true performance of the environmental rating.

[0109] Training unit 20 is used to train the sparse representation model using ESG evaluation data to obtain an ESG rinsing behavior prediction model.

[0110] Specifically, ESG evaluation data, including corporate annual reports and operational data, was acquired, and an ESG washing behavior prediction model was constructed through training a sparse representation model. Due to the special nature of financial data, the financial data used for ESG score prediction is typically prepared tabular data. This tabular data is quite large in both length and width dimensions, containing a large amount of redundant information. This information redundancy can lead to performance degradation, and it is also difficult to capture important features and correlations between tabular data. Lista's sparse representation-based algorithm can reduce the dimensionality of the data to sparsity, thereby capturing key features while maintaining good interpretability and generalization performance.

[0111] Input unit 30 is used to input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index.

[0112] Specifically, the predictive model was used to analyze the ESG evaluation data of a specific company, generating a quantitative index that reflects its ESG washing behavior tendencies.

[0113] Analysis unit 40 is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model using the SHAP algorithm to obtain SHAP analysis results. The SHAP analysis results show the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index.

[0114] Specifically, the SHAP algorithm was used to conduct a post-hoc interpretability analysis of the prediction model, revealing the specific contributions of each ESG indicator to the rinsing behavior index, i.e., the distribution of its influence.

[0115] Display unit 50 is used to display the ESG rinsing behavior index and SHAP analysis results.

[0116] Specifically, the ESG rinsing behavior index and SHAP analysis results are presented to users in a visual way, thus providing the possibility of a deeper understanding of the prediction results.

[0117] In this embodiment, the ESG rinsing behavior identification device obtains an ESG rinsing behavior prediction model by training a sparse representation model. This model can then predict the ESG rinsing behavior of the enterprise to be identified, resulting in an ESG rinsing behavior index. Since the sparse representation model has good interpretability, the SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model, yielding SHAP analysis results. The distribution of the influence of each indicator on the ESG rinsing behavior index not only characterizes the overall ESG rinsing level of the enterprise to be identified based on the ESG rinsing behavior index but also displays the distribution of indicators affecting the ESG rinsing behavior index, improving the interpretability of the identification results and solving the problem of poor interpretability of ESG rinsing behavior identification results in existing technologies.

[0118] Furthermore, in this embodiment, the training unit includes:

[0119] The preprocessing subunit is used to preprocess ESG evaluation data from multiple enterprises to obtain preprocessed ESG evaluation data. The preprocessing includes missing value handling, logarithmic processing, normalization processing, and text symbol removal.

[0120] The analysis subunit is used to perform sentiment analysis on the pre-processed ESG evaluation data of each enterprise to obtain ESG evaluation data of each enterprise with multiple sentiment types, including positive, negative, neutral and meaningless.

[0121] The training subunit is used to train the sparse representation model using ESG evaluation data of multiple sentiment types from all enterprises, so as to obtain an ESG rinsing behavior prediction model.

[0122] In this embodiment, ESG evaluation data from multiple enterprises are preprocessed, specifically including missing value handling, logarithmic transformation, normalization, and removal of text symbols to ensure data consistency and reduce noise interference. Next, sentiment analysis is performed on the preprocessed ESG evaluation data from each enterprise, categorizing the data into four sentiment types: positive, negative, neutral, and meaningless. This aims to comprehensively understand the attitudes and tendencies implied in the enterprises' ESG annual reports. Subsequently, a sparse representation model is trained using ESG evaluation data from all enterprises across multiple sentiment types to obtain an ESG rinsing behavior prediction model. This process not only considers the diversity of data but also fully explores the emotional value behind the text. Through the learning of the sparse representation model, key features can be identified in high-dimensional data environments, thereby effectively identifying and predicting potential ESG rinsing behaviors of enterprises. Furthermore, the preprocessing and sentiment analysis steps in this embodiment improve the model's training effect, enhance the accuracy and stability of the prediction model, and provide a solid data foundation and technical support for subsequent ESG rinsing behavior identification. In other embodiments, the classification of sentiment types can also be fine-tuned using different deep learning models, such as the BERT model, which will further improve the accuracy of sentiment analysis and better serve the task of identifying ESG rinsing behaviors.

[0123] Furthermore, in this embodiment, the above-mentioned training subunit includes:

[0124] The input module is used to input ESG evaluation data of multiple sentiment types from all enterprises into the training configuration page. The training configuration page is used to select indicators for training and configure the split ratio of the dataset. The dataset consists of a training set and a validation set.

[0125] The first training module is used to train the sparse representation model through the training configuration page to obtain the ESG rinsing behavior prediction model.

[0126] In this embodiment, a sparse representation model is trained using ESG evaluation data of multiple sentiment types from all enterprises to obtain an ESG washout behavior prediction model. The ESG evaluation data of all sentiment types are configured through the training configuration page, allowing for indicator selection and dataset partitioning. The dataset is divided into training and validation sets. During this process, the administrator can flexibly adjust the training parameters according to actual conditions to optimize model performance. By training the sparse representation model through the training configuration page, the model can learn the characteristic patterns of enterprise ESG washout behavior from multi-source data, thereby constructing a prediction model. This design effectively handles high-dimensional data in the financial field, reduces data redundancy, and improves the model's generalization ability and interpretability. When new data is introduced, the prediction model can quickly adapt and accurately predict the degree of ESG washout tendency of enterprises, providing timely and reliable decision support for regulatory agencies and investors. In other embodiments, different machine learning or deep learning models can also be used for training, but the sparse representation model is the preferred solution of this invention due to its advantages in handling high-dimensional data and maintaining model interpretability. In this way, the present invention can meet the specific needs of the financial sector, improve the accuracy and credibility of predictive models, and, through the combination of SHAP and large-scale models, provide in-depth interpretation of the results, enhancing the persuasiveness and practicality of the predictions. Based on this, enterprises, regulators, and investors can more accurately understand and respond to ESG washout practices, promoting sustainable development practices.

[0127] Furthermore, in this embodiment, the first training module includes:

[0128] The training submodule is used to train the sparse representation model through the training configuration page until the loss function value meets the first predetermined condition, thereby obtaining the predetermined ESG rinsing behavior prediction model.

[0129] The determination submodule is used to determine the predetermined ESG rinsing behavior prediction model as an ESG rinsing behavior prediction model when the output of the predetermined ESG rinsing behavior prediction model meets the second predetermined condition corresponding to the evaluation method. The evaluation methods include the MAE evaluation method and the R2 evaluation method.

[0130] In this embodiment, the sparse representation model is trained through the training configuration page until the loss function value meets the first predetermined condition, resulting in a predetermined ESG washing behavior prediction model 2. During this process, by adjusting the model parameters and training dataset, the sparse representation model can effectively learn and extract key features from high-dimensional multi-source data to predict corporate ESG washing behavior tendencies. If the output of the predetermined ESG washing behavior prediction model 2 meets the second predetermined conditions corresponding to the MAE evaluation method and the R2 evaluation method, this model is determined as the final ESG washing behavior prediction model. This process ensures the model's accuracy and generalization ability, enabling it to provide reliable results when predicting ESG washing behavior. By setting specific evaluation methods, such as the MAE evaluation method to assess the mean absolute error between the model's prediction results and the actual values, and the R2 evaluation method to measure the model's ability to explain data variations, model performance can be effectively detected, thereby ensuring the credibility of the prediction results. As model training progresses, the loss function value gradually decreases, reaching the first predetermined condition, indicating that the model's prediction accuracy has been optimized, enabling better identification of corporate ESG washing behavior. Of course, in other embodiments not shown, other machine learning models or deep learning models may also be selected for training and prediction, but the high efficiency and interpretability advantages of sparse representation models in processing financial data will be used as evaluation criteria for other models to ensure that the final ESG washing behavior prediction model is not only accurate, but also interpretable and understandable.

[0131] Furthermore, in this embodiment, the above-mentioned analysis subunit includes:

[0132] The acquisition module is used to acquire pre-trained large models;

[0133] The second training module is used to train the pre-trained large model using ESG evaluation data from multiple enterprises and corresponding manually labeled sentiment types to obtain a sentiment analysis model.

[0134] The analysis module is used to input the preprocessed ESG evaluation data of each enterprise into the sentiment analysis model for sentiment analysis, and obtain ESG evaluation data of multiple sentiment types for each enterprise.

[0135] In this embodiment, sentiment analysis is performed on the preprocessed ESG evaluation data of various enterprises to obtain ESG evaluation data of multiple sentiment types for each enterprise. The specific implementation process is as follows: First, a pre-trained large model is obtained, which has basic semantic understanding and text classification capabilities. Second, the pre-trained large model is trained using ESG evaluation data from multiple enterprises and corresponding manually labeled sentiment types to obtain a sentiment analysis model. This process involves fine-tuning the model parameters to more accurately identify the sentiment tendencies contained in the ESG evaluation data. Finally, the preprocessed ESG evaluation data of each enterprise is input into the sentiment analysis model for sentiment analysis to obtain ESG evaluation data of multiple sentiment types for each enterprise. This large-model-based sentiment analysis method can effectively process text data such as corporate annual reports and announcements, accurately identify positive, negative, or neutral sentiment tendencies, and provide more comprehensive and in-depth data support for subsequent ESG washing behavior identification, improving the accuracy and reliability of identification. At the same time, through fine-tuning and manually labeled sentiment types, this sentiment analysis model can adapt to the characteristics of different industries and enterprises, enhancing its applicability and interpretability in the financial field. In other embodiments not shown in the figure, the sentiment analysis model can be fine-tuned using different pre-trained large models or by combining industry-specific corpora to further enhance the model's generalization ability and industry adaptability.

[0136] Furthermore, in this embodiment, the preprocessing subunit includes:

[0137] The first preprocessing module is used to process the numerical ESG evaluation data by handling missing values, logarithmic transformation, and normalization to obtain preprocessed ESG evaluation data.

[0138] The second preprocessing module is used to remove text symbols from text-based ESG evaluation data to obtain preprocessed ESG evaluation data.

[0139] In this embodiment, ESG evaluation data from multiple enterprises are preprocessed to obtain preprocessed ESG evaluation data. Specifically, this includes handling missing values, logarithmic transformation, and normalization for numerical ESG evaluation data to ensure data consistency and model training stability. For textual ESG evaluation data, text symbols are removed to eliminate interference from non-semantic information and improve the quality of the textual data. Preprocessing of numerical data, by filling in missing values ​​and transforming the data scale, makes the data more suitable for subsequent model training and prediction. Simultaneously, logarithmic transformation and normalization effectively avoid the impact of extreme values ​​in the data on model performance. For textual data, the removal of non-text symbols improves the accuracy of text analysis, ensuring the precision of sentiment analysis and semantic understanding. The entire preprocessing process aims to provide a clean and standardized dataset for the identification and analysis of enterprise ESG washing behaviors, which is beneficial for improving model training efficiency and prediction accuracy, while also laying a solid foundation for model interpretability and result reliability. In this embodiment, numerical ESG evaluation data undergoes preprocessing, including missing value handling, logarithmic transformation, and normalization, making the data more suitable for model training and prediction needs, reducing noise and bias in the data, and improving the model's generalization ability and prediction accuracy. Textual ESG evaluation data is processed by removing text symbols, eliminating redundant information and making the text data cleaner, which is beneficial for the accuracy of sentiment analysis and semantic understanding, further enhancing the model's recognition performance.

[0140] Furthermore, in this embodiment, the above-mentioned display unit includes:

[0141] The acquisition sub-unit is used to acquire the minimum, maximum, and average values ​​of the ESG rinsing behavior index of all sample data in the industry where the company to be identified is located. The sample data consists of the ESG evaluation data of companies in the industry where the company to be identified is located.

[0142] The input sub-unit is used to input the ESG rinsing behavior index, SHAP analysis results, and the minimum, maximum, and average values ​​of the ESG rinsing behavior index into the large model to obtain the large model interpretation results, which are the interpretations of the ESG rinsing behavior index and SHAP analysis results.

[0143] The display sub-unit is used to present the ESG rinsing behavior index, SHAP analysis results, and large model interpretation results.

[0144] In this embodiment, when displaying the enterprise's ESG washing behavior index and SHAP analysis results, we obtained the minimum, maximum, and average values ​​of the ESG washing behavior index from all sample data within the industry of the enterprise to be identified. The sample data includes ESG evaluation data from enterprises in that industry. By inputting the ESG washing behavior index, SHAP analysis results, and these statistical values ​​into a large model based on the Prompt project, we can obtain the large model interpretation results, i.e., an in-depth analysis of the ESG washing behavior index and SHAP analysis results. Visualizing the ESG washing behavior index, SHAP analysis results, and large model interpretation results provides users with intuitive understanding and decision-making support, realizing a full-process service from data input and model prediction to result interpretation. This process not only improves the transparency of model predictions but also enhances the interpretability of results through large model analysis, enabling non-professional users to understand the logic behind model predictions, thereby improving the accuracy and fairness of decision-making. In other embodiments not shown in the figure, different visualization techniques can be employed to further enhance the user's understanding of the data. For example, a heatmap can be used to display SHAP values, with the intensity of color visually reflecting the degree of influence of each feature on the prediction results. Alternatively, a bar chart can be used to compare the ESG rinsing behavior index with the industry average, helping users quickly pinpoint a company's relative performance. These alternatives can also achieve the invention's objectives, providing diverse and personalized results presentation methods.

[0145] The aforementioned ESG rinsing behavior identification device includes a processor and a memory. The acquisition unit, training unit, input unit, analysis unit, and display unit are all stored as program units in the memory. The processor executes these program units stored in the memory to achieve the corresponding functions. All of the above modules are located in the same processor; alternatively, the modules may be located in different processors in any combination.

[0146] The processor contains a kernel, which retrieves the corresponding program unit from memory. One or more kernels can be configured, and adjusting kernel parameters can address the poor interpretability of ESG rinsing behavior recognition results in existing technologies.

[0147] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0148] This invention provides a computer-readable storage medium including a stored program, wherein the program, when running, controls the device containing the computer-readable storage medium to perform the ESG rinsing behavior.

[0149] Specifically, the methods for identifying ESG rinsing behavior include:

[0150] Step S201: Obtain ESG evaluation data from multiple companies, including company annual report data and company operating data;

[0151] Step S202: Train the sparse representation model using ESG evaluation data to obtain the ESG rinsing behavior prediction model.

[0152] Step S203: Input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index.

[0153] Step S204: The SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model to obtain the SHAP analysis results. The SHAP analysis results show the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index.

[0154] Step S205: Display the ESG rinsing behavior index and SHAP analysis results.

[0155] This invention provides a processor for running a program, wherein the program executes the method for identifying ESG rinsing behavior.

[0156] Specifically, the methods for identifying ESG rinsing behavior include:

[0157] Step S201: Obtain ESG evaluation data from multiple companies, including company annual report data and company operating data;

[0158] Step S202: Train the sparse representation model using ESG evaluation data to obtain the ESG rinsing behavior prediction model.

[0159] Step S203: Input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index.

[0160] Step S204: The SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model to obtain the SHAP analysis results. The SHAP analysis results show the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index.

[0161] Step S205: Display the ESG rinsing behavior index and SHAP analysis results.

[0162] This invention provides an ESG assessment system, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0163] Step S201: Obtain ESG evaluation data from multiple companies, including company annual report data and company operating data;

[0164] Step S202: Train the sparse representation model using ESG evaluation data to obtain the ESG rinsing behavior prediction model.

[0165] Step S203: Input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index.

[0166] Step S204: The SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model to obtain the SHAP analysis results. The SHAP analysis results show the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index.

[0167] Step S205: Display the ESG rinsing behavior index and SHAP analysis results.

[0168] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0169] Step S201: Obtain ESG evaluation data from multiple companies, including company annual report data and company operating data;

[0170] Step S202: Train the sparse representation model using ESG evaluation data to obtain the ESG rinsing behavior prediction model.

[0171] Step S203: Input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index.

[0172] Step S204: The SHAP algorithm is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model to obtain the SHAP analysis results. The SHAP analysis results show the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index.

[0173] Step S205: Display the ESG rinsing behavior index and SHAP analysis results.

[0174] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0175] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0176] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0177] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0178] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0179] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0180] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0181] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0182] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0183] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0184] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0185] 1) In the ESG rinsing behavior identification method of this application, an ESG rinsing behavior prediction model is obtained by training a sparse representation model. The ESG rinsing behavior prediction model can then be used to predict the ESG rinsing behavior of the enterprise to be identified, and obtain the ESG rinsing behavior index. Since the sparse representation model has good interpretability, the ESG rinsing behavior prediction model is subjected to ex-post interpretability analysis using the SHAP algorithm to obtain the SHAP analysis results. The distribution of the influence of each indicator on the ESG rinsing behavior index can not only characterize the overall ESG rinsing degree of the enterprise to be identified based on the ESG rinsing behavior index, but also show the distribution of indicators affecting the ESG rinsing behavior index, thereby improving the interpretability of the identification results and solving the problem of poor interpretability of the ESG rinsing behavior identification results in the prior art.

[0186] 2) In the ESG rinsing behavior identification device of this application, an ESG rinsing behavior prediction model is obtained by training a sparse representation model. The ESG rinsing behavior prediction model can then be used to predict the ESG rinsing behavior of the enterprise to be identified, and obtain the ESG rinsing behavior index. Since the sparse representation model has good interpretability, the ESG rinsing behavior prediction model is subjected to ex-post interpretability analysis using the SHAP algorithm to obtain the SHAP analysis results. The distribution of the influence of each indicator on the ESG rinsing behavior index can not only characterize the overall ESG rinsing degree of the enterprise to be identified based on the ESG rinsing behavior index, but also show the distribution of indicators affecting the ESG rinsing behavior index, thereby improving the interpretability of the identification results and solving the problem of poor interpretability of the ESG rinsing behavior identification results in the prior art.

[0187] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for identifying ESG rinsing behavior, characterized in that, include: Acquire ESG evaluation data from multiple companies, including company annual report data and company operating data; The sparse representation model is trained using the ESG evaluation data to obtain an ESG rinsing behavior prediction model. Input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index. The SHAP algorithm was used to perform a post-hoc interpretability analysis on the ESG rinsing behavior prediction model to obtain the SHAP analysis results. The SHAP analysis results show the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index. The ESG rinsing behavior index and the SHAP analysis results are presented.

2. The method according to claim 1, characterized in that, The sparse representation model is trained using the ESG evaluation data to obtain an ESG rinsing behavior prediction model, including: The ESG evaluation data of multiple enterprises are preprocessed to obtain preprocessed ESG evaluation data. The preprocessing includes missing value processing, logarithmic processing, normalization processing, and text symbol removal processing. Sentiment analysis is performed on the preprocessed ESG evaluation data of each enterprise to obtain ESG evaluation data of each enterprise with multiple sentiment types, including positive, negative, neutral and meaningless. The sparse representation model is trained using ESG evaluation data of multiple sentiment types from all the aforementioned enterprises to obtain the ESG rinsing behavior prediction model.

3. The method according to claim 2, characterized in that, The sparse representation model is trained using ESG evaluation data of multiple sentiment types from all the aforementioned enterprises to obtain the ESG rinsing behavior prediction model, including: Input the ESG evaluation data of multiple sentiment types of all the enterprises into the training configuration page. The training configuration page is used to select the indicators for training and configure the division ratio of the dataset. The dataset consists of a training set and a validation set. The sparse representation model is trained through the training configuration page to obtain the ESG rinsing behavior prediction model.

4. The method according to claim 3, characterized in that, The sparse representation model is trained through the training configuration page to obtain the ESG rinsing behavior prediction model, including: The sparse representation model is trained through the training configuration page until the loss function value meets the first predetermined condition, thereby obtaining the predetermined ESG rinsing behavior prediction model. If the output of the predetermined ESG rinsing behavior prediction model meets the second predetermined condition corresponding to the evaluation method, the predetermined ESG rinsing behavior prediction model is determined as the ESG rinsing behavior prediction model. The evaluation method includes the MAE evaluation method and R... 2 Evaluation methods.

5. The method according to claim 2, characterized in that, Sentiment analysis is performed on the preprocessed ESG evaluation data of each enterprise to obtain ESG evaluation data of each enterprise in multiple sentiment types, including: Obtain a pre-trained large model; The pre-trained large model is trained using ESG evaluation data from multiple companies and corresponding manually labeled sentiment types to obtain a sentiment analysis model. The preprocessed ESG evaluation data of each enterprise is input into the sentiment analysis model for sentiment analysis, resulting in ESG evaluation data of multiple sentiment types for each enterprise.

6. The method according to claim 2, characterized in that, The ESG evaluation data of multiple companies are preprocessed to obtain preprocessed ESG evaluation data, including: The numerical ESG evaluation data is processed by missing value handling, logarithmic processing, and normalization to obtain the preprocessed ESG evaluation data. The text-based ESG evaluation data is processed by removing text symbols to obtain the preprocessed ESG evaluation data.

7. The method according to any one of claims 1 to 6, characterized in that, The ESG rinsing behavior index and the SHAP analysis results are presented, including: Obtain the minimum, maximum, and average values ​​of the ESG rinsing behavior index for all sample data of the industry in which the enterprise to be identified is located, where the sample data is the ESG evaluation data of the enterprises in the industry in which the enterprise to be identified is located. The ESG rinsing behavior index, the SHAP analysis results, and the minimum, maximum, and average values ​​of the ESG rinsing behavior index are input into the large model to obtain the large model interpretation results, which are the interpretations of the ESG rinsing behavior index and the SHAP analysis results. The ESG rinsing behavior index, the SHAP analysis results, and the large model interpretation results are presented.

8. A device for identifying ESG rinsing behavior, characterized in that, include: The acquisition unit is used to acquire ESG evaluation data from multiple enterprises, including enterprise annual report data and enterprise operating data. The training unit is used to train the sparse representation model using the ESG evaluation data to obtain an ESG rinsing behavior prediction model. The input unit is used to input the ESG evaluation data of the enterprise to be identified into the ESG rinsing behavior prediction model to obtain the ESG rinsing behavior index. The analysis unit is used to perform post-hoc interpretability analysis on the ESG rinsing behavior prediction model using the SHAP algorithm to obtain SHAP analysis results. The SHAP analysis results are the distribution of the influence of each index of the ESG evaluation data on the ESG rinsing behavior index. The display unit is used to display the ESG rinsing behavior index and the SHAP analysis results.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the ESG rinsing behavior identification method according to any one of claims 1 to 7.

10. An ESG evaluation system, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including a method for performing an identification method for performing an ESG rinsing behavior as described in any one of claims 1 to 7.

Images