Enterprise ESG risk assessment system and method

Abstract

The invention discloses an enterprise ESG risk assessment system and method, and the method comprises the steps: collecting environment assessment data, management related files and enterprise social media data of an enterprise, and collecting social assessment quantitative data; dividing the enterprise social media data into social related data and governance related data by using a data classification model, and obtaining a comment positive-negative ratio of the social related data based on an emotion dictionary; according to a formula, calculating to obtain an environmental risk index and a social risk index, and according to the text analysis large model, obtaining a governance risk index; obtaining the weight of each risk index by using an analytic hierarchy process, and obtaining an ESG evaluation index through a weighted summation method; judging whether the risk indexes and the ESG evaluation indexes exceed corresponding threshold values or not; if yes, sending an early warning signal; and if not, sending a normal signal. The invention relates to the technical field of enterprise risk assessment, and solves the technical problem that the assessment result is not accurate enough due to the fact that traditional ESG assessment data is difficult to quantify.

Description

Technical Field

[0001] The present invention belongs to the technical field of enterprise risk assessment, and specifically relates to an enterprise ESG risk assessment system and method. Background Art

[0002] In today's society, as the concept of sustainable development continues to deepen, the environmental, social and corporate governance (ESG) performance of enterprises has received more and more attention. ESG risk assessment is crucial to the long-term and stable development of enterprises. It can not only help enterprises identify potential environmental and social risks, formulate response strategies in advance, and reduce economic losses caused by environmental damage, lack of social responsibility and other problems, but also enhance the corporate image, enhance investor confidence, and attract more long-term capital investment.

[0003] However, the existing ESG risk assessment methods lack scientific and effective quantitative means in the assessment process. For many indicators in the environmental and social dimensions, such as employee satisfaction, community influence, and the actual effects of corporate environmental protection measures, they are mostly based on qualitative descriptions, making it difficult to conduct accurate quantitative analysis. This qualitative assessment method makes the ESG assessment results of different companies lack comparability, because the subjective judgment of the assessors and differences in evaluation criteria can easily introduce large deviations. At the same time, due to the lack of support from quantitative data, it is difficult for companies to accurately identify their own risk levels and improvement directions in ESG, and it is impossible to formulate targeted strategies to improve ESG performance. It also makes it difficult for investors and other stakeholders to make reasonable decisions based on objective and accurate assessments. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an enterprise ESG risk assessment system and method, which are used to solve the technical problem that traditional ESG assessment data is difficult to quantify and process, resulting in inaccurate assessment results.

[0005] To achieve the above objectives, a first aspect of the present invention provides a risk assessment system based on enterprise ESG, comprising: Data collection module: used to collect the environmental assessment data of enterprises, manage relevant documents and enterprise social media data, and collect social assessment quantitative data; wherein the enterprise social media data is obtained through crawler technology; Data processing module: used to divide enterprise social media data into social-related data and governance-related data using a data classification model, and obtain the positive and negative ratio of comments on social-related data based on the sentiment dictionary; Risk assessment module: used to calculate environmental risk index and social risk index based on formulas from environmental assessment data and social assessment data, and to obtain governance risk index based on text analysis big model and management-related documents and governance-related data; and using the analytic hierarchy process to obtain the weights of the environmental risk index, social risk index and governance risk index, and obtaining the ESG evaluation index through the weighted summation method; wherein the social evaluation data includes social evaluation quantitative data and the positive and negative ratio of comments; Risk warning module: determines whether the environmental risk index, social risk index, governance risk index and ESG assessment index exceed the corresponding preset thresholds; if yes, a warning signal is sent; if not, a normal signal is sent.

[0006] Based on the above technical modules, the present invention realizes comprehensive, detailed and efficient risk assessment and early warning for enterprises under the ESG framework. Specifically, the present invention realizes the collection of data from multiple channels and conducts in-depth mining and classification processing to form a multi-dimensional and interrelated risk index, covering key areas such as environment, society, and governance, and then comprehensively considers various factors to obtain an objective and accurate ESG evaluation index. At the same time, it can monitor the relationship between these indexes and preset thresholds in real time, and issue early warning signals or normal signals in a timely manner, providing strong support for enterprises to prevent and control risks in advance, optimize operating strategies, and make reasonable decisions for investors and other relevant entities, thereby improving the scientificity and foresight of enterprise risk management.

[0007] Furthermore, the environmental assessment data includes the total carbon emissions, total power generation, total energy consumption, energy conversion efficiency, total water use, water recycling rate, total wastewater discharge and wastewater discharge compliance rate of the enterprise within a preset period; The management-related documents include the company's articles of association, minutes of board meetings, strategic planning documents, internal management system manuals, financial reports and audit reports, compliance policy documents, human resources management documents and information disclosure documents; The social assessment quantitative data include product quality complaint rate, product return rate, employee turnover rate and community project participation rate.

[0008] Furthermore, the data classification model is constructed based on a deep learning algorithm, including: A1, crawling enterprise social media data using crawler technology and preprocessing to obtain text dataset; A2, classify several paragraphs in the text dataset to obtain several paragraph annotations, and use the paragraph vector method based on word embedding to map several paragraphs into the vector space to obtain several paragraph vectors; where paragraph annotations are represented by 0 and 1, 0 represents social category, and 1 represents governance category; A3, build a neural network model based on deep learning, and set the training parameters, loss function and optimizer of the neural network model to obtain the initial model; A4: Divide a number of paragraph annotations and a number of paragraph vectors into a training set, a validation set, and a test set according to a preset ratio, input the training set and the validation set into the initial model for iterative training and update optimization, and obtain the optimal performance model based on the validation accuracy; A5: Input the test set into the best performing model for evaluation, obtain the test accuracy, and determine whether the test accuracy reaches the preset accuracy threshold; if yes, mark the best performing model as the data classification model; if no, adjust the training parameters, loss function, and optimizer to A4.

[0009] When collecting data, since corporate social media data comes from a wide range of sources and the amount of data is huge, in order to make full use of it, it is necessary to classify and sort it out to extract effective information at the social and governance levels. However, manual classification is inefficient, and it is time-consuming and laborious to judge the category sentence by sentence in the face of massive data. Therefore, a data classification model based on deep learning algorithms is designed to use its automation and intelligence advantages to efficiently and accurately divide data and provide high-quality classified data for ESG risk assessment.

[0010] Furthermore, the positive and negative ratio of comments of social related data obtained based on the sentiment dictionary includes: B1, preprocessing social related data to obtain preprocessed text; B2, classifying and assigning values ​​to a number of words in the preprocessed text to obtain a sentiment dictionary containing a number of positive words, a number of negative words and corresponding scores; B3, based on the sentiment dictionary, obtain the sentiment tendency of several sentences in the preprocessed text, and calculate the ratio of several positive sentences to several negative sentences in the sentiment tendency to obtain the positive and negative ratio of the comments.

[0011] In the present invention, by calculating the positive and negative ratio of comments based on the sentiment dictionary, the sentiment tendency in the social related data is quantified, and the overall emotional attitude of the public towards the company is displayed in a concise and clear way; and as an important indicator of the social dimension under the ESG framework, it can provide a unique perspective for comprehensive risk assessment from the perspective of social reputation and public cognition.

[0012] Furthermore, the classification and assignment of several words of the preprocessed text includes: B2-1, select a number of initial words for classification and assignment, and obtain seed words of positive and negative categories and corresponding first scores; B2-2, perform word segmentation on the preprocessed text based on the N-gram model to obtain several words; B2-3. Use a word vector model to convert a number of words into a number of word vectors, and use cosine similarity to calculate the similarity values between the number of words and the seed words, obtaining a number of positive similarities and a number of negative similarities; B2-4. According to a preset assignment rule, classify and assign values to the number of words based on the number of positive similarities and the number of negative similarities.

[0013] Based on the seed words, with the help of natural language processing technology, it is possible to continuously expand the vocabulary range covered by the sentiment dictionary, and determine the words belonging to the sentiment dictionary by calculating the similarity between the words and the seed words, so that it not only includes the initially set limited vocabulary, but also can incorporate a large number of related and semantically similar words in the text, enhancing the adaptability of the dictionary to various text expressions.

[0014] Further, the preset assignment rule includes: B2-41. Set a first similarity threshold A, a second similarity threshold B, and a third similarity threshold C, and when the similarity belongs to [0, A), assign a value of 0, when the similarity belongs to [A, B), assign a third score value, and when the similarity belongs to [B, C], assign a second score value, obtaining a similarity assignment table; where A < B < C, 0 < the third score value < the second score value < the first score value; B2-42. For each current word, screen out the highest values among the number of positive similarities and the number of negative similarities, obtaining the highest positive degree and the highest negative degree; B2-43. Determine whether the highest positive degree is greater than the highest negative degree; if yes, mark the current word as a positive word and jump to B2-44; if no, mark the current word as a negative word and jump to B2-44; B2-44. Assign a value to the current word according to the similarity assignment table, obtaining the corresponding score value.

[0015] In the sentiment dictionary, simply distinguishing between positive and negative categories is not sufficient to fully explore the emotional value of words. Therefore, it is necessary to quantify this emotional degree difference by setting multiple thresholds and corresponding score values. Different similarity thresholds correspond to different score values, which can reflect the strength of the emotional tendency of words, more delicately reflect the emotional information conveyed by words in the text, enrich the connotation of the sentiment dictionary, and contribute to more accurately analyzing the overall emotional tendency of sentences subsequently.

[0016] Further, obtaining the emotional tendency of several sentences in the preprocessed text based on the sentiment dictionary includes: B3-1. Perform word segmentation on several sentences in the preprocessed text based on the N-gram model, obtaining several segmented words; B3-2. Match the several segmented words in each sentence with the sentiment dictionary, obtaining several matching words and several corresponding score values; B3-3, summing up the corresponding scores of the positive words in the matching words, and subtracting the sum of the corresponding scores of the negative words to obtain a sentiment score; B3-4, mark the sentences with sentiment scores greater than 0 as positive sentences; mark the sentences with sentiment scores equal to 0 as neutral sentences; mark the sentences with sentiment scores less than 0 as negative sentences.

[0017] By matching the sentiment dictionary word by word and calculating the score, the overall impact of each sentiment-inclined word in the sentence is fully considered, avoiding the one-sidedness and inaccuracy of judging the sentiment of the sentence only from part of the words or subjective assumptions, making the judgment of the sentiment tendency of the sentence more scientific, objective and in line with the actual semantics.

[0018] Furthermore, the environmental risk index and social risk index are calculated according to the formula, including: C1, divide several indicators in environmental assessment data and social assessment data to obtain positive indicators E + and negative indicator E - ; Among them, the positive indicators include energy conversion efficiency, water recycling rate, total power generation, wastewater discharge compliance rate, community project participation rate and positive and negative ratio of comments; the negative indicators include total carbon emissions, total energy consumption, total water use, total wastewater discharge, product quality complaint rate, product return rate and employee turnover rate; C2, according to the formula ERI=[∑ 4 i=1 (w i ×(E - i / AE - ))] / [∑ 4 j=1 (w j ×(E + j / AE + ))+∑ 4 i=1 (w i ×(E - i / AE - ))] to calculate the environmental risk index ERI; where w i and w j represents the indicator weight, E - i and E + j Respectively represent the negative and positive indicators in environmental assessment data, AE - and AE + They represent the mean of negative indicators and the mean of positive indicators in environmental assessment data respectively; C3, according to the formula SRI=[∑ 3 p=1 (w p ×(E - p / BE - ))] / [∑ 2 q=1 (w q ×(E + q / BE + ))+∑ 3 p=1 (w p ×(E - p / BE - ))] to calculate the social risk index SRI; where w p and w q represents the indicator weight, E - p and E + q Represent the negative and positive indicators in social evaluation data, BE - and BE + They represent the mean of negative indicators and the mean of positive indicators in social evaluation data respectively.

[0019] In the field of enterprise risk assessment, in order to enable different enterprises to be compared and analyzed based on unified standards, various types of data need to be converted into standardized risk indexes; and different indicators have different ways and degrees of impact on risk, some of which are positive in promoting risk reduction, while others are negative in increasing risk. By classifying and assigning corresponding calculation methods, the characteristics of these indicators can be accurately reflected, and the real role of environmental and social factors in the composition of enterprise risks can be more scientifically reflected.

[0020] Furthermore, the governance risk index is obtained based on the text analysis big model and management-related documents and governance-related data, including: D1, converting management-related documents and governance-related data into plain text format data to obtain text data; D2, determining several dimensions of enterprise risk assessment, and inputting the text data, the several dimensions and the first instruction statement into a text analysis big model to obtain risk points of the several dimensions and the frequency of occurrence of the risk points; wherein the text analysis big model represents an existing language generation big model; D3, input the risk points of several dimensions, text data and the second instruction statement into the text analysis model to obtain the severity score of the risk point; D4, assign weights to several dimensions according to the nature of the enterprise, and use the formula GRI=∑ n=1∑ m=1 (S mn ×F mn )×W n The governance risk index is calculated; where S mn Indicates the severity score of the mth risk point in the nth dimension, F mn represents the occurrence frequency of the mth risk point in the nth dimension, W n Represents the weight of the nth dimension.

[0021] Most data related to corporate governance exist in text form and are complex in content. The use of text analysis big models can leverage their natural language processing capabilities to quickly and comprehensively dig out deep-level risk information and severity, effectively quantify governance risks, and make risk assessment results more convincing and practical.

[0022] A first aspect of the present invention provides a risk assessment method based on enterprise ESG, comprising: S1, collect the environmental assessment data, management-related documents and corporate social media data of the enterprise, and collect social assessment quantitative data; wherein the corporate social media data is obtained through crawler technology; S2, using the data classification model to divide the enterprise social media data into social-related data and governance-related data, and obtain the positive and negative ratio of comments on social-related data based on the sentiment dictionary; S3, calculating the environmental risk index and social risk index by using the environmental assessment data and social assessment data according to the formula, and obtaining the governance risk index according to the text analysis big model and the management-related documents and governance-related data; wherein the social assessment data includes the social assessment quantitative data and the positive and negative ratio of comments; S4, using the analytic hierarchy process to obtain the weights of the environmental risk index, social risk index and governance risk index, and using the weighted sum method to obtain the ESG assessment index; S5, determines whether the environmental risk index, social risk index, governance risk index and ESG assessment index exceed the corresponding preset thresholds; if yes, sends a warning signal; if not, sends a normal signal.

[0023] Compared with the prior art, the present invention has the following beneficial effects: Through comprehensive data collection and in-depth analysis, the present invention constructs a risk indicator system covering multiple dimensions of environment, society, and governance. In terms of environment, key indicators such as total carbon emissions and energy conversion efficiency are considered; in terms of society, factors such as product quality complaint rate, employee turnover rate, and positive and negative ratio of social media comments are included; in terms of governance, risk points and corresponding indexes are determined by analyzing management-related documents such as company articles of association and minutes of board meetings; it can accurately reflect the real risk status of the enterprise, improve the effectiveness of early warning, and ensure that the enterprise can accurately identify potential risks; The data processing module uses deep learning algorithms to build a data classification model to accurately classify corporate social media data. It also builds a sentiment dictionary to analyze the positive and negative ratio of comments in social-related data, evaluates the image and reputation of the company in the public's mind, and increases the comprehensiveness and reliability of the social dimension in ESG assessment. At the same time, it combines the company's management-related documents and governance-related data for analysis, comprehensively evaluates the company's internal management and governance structure, makes governance risk assessment more comprehensive, and thus improves the overall comprehensiveness of ESG assessment; When calculating the governance risk index, the risk assessment module uses the text analysis model to determine the severity and frequency of risk points in different dimensions of governance; in addition, the analytic hierarchy process is used to determine the weights of each risk index, and the final ESG assessment index is obtained through the weighted summation method. These methods fully consider the complexity and interrelationships of risks in different dimensions, ensuring that the assessment results are scientific and reasonable and can objectively reflect the ESG risk level of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 A schematic diagram of the framework of an enterprise ESG risk assessment system provided by the present invention; Figure 2 A flowchart of an enterprise ESG risk assessment method provided by the present invention; Figure 3 A schematic diagram of a flow chart for obtaining a positive-negative ratio of comments in a data processing module provided by the present invention; Figure 4 A schematic diagram of a flow chart for determining the sentiment tendency of a sentence provided by the present invention. DETAILED DESCRIPTION

[0026] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] See also Figure 1-Figure 4 The first aspect of the present invention provides an enterprise ESG risk assessment system, including: Data collection module: used to collect the company's environmental assessment data, manage relevant documents and corporate social media data, and collect social assessment quantitative data; among them, corporate social media data is obtained through crawler technology; Data processing module: used to divide enterprise social media data into social-related data and governance-related data using a data classification model, and obtain the positive and negative ratio of comments on social-related data based on the sentiment dictionary; Risk assessment module: used to calculate environmental risk index and social risk index based on formulas from environmental assessment data and social assessment data, and to obtain governance risk index based on text analysis big model and management-related documents and governance-related data; The weights of the environmental risk index, social risk index and governance risk index are obtained by using the analytic hierarchy process, and the ESG evaluation index is obtained by the weighted summation method; among which, the social evaluation data includes the quantitative data of social evaluation and the positive and negative ratio of comments; Risk warning module: determines whether the environmental risk index, social risk index, governance risk index and ESG assessment index exceed the corresponding preset thresholds; if yes, a warning signal is sent; if not, a normal signal is sent.

[0028] It should be noted that the data acquisition module, data processing module, risk assessment module and risk warning module of the present invention are communicatively connected, and the sentiment dictionary, data classification model and text analysis model are built into the core computing unit jointly constituted by the data processing module and the risk assessment module.

[0029] In order to comprehensively evaluate the comprehensive performance of enterprises in multiple dimensions such as environmental protection, internal governance and social influence, the data collection module of the present invention comprehensively collects the environmental assessment data, management-related documents, enterprise social media data and social assessment quantitative data of the enterprise; among them, the enterprise social media data is obtained through crawler technology; the environmental assessment data includes but is not limited to the total carbon emissions, total power generation, total energy consumption, energy conversion efficiency, total water use, water recycling rate, total wastewater discharge and wastewater discharge compliance rate of the enterprise within a preset period; management-related documents include but are not limited to the company's articles of association, board meeting minutes, strategic planning documents, internal management system manuals, financial reports and audit reports, compliance policy documents, human resources management documents and information disclosure documents; social assessment quantitative data includes but is not limited to product quality complaint rate, product return rate, employee turnover rate and community project participation rate.

[0030] It should be noted that the core purpose of the present invention to obtain corporate social media data through crawler technology is to comprehensively assess the social dimension risks faced by the enterprise. During the acquisition process, no personal privacy data will be collected, only public information directly related to the enterprise will be collected, and strict filtering and avoidance measures will be taken for data that may involve personal identification and sensitive information. In addition, the acquired data will only be used for research, analysis and model building related to corporate risk assessment, and will not be used for any profit-making purposes, will not be sold to third parties, and will not be used for commercial marketing activities or data transactions. It is only used to provide risk assessment reports for enterprises to help improve risk management levels.

[0031] Then in the data processing module, the data classification model in the core computing unit is first called to classify the collected corporate social media data to obtain social-related data and governance-related data. Then, the social-related data is analyzed based on the constructed sentiment dictionary to obtain the positive and negative ratio of comments.

[0032] Among them, the data classification model is built based on the deep learning algorithm, and its construction process is as follows: A1, crawling enterprise social media data using crawler technology and preprocessing to obtain text dataset; A2, classify several paragraphs in the text dataset to obtain several paragraph annotations, and use the paragraph vector method based on word embedding to map several paragraphs into the vector space to obtain several paragraph vectors; where paragraph annotations are represented by 0 and 1, 0 represents social category, and 1 represents governance category; A3, build a neural network model based on deep learning, and set the training parameters, loss function and optimizer of the neural network model to obtain the initial model; A4: Divide a number of paragraph annotations and a number of paragraph vectors into a training set, a validation set, and a test set according to a preset ratio, input the training set and the validation set into the initial model for iterative training and update optimization, and obtain the optimal performance model based on the validation accuracy; A5: Input the test set into the best performing model for evaluation, obtain the test accuracy, and determine whether the test accuracy reaches the preset accuracy threshold; if yes, mark the best performing model as the data classification model; if no, adjust the training parameters, loss function, and optimizer to A4.

[0033] Corporate social media data covers a wide range of content and a large amount of data, including social-related information such as corporate social responsibility activities and public welfare participation, as well as governance-related matters such as strategic decision-making and management changes. Traditional methods are inefficient and difficult to guarantee accuracy when processing massive amounts of data. Deep learning algorithms, with the help of parallel computing and optimized training algorithms, can quickly process large amounts of data and complete the classification task of large amounts of social media data in a short period of time, providing system processing efficiency and timeliness of risk assessment.

[0034] The construction process of the sentiment dictionary of the present invention is as follows: Data preprocessing stage (B1): Before starting to build the sentiment dictionary, the social data is first preprocessed, including removing noise from the data, such as irrelevant special characters, HTML tags, etc. At the same time, the text is standardized, all words are unified into lowercase, and stem extraction or word form restoration operations may be performed to obtain preprocessed text; Seed vocabulary determination (B2-1): Predetermine a number of initial words with clear emotional tendencies, such as "excellent" and "outstanding" are classified as positive categories, and "bad" and "awful" are classified as negative categories. Each seed word is assigned a first score to reflect the emotional intensity of the word. The higher the score, the stronger the emotional tendency. For example, the positive seed word "excellent" and the negative seed word "bad" are both assigned a first score of 5; Word segmentation (B2-2): Use the N-gram model to perform word segmentation on the preprocessed text. The N-gram model can divide the text into word sequences of different lengths. For example, when N=1, the text will be segmented into single words; when N=2, it will generate two-word combinations, such as "enterprise development", etc.; and obtain several words; Word vector conversion and similarity calculation (B2-3): With the help of a word vector model, several words obtained by word segmentation are converted into corresponding word vectors; then, the cosine similarity formula is used to calculate the similarity values of each word with positive and negative seed words. For example, for the word "efficient", its similarity with the positive seed word "excellent" is calculated through cosine similarity to obtain a positive similarity value; at the same time, its similarity with the negative seed word "terrible" is calculated to obtain a negative similarity value; several positive similarities and several negative similarities corresponding to all words are obtained; Word classification and assignment (B2-4): First, set the similarity threshold and assignment table (B2-41): Set three different similarity thresholds, namely the first similarity threshold A, the second similarity threshold B, and the third similarity threshold C, and satisfy A < B < C. According to these thresholds, a similarity assignment table is formulated. When the similarity is in the interval [0, A), the corresponding word is assigned 0, indicating that the word has no emotional tendency; when the similarity is in the interval [A, B), it is assigned the third score, indicating that the word has a certain degree of emotional tendency; when the similarity is in the interval [B, C], it is assigned the second score, meaning that the emotional tendency of the word is stronger. For example, assume A = 0.2, B = 0.4, C = 0.6, the third score is 1, the second score is 3, and the first score is 5; Then, determine the word emotional category (B2-42, B2-43): For each current word, find the highest values among its similarities with positive and negative seed words, namely the highest positive degree and the highest negative degree. Then compare these two values. If the highest positive degree is greater than the highest negative degree, mark the current word as a positive word; otherwise, mark it as a negative word. For example, for the word "innovative", its highest positive degree is 0.5 and the highest negative degree is 0.1. Since 0.5 > 0.1, "innovative" is marked as a positive word; Next, perform word assignment (B2-44): According to the similarity assignment table, assign values to the current words whose emotional categories have been marked. For example, for the above-mentioned "innovative" marked as a positive word, its highest positive degree 0.5 is in the interval [B, C] (assuming B = 0.4, C = 0.6). According to the assignment table, it will be assigned the second score, which is 3; finally, an emotional dictionary containing several positive words, several negative words, and their corresponding scores is obtained.

[0035] In the present invention, a number of words in the enterprise social media data are converted into word vectors through a vector model, and the cosine similarity formula is used to calculate the similarity value with the seed word, quantify the semantic association between words, accurately discover words with similar sentiment tendencies to the seed words, limit and expand the coverage of sentiment words, and allow more words to be accurately classified into corresponding sentiment categories; at the same time, different similarity thresholds and assignment tables are set to more carefully reflect the strength of the sentiment tendency of words, so that the sentiment dictionary can quantify the sentiment of words more scientifically and reasonably.

[0036] In this embodiment, obtaining the positive and negative ratio of comments in social related data based on the constructed sentiment dictionary includes the following process: First, determine the sentiment tendency of several sentences in social related data (B3): Sentence segmentation (B3-1): Use the N-gram model to segment several sentences in the preprocessed text to obtain several segmented words; Vocabulary matching and score acquisition (B3-2): Compare the word segments in each sentence with the constructed sentiment dictionary one by one, find the matching words, and obtain their corresponding scores. For example, for the sentence "This company's product quality is great", the word segments are "this" "company" "of" "product" "quality" "great", among which "great" is marked as a positive word in the sentiment dictionary, assuming the corresponding score is 3; Calculate the sentiment score (B3-3): Add the corresponding scores of all positive words in the sentence, and then subtract the sum of the corresponding scores of negative words to get the sentiment score of the sentence. For example, if the sum of the positive word scores in a sentence is 5 and the sum of the negative word scores is 0, then the sentiment score of the sentence is 5; Mark sentence sentiment tendency (B3-4): Determine the sentiment tendency of a sentence based on the sentiment score. If the sentiment score is greater than 0, mark the sentence as a positive sentence; if the sentiment score is equal to 0, mark it as a neutral sentence; if the sentiment score is less than 0, mark it as a negative sentence; Finally, the ratio of positive and negative sentences in all sentences is calculated to obtain the positive-negative ratio of comments, thus completing the sentiment analysis of social related data.

[0037] In the ESG evaluation system, the evaluation indicators of the social dimension cover multiple aspects such as employee rights protection, community relationship maintenance, and customer service quality. However, these indicators are often based on qualitative data, such as employees' subjective descriptions of the work environment, the community's general evaluation of the company, and customer feedback on service attitudes. Due to the lack of quantitative means, when incorporating this information into the ESG evaluation, it is easily affected by factors such as the evaluator's personal understanding and differences in judgment standards, resulting in the ESG evaluation being too subjective and difficult to ensure the consistency and comparability of evaluation results for different companies.

[0038] Therefore, the present invention obtains corporate social media data and mines out content closely related to the social dimension, providing an innovative and effective way to quantify the social dimension in ESG assessment. On social media platforms, employees may share their views on the company, and customers may also express their feelings about products or services. By analyzing these data and using the constructed sentiment dictionary to obtain the positive and negative ratio of comments, key information such as employee loyalty and customer satisfaction can be reflected concisely and intuitively. For example, if the positive and negative ratio of comments is greater than 1, it means that employees have a high degree of recognition of the company and customers are relatively satisfied with the products or services, which indirectly reflects the company's good performance in the social dimension; conversely, if the positive and negative ratio of comments is less than 1, it means that the company has problems in employee care, customer service, etc.

[0039] In the risk assessment module, the environmental assessment data and social assessment data (including social assessment quantitative data and positive and negative ratios of comments) are calculated according to the formula to obtain the environmental risk index and social risk index, and the governance assessment data are analyzed using the text analysis big model to obtain the governance risk index; Specifically, firstly, several indicators included in the environmental assessment data and social assessment data are classified and divided: those indicators that show positive impact on the environment or society are classified as positive indicators E+, such as energy conversion efficiency, water recycling rate, total power generation, wastewater discharge compliance rate, community project participation rate and positive-negative ratio of comments, etc.; while those indicators reflecting negative impacts are classified as negative indicators E-, such as total carbon emissions, total energy consumption, total water use, total wastewater discharge, product quality complaint rate, product return rate and employee turnover rate, etc.; Then, according to the formula ERI=[∑ 4 i=1 (w i ×(E - i / AE - ))] / [∑ 4 j=1 (w j ×(E + j / AE + ))+∑ 4 i=1 (w i ×(E - i / AE - ))] to calculate the environmental risk index ERI; where w i and w j represents the indicator weight, E - i and E + jRespectively represent the negative and positive indicators in environmental assessment data, AE - and AE + They represent the mean of negative indicators and the mean of positive indicators in environmental assessment data respectively; According to the formula SRI=[∑ 3 p=1 (w p ×(E - p / BE - ))] / [∑ 2 q=1 (w q ×(E + q / BE + ))+∑ 3 p=1 (w p ×(E - p / BE - ))] to calculate the social risk index SRI; where w p and w q represents the indicator weight, E - p and E + q Represent the negative and positive indicators in social evaluation data, BE - and BE + They represent the mean of negative indicators and the mean of positive indicators in social evaluation data respectively.

[0040] In this embodiment, the calculation process of the governance risk index is as follows: First, the collected management-related documents and governance-related data are processed and converted into plain text format to obtain text data that is convenient for subsequent analysis; Next, predetermine several dimensions of corporate governance risk assessment, including but not limited to board governance, internal audit, compliance management, and information disclosure; Replace the text data and several dimensions into the corresponding positions in the first instruction statement ("Please analyze the risk points of [several dimensions] in the [text data] according to the ESG risk assessment theory, and count the frequency of each risk point in the text"), and then input them into the existing language generation text analysis model. The text analysis model will analyze and output the risk points in each dimension and the corresponding frequency of each risk point in the text data based on these input information; Then replace the risk points and text data of several dimensions obtained into the corresponding positions in the second instruction statement ("Please judge the severity of [risk point] in [several dimensions] of [text data] based on ESG risk assessment theory, and express it with a score of 1-5, 1 for extremely low and 5 for extremely high") and input it into the text analysis model again. After being processed by the model, the severity score corresponding to each risk point will be output to quantify the severity of the risk point; the [] symbol in the instruction statement indicates the replacement position; According to the nature of the enterprise, the expert evaluation method is used to reasonably assign corresponding weights to each dimension; then, the formula GRI=∑ n=1 ∑ m=1 (S mn ×F mn )×W n To calculate the Governance Risk Index (GRI). In this formula, S mn Indicates the severity score of the mth risk point in the nth dimension, F mn represents the occurrence frequency of the mth risk point in the nth dimension, W n It represents the weight of the nth dimension. According to the relevant scores, frequencies and dimension weights of risk points under each dimension, they are substituted into the formula for calculation, and finally the governance risk index GRI is obtained.

[0041] Next, the analytic hierarchy process is used to obtain the weights of the environmental risk index, social risk index and governance risk index. The analytic hierarchy process will comprehensively consider various factors and weigh the relative importance of each index in the overall assessment to determine a reasonable weight value, and then obtain the ESG evaluation index through the weighted summation method.

[0042] Finally, in the risk warning module, the enterprise is judged whether there is a risk by judging whether the environmental risk index, social risk index, governance risk index and ESG assessment index exceed the corresponding preset thresholds. If a certain index exceeds the corresponding preset threshold, an early warning signal is immediately sent to remind the enterprise management personnel to take corresponding measures in response to the early warning signal. At the same time, it should be noted that this embodiment uses four index thresholds for risk warning, so the early warning signal sent will also send different early warning information according to the index type, intuitively displaying the source of risk for the enterprise and realizing intelligent enterprise risk early warning.

[0043] A second aspect of the present invention provides a risk assessment method based on enterprise ESG, including: S1, collects the environmental assessment data of enterprises, manages relevant documents and enterprise social media data, and collects quantitative data of social assessment; among which, the enterprise social media data is obtained through crawler technology; S2, using the data classification model to divide the enterprise social media data into social-related data and governance-related data, and obtain the positive and negative ratio of comments on social-related data based on the sentiment dictionary; S3, calculates the environmental risk index and social risk index based on the formula of environmental assessment data and social assessment data, and obtains the governance risk index based on the text analysis big model and management-related documents and governance-related data; among which, the social assessment data includes social assessment quantitative data and positive and negative ratios of comments; S4, using the analytic hierarchy process to obtain the weights of the environmental risk index, social risk index and governance risk index, and using the weighted sum method to obtain the ESG assessment index; S5, determines whether the environmental risk index, social risk index, governance risk index and ESG assessment index exceed the corresponding preset thresholds; if yes, sends a warning signal; if not, sends a normal signal.

[0044] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0045] Working principle of the present invention: Based on the ESG framework, the present invention conducts a comprehensive assessment and early warning of corporate risks from three main dimensions: environment, society, and governance. By collecting rich data through multiple channels, the data is deeply processed and analyzed using technical means such as deep learning algorithms and text analysis models, and various types of data are converted into quantifiable risk indexes. The ESG evaluation index is obtained by comprehensively considering the weights of each index, and finally compared with the preset threshold to achieve risk warning. The entire process realizes data-driven intelligent assessment and management of corporate risks, providing a scientific basis for enterprises to accurately identify potential risks and optimize operational strategies, while also providing strong support for decision-making by investors and other relevant entities.

[0046] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An enterprise ESG risk assessment system, characterized in that: include: Data collection module: used to collect the environmental assessment data of enterprises, manage relevant documents and enterprise social media data, and collect social assessment quantitative data; wherein the enterprise social media data is obtained through crawler technology; Data processing module: used to divide enterprise social media data into social-related data and governance-related data using a data classification model, and obtain the positive and negative ratio of comments on social-related data based on the sentiment dictionary; Risk assessment module: used to calculate environmental risk index and social risk index based on formulas from environmental assessment data and social assessment data, and to obtain governance risk index based on text analysis big model and management-related documents and governance-related data; and using the analytic hierarchy process to obtain the weights of the environmental risk index, social risk index and governance risk index, and obtaining the ESG evaluation index through the weighted summation method; wherein the social evaluation data includes social evaluation quantitative data and the positive and negative ratio of comments; Risk warning module: determines whether the environmental risk index, social risk index, governance risk index and ESG assessment index exceed the corresponding preset thresholds; if yes, a warning signal is sent; if not, a normal signal is sent.

2. The enterprise ESG risk assessment system according to claim 1, characterized in that: The environmental assessment data include the total carbon emissions, total power generation, total energy consumption, energy conversion efficiency, total water use, water recycling rate, total wastewater discharge and wastewater discharge compliance rate of the enterprise within the preset period; The management-related documents include the company's articles of association, minutes of board meetings, strategic planning documents, internal management system manuals, financial reports and audit reports, compliance policy documents, human resources management documents and information disclosure documents; The social assessment quantitative data include product quality complaint rate, product return rate, employee turnover rate and community project participation rate.

3. The enterprise ESG risk assessment system according to claim 1, characterized in that: The data classification model is built based on a deep learning algorithm, including: A1, crawling enterprise social media data using crawler technology and preprocessing to obtain text dataset; A2, classify several paragraphs in the text dataset to obtain several paragraph annotations, and use the paragraph vector method based on word embedding to map several paragraphs into the vector space to obtain several paragraph vectors; A3, build a neural network model based on deep learning, and set the training parameters, loss function and optimizer of the neural network model to obtain the initial model; A4: Divide a number of paragraph annotations and a number of paragraph vectors into a training set, a validation set, and a test set according to a preset ratio, input the training set and the validation set into the initial model for iterative training and update optimization, and obtain the optimal performance model based on the validation accuracy; A5: Input the test set into the best performing model for evaluation, obtain the test accuracy, and determine whether the test accuracy reaches the preset accuracy threshold; if yes, mark the best performing model as the data classification model; if no, adjust the training parameters, loss function, and optimizer to A4.

4. The enterprise ESG risk assessment system according to claim 1, characterized in that: The positive and negative ratio of comments of social related data obtained based on the sentiment dictionary includes: B1, preprocessing social related data to obtain preprocessed text; B2, classifying and assigning values ​​to a number of words in the preprocessed text to obtain a sentiment dictionary containing a number of positive words, a number of negative words and corresponding scores; B3. Obtain the sentiment tendency of several statements in the preprocessed text based on the sentiment dictionary, and calculate the ratio of several positive statements and several negative statements in the sentiment tendency to obtain the positive-negative ratio of the comments.

5. The enterprise ESG risk assessment system according to claim 4, characterized in that: The classification and assignment of several words in the preprocessed text include: B2-1. Select several initial words for classification and assignment to obtain the seed words and corresponding first scores for the positive and negative categories; B2-2. Perform word segmentation on the preprocessed text based on the N-gram model to obtain several words; B2-3. Use the word vector model to convert several words into several word vectors, and calculate the similarity values between several words and the seed words using cosine similarity to obtain several positive similarities and several negative similarities; B2-4. Classify and assign several words according to several positive similarities and several negative similarities according to the preset assignment rules.

6. The enterprise ESG risk assessment system according to claim 5, characterized in that: The preset assignment rules include: B2-41. Set the first similarity threshold A, the second similarity threshold B, and the third similarity threshold C, and when the similarity belongs to [0, A), assign a value of 0, when the similarity belongs to [A, B), assign a value of the third score, and when the similarity belongs to [B, C], assign a value of the second score to obtain the similarity assignment table; where A < B < C, 0 < the third score < the second score < the first score; B2-42. For each current word, select the highest value among several positive similarities and several negative similarities to obtain the highest positive degree and the highest negative degree; B2-43. Determine whether the highest positive degree is greater than the highest negative degree; if so, mark the current word as a positive word and jump to B2-44; otherwise, mark the current word as a negative word and jump to B2-44; B2-44. Assign a value to the current word according to the similarity assignment table to obtain the corresponding score.

7. The enterprise ESG risk assessment system according to claim 4, characterized in that: The obtaining of the sentiment tendency of several statements in the preprocessed text based on the sentiment dictionary includes: B3-1. Perform word segmentation on several statements in the preprocessed text based on the N-gram model to obtain several segmented words; B3-2. Match several segmented words in each statement with the sentiment dictionary to obtain several matching words and several corresponding scores; B3-3. Sum the corresponding scores of positive words among several matching words, and subtract the sum result of the corresponding scores of negative words to obtain the sentiment score; B3-4. Mark the statements with a sentiment score greater than 0 as positive statements; mark the statements with a sentiment score equal to 0 as neutral statements; mark the statements with a sentiment score less than 0 as negative statements.

8. The enterprise ESG risk assessment system according to claim 1, characterized in that: The calculation of the environmental risk index and the social risk index according to the formula includes: C1, divide several indicators in environmental assessment data and social assessment data to obtain positive indicators E + and negative indicator E - ; Among them, the positive indicators include energy conversion efficiency, water recycling rate, total power generation, wastewater discharge compliance rate, community project participation rate and positive and negative ratio of comments; the negative indicators include total carbon emissions, total energy consumption, total water use, total wastewater discharge, product quality complaint rate, product return rate and employee turnover rate; C2, according to the formula ERI=[∑ 4 i=1 (w i ×(E - i / AE - ))] / [∑ 4 j=1 (w j ×(E + j / AE + ))+∑ 4 i=1 (w i ×(E - i / AE - ))] to calculate the environmental risk index ERI; where w i and w j represents the indicator weight, E - i and E + j Respectively represent the negative and positive indicators in environmental assessment data, AE - and AE + They represent the mean of negative indicators and the mean of positive indicators in environmental assessment data respectively; C3, according to the formula SRI=[∑ 3 p=1 (w p ×(E - p / BE - ))] / [∑ 2 q=1 (w q ×(E + q / BE + ))+∑ 3 p=1 (w p ×(E - p / BE - ))] to calculate the social risk index SRI; where w p and w q represents the indicator weight, E - p and E + q Represent the negative and positive indicators in social evaluation data, BE - and BE + They represent the mean of negative indicators and the mean of positive indicators in social evaluation data respectively.

9. The enterprise ESG risk assessment system according to claim 1, characterized in that: The obtaining of the governance risk index according to the text analysis large model and management-related documents and governance-related data includes: D1. Convert management-related documents and governance-related data into plain text format data to obtain text data; D2. Determine several dimensions of enterprise risk assessment, and input the text data, several dimensions, and the first instruction statement into the text analysis large model to obtain several risk points of the dimensions and the occurrence frequencies of the risk points; where the text analysis large model is an existing language generation type large model; D3, input the risk points of several dimensions, text data and the second instruction statement into the text analysis model to obtain the severity score of the risk point; D4, assign weights to several dimensions according to the nature of the enterprise, and use the formula GRI=∑ n=1 ∑ m=1 (S mn ×F mn )×W n The governance risk index is calculated; where S mn Indicates the severity score of the mth risk point in the nth dimension, F mn represents the occurrence frequency of the mth risk point in the nth dimension, W n Represents the weight of the nth dimension.

10. An enterprise ESG risk assessment method, applied to an enterprise ESG risk assessment system according to any one of claims 1 to 9, characterized in that: include: S1, collect the environmental assessment data, management-related documents and corporate social media data of the enterprise, and collect social assessment quantitative data; wherein the corporate social media data is obtained through crawler technology; S2, using the data classification model to divide the enterprise social media data into social-related data and governance-related data, and obtain the positive and negative ratio of comments on social-related data based on the sentiment dictionary; S3, calculating the environmental risk index and social risk index by using the environmental assessment data and social assessment data according to the formula, and obtaining the governance risk index according to the text analysis big model and the management-related documents and governance-related data; wherein the social assessment data includes the social assessment quantitative data and the positive and negative ratio of comments; S4, using the analytic hierarchy process to obtain the weights of the environmental risk index, social risk index and governance risk index, and using the weighted sum method to obtain the ESG assessment index; S5, determines whether the environmental risk index, social risk index, governance risk index and ESG assessment index exceed the corresponding preset thresholds; if yes, sends a warning signal; if not, sends a normal signal.

Images