Bank ESG index system

Abstract

The invention provides a bank ESG index system, and the system is characterized in that the system comprises a cloud server which is used for storing patent data, enterprise financial indexes, a first-class policy file, and a second-class policy file; the at least one user terminal is in communication connection with the cloud server through a network, the cloud server comprises a service side communication part, a data storage part, an index and weight updating part, a three-level index and weight storage part, an ESG data extraction part and an ESG index calculation part, and the user terminal comprises an input storage part and a user side communication part. In a word, the system can obtain a more accurate ESG index evaluation result.

Description

Technical Field

[0001] The present invention relates to ESG index determination, and in particular to a bank ESG index system. Background Art

[0002] ESG is the abbreviation of environment, social responsibility and corporate governance. By focusing on the impact of enterprises on society from the three dimensions of ESG, we can see the sustainable development capabilities of enterprises. According to the theory of sustainable development, economic sustainability is the premise of social sustainability. As a micro-unit of economic development, enterprises can only achieve long-term sustainable development for themselves and the entire economy and society by balancing their own benefits, environmental interests and social interests.

[0003] Although there are some methods for determining ESG indexes both internationally and domestically, the existing technologies rarely take into account macroeconomic control factors when formulating ESG indexes. In particular, some foreign rating agencies have failed to consider China's localized rating methods, which has led to deviations in the rating results. In addition, the rating methods are mainly determined manually, which is costly, with low levels of automation and intelligence. They cannot be adjusted in a timely manner as the industry changes, and are not very timely.

[0004] In short, the existing technology for ESG index evaluation of banks still has problems such as low degree of automation and large deviation in ESG index evaluation results. Summary of the invention

[0005] The present invention is made to solve the above-mentioned problems, and its purpose is to provide a bank ESG index system.

[0006] The present invention provides a bank ESG index system, which is used for multiple users to obtain the bank's ESG index based on the bank's ESG report, patent data and corporate financial indicators, as well as a first-category policy document with macro-control content and a second-category policy document with specific reward and punishment measures. The system has the following characteristics, including: a cloud server, used to store patent data, corporate financial indicators, first-category policy documents and second-category policy documents; at least one user terminal, which is connected to the cloud server through a network, wherein the cloud server includes a service-side communication unit, a data storage unit, an indicator and weight update unit, a three-level indicator and weight storage unit, an ESG data extraction unit and an ESG index calculation unit; the user terminal includes an input storage unit and a user-side communication unit, the input storage unit is used to store the ESG report, patent data, corporate financial indicators, first-category policy documents and second-category policy documents input by the user, the user-side communication unit is used to send the patent data, corporate financial indicators, first-category policy documents and second-category policy documents in the input storage unit as update data and ESG reports to the cloud server, and then receive the ESG index from the cloud server, and the service-side communication unit is used to receive the update data sent by the user terminal The data storage unit is used to update the stored patent data, corporate financial indicators, first-category policy documents and second-category policy documents according to the updated data. The three-level indicator and weight storage unit is used to store the three-level indicators and the corresponding three-level weights. The three-level indicators include patent indicators, which are part of the patent categories. The indicator and weight updating unit is used to update the three-level indicators and the three-level weights in the three-level indicator and weight storage unit according to the patent data, corporate financial indicators, first-category policy documents and second-category policy documents stored in the data storage unit. The ESG data extraction unit is used to extract concise sentences from the ESG report, and then extract the corresponding data items, texts and text-related information from the concise sentences according to each of the three-level indicators in the three-level indicator and weight storage unit. The number of patents corresponding to the patent indicator is extracted from the patent data in the data storage unit, and the extracted number of patents, data items, texts and text-related information are used as the indicator data of the three-level indicators. The ESG index calculation unit is used to weighted average the three-level weights in the three-level indicator and weight storage unit and the corresponding indicator data extracted by the ESG data extraction unit to obtain the ESG index.

[0007] The bank ESG index system provided by the present invention may also have the following features: wherein, the indicator and weight updating unit includes a patent processing unit, an indicator extraction unit and an updating unit; the patent processing unit is used to classify the patent data in the data storage unit, obtain various patent categories and corresponding initial patent category weights, and select some patent categories as initial patent indicators; the indicator extraction unit includes an indicator extraction model, which is used to extract initial three-level indicators and corresponding initial weights from the first category policy documents and the second category policy documents in the data storage unit according to the indicator extraction model; the updating unit includes a regression model, which is used to update the three-level indicators and the three-level weights in the three-level indicator and weight storage unit according to the initial weights, initial patent category weights, initial patent indicators, initial three-level indicators, corporate financial indicators and indicator data.

[0008] The bank ESG index system provided by the present invention may also have the following features: wherein the specific process by which the updating unit updates the three-level indicators and the three-level weights is as follows: using the initial weights and the initial patent category weights as the weights to be optimized, using the initial patent indicators and the initial three-level indicators as the updated three-level indicators, obtaining the corresponding indicator data through the ESG data extraction unit, performing weighted averaging of the indicator data and the corresponding weights to be optimized to obtain the initial ESG index, and then inputting the initial ESG index and the corporate financial indicators into the regression model to obtain the optimized weights of each three-level indicator, and using each updated three-level indicator and the corresponding optimized weights as the three-level indicators and the three-level indicators and the three-level weights in the weight storage unit.

[0009] The bank ESG index system provided by the present invention may also have the following features: wherein the indicator extraction model is a first indicator extraction model and a second indicator extraction model, the first indicator extraction model includes a first Snorkel-MeTaL model and a first LSTM model, the second indicator extraction model includes a second Snorkel-MeTaL model and a second LSTM model, and the specific process of extracting the initial three-level indicators and the corresponding initial weights according to the first type of policy documents and the second type of policy documents includes the following steps: step S1, inputting the first type of policy documents into the first Snorkel-MeTaL model for annotation to obtain first annotated data; step S2, performing RoBERTa encoding on the first annotated data, and then inputting it into the first LSTM model Extract keywords as secondary indicators; step S3, obtain the weight of each secondary indicator according to the frequency of occurrence of the secondary indicator and the number of corresponding first-category policy documents; step S4, input the second-category policy documents into the second Snorkel-MeTaL model for annotation to obtain second annotated data; step S5, RoBERTa encode the second annotated data, and then input it into the second LSTM model, and extract keywords as key third-level indicators in combination with the secondary indicators; step S6, use synonyms of key third-level indicators as ordinary third-level indicators, and the set of ordinary third-level indicators and key third-level indicators as initial third-level indicators; step S7, obtain the initial weight of the initial third-level indicator according to the frequency of occurrence of the initial third-level indicator in the second-category policy documents and the weight of the corresponding secondary indicators.

[0010] The bank ESG index system provided by the present invention may also have the following features: wherein, the construction process of the first Snorkel-MeTaL model and the second Snorkel-MeTaL model includes the following steps: step T1, manually annotating some of the first-category policy documents as the first training data set, and manually annotating some of the second-category policy documents as the second training data set; step T2, compiling multiple annotation rules as annotation functions, and inputting the first training data set and the second training data set into the Snorkel-MeTaL model initialized with different annotation functions, respectively, to obtain corresponding training annotation data; step T3, checking each training annotation data according to the voting model, selecting the annotation function corresponding to the training annotation data with the best performance in the first training data set as the first annotation function, selecting the annotation function corresponding to the training annotation data with the best performance in the second training data set as the second annotation function, the Snorkel-MeTaL model with the first annotation function is the first Snorkel-MeTaL model, and the Snorkel-MeTaL model with the second annotation function is the second Snorkel-MeTaL model.

[0011] The bank ESG index system provided by the present invention may also have the following features: wherein the first labeled data and the second labeled data are both phrases with labels, the structure of the phrases is a verb plus a noun, and the labels are positive, negative or neutral.

[0012] The bank ESG index system provided by the present invention may also have the following features: wherein, the specific process of extracting concise sentences from the ESG report is: converting the content of the ESG report into well-defined sentences, then removing irrelevant punctuation marks in the sentences, setting rules to extract grammatically valid sentences, and finally removing stop words in the grammatically valid sentences to obtain concise sentences.

[0013] The bank ESG index system provided by the present invention may also have the following features: when the user-side communication unit only sends the ESG report to the cloud server, the indicator and weight updating unit does not update the third-level indicators and third-level weights in the third-level indicator and weight storage unit, and the third-level indicators and third-level weights in the third-level indicator and weight storage unit are directly used to calculate the ESG index.

[0014] Functions and Effects of the Invention

[0015] According to the bank ESG index system involved in the present invention, because the number of patent applications of the bank in some patent categories is quantified as a data indicator of the bank's technological development level, that is, as a third-level indicator of ESG, it reflects the bank's technological development level, extracts keywords from two types of policy documents to construct third-level indicators and third-level weights, incorporates policy influencing factors into the consideration scope of the ESG index, obtains China's localized ESG rating results, adjusts the third-level weights through corporate financial indicators, and enhances the correlation between the actual business conditions of the enterprise and the ESG index. Therefore, the bank ESG index system of the present invention can obtain more accurate ESG index evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic diagram of the framework of the bank ESG index system in an embodiment of the present invention;

[0017] Figure 2 is a schematic diagram of a framework of a user terminal in an embodiment of the present invention;

[0018] Figure 3 is a schematic diagram of a cloud server framework in an embodiment of the present invention;

[0019] Figure 4 is a schematic diagram of a framework of an indicator and weight updating unit in an embodiment of the present invention;

[0020] Figure 5 is a schematic diagram of a process of constructing a first indicator extraction model and a second indicator extraction model in an embodiment of the present invention;

[0021] Figure 6 is a schematic diagram of a process for extracting initial three-level indicators and corresponding initial weights in an embodiment of the present invention;

[0022] Figure 7 It is a flow chart of obtaining a bank ESG index in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments and accompanying drawings specifically illustrate the bank ESG index system of the present invention.

[0024] Figure 1 It is a schematic diagram of the framework of the bank ESG index system in an embodiment of the present invention.

[0025] like Figure 1 As shown, the bank ESG index system 100 includes a cloud server 101 and at least one user terminal 102 that is connected to the cloud server 101 through a network, and is used for multiple users to transmit the bank's ESG report, patent data and corporate financial indicators, as well as the first type of policy documents with macro-control content and the second type of policy documents with specific reward and punishment measures from the user terminal 102 to the cloud server 101, and obtain the corresponding bank's ESG index.

[0026] In this embodiment, the first type of policy documents are development plans, plans and policy interpretation news, and the second type of policy documents are specific reward and punishment measures, including rewards for green enterprises and qualitative standards or specific quantitative indicators for green enterprise assessment.

[0027] Figure 2 It is a schematic diagram of the framework of a user terminal in an embodiment of the present invention.

[0028] like Figure 2 As shown, each user terminal 102 includes an input storage unit 1021 and a user side communication unit 1022 .

[0029] The input storage unit 1021 is used to store the ESG report, patent data, corporate financial indicators, first-class policy documents and second-class policy documents input by the user. The user-side communication unit 1022 is used to send the patent data, corporate financial indicators, first-class policy documents and second-class policy documents in the input storage unit 1021 as update data and ESG reports to the cloud server 101, and then receive the ESG index from the cloud server 101.

[0030] Figure 3 It is a schematic diagram of the framework of the cloud server in an embodiment of the present invention.

[0031] like Figure 3 As shown, the cloud server 101 includes a service-side communication unit 1011, a data storage unit 1012, a three-level indicator and weight storage unit 1013, an indicator and weight updating unit 1014, an ESG data extraction unit 1015 and an ESG index calculation unit 1016.

[0032] The service-side communication unit 1011 is used to receive update data and ESG reports sent by the user terminal 102 , and send the ESG index corresponding to the ESG report to the user terminal 102 .

[0033] The data storage unit 1012 is used to update the stored patent data, enterprise financial indicators, first-category policy documents and second-category policy documents according to the updated data.

[0034] The three-level indicator and weight storage unit 1013 is used to store the three-level indicators and the corresponding three-level weights. The three-level indicators include patent indicators, and the patent indicators are some patent categories.

[0035] Figure 4 It is a schematic diagram of the framework of the indicator and weight updating unit in the embodiment of the present invention.

[0036] like Figure 4 As shown, the indicator and weight updating unit 1014 includes a patent processing unit 10141, an indicator extraction unit 10142 and an updating unit 10143.

[0037] The indicator and weight updating unit 1014 is used to update the third-level indicators and third-level weights in the third-level indicator and weight storage unit 1013 according to the patent data, corporate financial indicators, first-category policy documents and second-category policy documents stored in the data storage unit 1012.

[0038] The patent processing unit 10141 is used to classify the patent data in the data storage unit 1012, obtain various patent categories and corresponding initial patent category weights, and select some patent categories as initial patent indicators.

[0039] The indicator extraction unit 10142 includes an indicator extraction model, which is used to extract initial three-level indicators and corresponding initial weights from the first type of policy files and the second type of policy files of the data storage unit 1012 according to the indicator extraction model.

[0040] Among them, the indicator extraction model is a first indicator extraction model and a second indicator extraction model. The first indicator extraction model includes a first Snorkel-MeTaL model and a first LSTM model, and the second indicator extraction model includes a second Snorkel-MeTaL model and a second LSTM model.

[0041] Figure 5It is a flowchart of constructing a first indicator extraction model and a second indicator extraction model in an embodiment of the present invention.

[0042] like Figure 5 As shown, the construction process of the first Snorkel-MeTaL model and the second Snorkel-MeTaL model includes the following steps:

[0043] Step T1, manually annotate some of the first category policy documents as the first training data set, and manually annotate some of the second category policy documents as the second training data set.

[0044] Step T2, compile multiple labeling rules as labeling functions, and input the first training data set and the second training data set into the Snorkel-MeTaL model initialized with different labeling functions to obtain corresponding training labeling data.

[0045] Step T3, check each training labeled data according to the voting model, select the labeling function corresponding to the training labeled data with the best performance in the first training data set as the first labeling function, and select the labeling function corresponding to the training labeled data with the best performance in the second training data set as the second labeling function.

[0046] The Snorkel-MeTaL model with the first annotation function is a first Snorkel-MeTaL model, and the Snorkel-MeTaL model with the second annotation function is a second Snorkel-MeTaL model.

[0047] Figure 6 It is a schematic diagram of a process of extracting initial three-level indicators and corresponding initial weights in an embodiment of the present invention.

[0048] like Figure 6 As shown, according to the first type of policy documents and the second type of policy documents, the specific process of extracting the initial three-level indicators and the corresponding initial weights includes the following steps:

[0049] Step S1: input a first type of policy document into a first Snorkel-MeTaL model for annotation to obtain first annotation data.

[0050] Step S2: RoBERTa encode the first labeled data and then input it into the first LSTM model to extract keywords as secondary indicators.

[0051] Step S3, obtaining the weight of each secondary indicator according to the occurrence frequency of the secondary indicator and the number of corresponding first-category policy documents.

[0052] Step S4: input the second type of policy documents into the second Snorkel-MeTaL model for annotation to obtain second annotation data.

[0053] Step S5, RoBERTa encode the second labeled data, then input it into the second LSTM model, and extract keywords as key third-level indicators in combination with the second-level indicators.

[0054] Step S6: taking synonyms of key third-level indicators as common third-level indicators, and the set of common third-level indicators and key third-level indicators as initial third-level indicators.

[0055] Step S7, according to the occurrence frequency of the initial third-level indicator in the second type of policy documents and the weight of the corresponding second-level indicator, the initial weight of the initial third-level indicator is obtained. In this embodiment, the initial weight corresponding to the common third-level indicator is lower than the initial weight corresponding to the key third-level indicator.

[0056] Among them, the first labeled data and the second labeled data are both phrases with labels, the structure of the phrases is a verb plus a noun, and the labels are positive, negative or neutral. In this embodiment, the verbs of the phrases with positive labels include promote, insist, promote and advance, and the corresponding weights are positive numbers, and the verbs of the phrases with negative labels include reduce, limit and control, and the corresponding weights are negative numbers.

[0057] The updating unit 10143 includes a regression model, which is used to update the third-level indicators and third-level weights in the third-level indicator and weight storage unit 1013 according to the initial weights, initial patent category weights, initial patent indicators, initial third-level indicators, corporate financial indicators and indicator data.

[0058] Among them, the specific process of updating the three-level indicators and the three-level weights by the updating unit 10143 is as follows: the initial weights and the initial patent category weights are used as the weights to be optimized, the initial patent indicators and the initial three-level indicators are used as the updated three-level indicators, the corresponding indicator data are obtained through the ESG data extraction unit 1015, the indicator data and the corresponding weights to be optimized are weighted averaged to obtain the initial ESG index, and then the initial ESG index and the corporate financial indicators are input into the regression model to obtain the optimized weights of each three-level indicator, and each updated three-level indicator and the corresponding optimized weights are used as the three-level indicators and the three-level indicators and the three-level weights in the weight storage unit 1013. In this embodiment, the gradient descent method is used to update the three-level weights until the regression model converges, and the optimized weights are obtained.

[0059] The ESG report corresponding to the indicator data used in the initial ESG index corresponds to the corporate financial indicators input into the regression model.

[0060] The ESG data extraction unit 1015 is used to extract concise sentences from the ESG report, and then extract the corresponding data items, texts and text-related information from the concise sentences according to each third-level indicator in the third-level indicator and weight storage unit 1013, and extract the number of patents corresponding to the patent indicator from the patent data in the data storage unit 1012, and then use the extracted number of patents, data items, texts and text-related information as indicator data of the third-level indicator.

[0061] The specific process of extracting concise sentences from the ESG report is as follows: converting the content of the ESG report into well-defined sentences, then removing irrelevant punctuation marks in the sentences, setting rules to extract grammatically valid sentences, and finally removing stop words in the grammatically valid sentences to obtain concise sentences. In this embodiment, irrelevant punctuation marks are removed from ESG reports with different PDF formats using the re library in the python language, only necessary text data is output, and then rules are set to extract grammatically valid sentences to obtain valid sentences.

[0062] The ESG index calculation unit 1016 is used to perform weighted averaging on the three-level weights in the three-level index and weight storage unit 1013 and the corresponding index data extracted by the ESG data extraction unit 1015 to obtain the ESG index.

[0063] Among them, when the user-side communication unit 1022 only sends the ESG report to the cloud server 101, the indicator and weight updating unit 1014 does not update the third-level indicators and third-level weights in the third-level indicator and weight storage unit 1013, and the third-level indicators and third-level weights in the third-level indicator and weight storage unit 1013 are directly used to calculate the ESG index.

[0064] Figure 7 It is a flow chart of obtaining a bank ESG index in an embodiment of the present invention.

[0065] like Figure 7 As shown, obtaining the bank ESG index includes the following steps:

[0066] Step R1, the input storage unit 1021 stores the ESG report, patent data, corporate financial indicators, first-category policy documents and second-category policy documents input by the user.

[0067] In step R2, the user-side communication unit 1022 sends the patent data, enterprise financial indicators, first-category policy documents and second-category policy documents in the input storage unit 1021 as update data and an ESG report to the cloud server 101.

[0068] Step R3 : the service-side communication unit 1011 receives the update data and ESG report sent by the user terminal 102 .

[0069] Step R4, determine whether update data is received, if so, go to step R5, if not, go to step R7.

[0070] Step R5, the data storage unit 1012 updates the stored patent data, enterprise financial indicators, first-category policy documents and second-category policy documents according to the updated data.

[0071] Step R6, the indicator and weight updating unit 1014 updates the third-level indicators and third-level weights in the third-level indicator and weight storage unit 1013 according to the patent data, corporate financial indicators, first-category policy documents and second-category policy documents stored in the data storage unit 1012.

[0072] Step R7, the ESG data extraction unit 1015 extracts simplified sentences from the ESG report, and then extracts corresponding data items, texts and text-related information from the simplified sentences according to each third-level indicator in the third-level indicator and weight storage unit 1013, and extracts the number of patents corresponding to the patent indicator from the patent data in the data storage unit 1012, and then uses the extracted number of patents, data items, texts and text-related information as indicator data of the third-level indicator.

[0073] In step R8, the ESG index calculation unit 1016 performs weighted average on the three-level weights in the three-level index and weight storage unit 1013 and the corresponding index data extracted by the ESG data extraction unit to obtain the ESG index.

[0074] In step R9 , the service-side communication unit 1011 sends the ESG index corresponding to the ESG report to the user terminal 102 .

[0075] Functions and Effects of the Embodiments

[0076] According to the bank ESG index system involved in this embodiment, the number of patent applications of the bank in some patent categories is quantified as a data indicator of the bank's technological development level, that is, as a third-level indicator of ESG, thereby reflecting the bank's technological development level. Keywords are extracted from two types of policy documents to construct third-level indicators and third-level weights, and policy influencing factors are included in the consideration scope of the ESG index to obtain the localized ESG rating results in China. The third-level weights are adjusted through corporate financial indicators to enhance the correlation between the actual business conditions of the enterprise and the ESG index. In short, this method can obtain more accurate ESG index evaluation results.

[0077] The above-mentioned embodiments are preferred examples of the present invention and are not intended to limit the protection scope of the present invention.

Claims

1. A bank ESG index system is used for multiple users to obtain the ESG index of the bank based on the bank's ESG report, patent data and corporate financial indicators, as well as the first type of policy documents with macroeconomic control content and the second type of policy documents with specific reward and punishment measures, characterized in that: include: A cloud server, used to store the patent data, the enterprise financial indicators, the first category policy documents and the second category policy documents; At least one user terminal is connected to the cloud server through a network. The cloud server includes a service-side communication unit, a data storage unit, an index and weight update unit, a three-level index and weight storage unit, an ESG data extraction unit, and an ESG index calculation unit. The user terminal includes an input storage unit and a user side communication unit. The input storage unit is used to store the ESG report, the patent data, the corporate financial indicators, the first type of policy documents and the second type of policy documents input by the user, The user-side communication unit is used to send the patent data, the enterprise financial indicators, the first type of policy documents and the second type of policy documents in the input storage unit as update data and the ESG report to the cloud server, and then receive the ESG index from the cloud server. The service-side communication unit is used to receive the update data and the ESG report sent by the user terminal, and send the ESG index corresponding to the ESG report to the user terminal, The data storage unit is used to update the stored patent data, the enterprise financial indicators, the first type of policy documents and the second type of policy documents according to the update data, The three-level indicator and weight storage unit is used to store three-level indicators and corresponding three-level weights. The three-level indicators include patent indicators, which are some patent categories. The indicator and weight updating unit is used to update the third-level indicators and the third-level weights in the third-level indicator and weight storage unit according to the patent data, the enterprise financial indicators, the first-type policy documents and the second-type policy documents stored in the data storage unit, The ESG data extraction unit is used to extract concise sentences from the ESG report, and then extract corresponding data items, texts and text-related information from the concise sentences according to each of the three-level indicators in the three-level indicators and weight storage unit, extract the number of patents corresponding to the patent indicators from the patent data in the data storage unit, and use the extracted number of patents, the data items, the texts and the text-related information as indicator data of the three-level indicators. The ESG index calculation unit is used to perform weighted averaging on the three-level weights in the three-level index and weight storage unit and the corresponding index data extracted by the ESG data extraction unit to obtain the ESG index.

2. The bank ESG index system according to claim 1, characterized in that: in, The index and weight updating unit includes a patent processing unit, an index extraction unit and an updating unit. The patent processing unit is used to classify the patent data in the data storage unit, obtain various patent categories and corresponding initial patent category weights, and select some of the patent categories as initial patent indicators. The indicator extraction unit includes an indicator extraction model, which is used to extract initial three-level indicators and corresponding initial weights from the first type of policy files and the second type of policy files in the data storage unit according to the indicator extraction model. The updating unit includes a regression model, which is used to update the third-level indicators and the third-level weights in the third-level indicator and weight storage unit according to the initial weights, the initial patent category weights, the initial patent indicators, the initial third-level indicators, the corporate financial indicators and the indicator data.

3. The bank ESG index system according to claim 2, characterized in that: in, The specific process of the updating unit updating the three-level indicators and the three-level weights is as follows: The initial weights and the initial patent category weights are used as weights to be optimized, the initial patent indicators and the initial three-level indicators are used as updated three-level indicators, the corresponding indicator data are obtained through the ESG data extraction unit, the indicator data and the corresponding weights to be optimized are weighted averaged to obtain the initial ESG index, and then the initial ESG index and the corporate financial indicators are input into the regression model to obtain the optimized weights of each of the three-level indicators, and each of the updated three-level indicators and the corresponding optimized weights are used as the three-level indicators and the three-level weights in the weight storage unit.

4. The bank ESG index system according to claim 2, Features: The indicator extraction model is a first indicator extraction model and a second indicator extraction model, the first indicator extraction model includes a first Snorkel-MeTaL model and a first LSTM model, the second indicator extraction model includes a second Snorkel-MeTaL model and a second LSTM model, According to the first type of policy documents and the second type of policy documents, the specific process of extracting the initial three-level indicators and the corresponding initial weights includes the following steps: Step S1, inputting the first type of policy document into the first Snorkel-MeTaL model for annotation to obtain first annotation data; Step S2, performing RoBERTa encoding on the first labeled data, and then inputting the first LSTM model to extract keywords as secondary indicators; Step S3, obtaining the weight of each of the secondary indicators according to the occurrence frequency of the secondary indicators and the number of the corresponding first-category policy documents; Step S4, inputting the second type of policy document into the second Snorkel-MeTaL model for annotation to obtain second annotation data; Step S5, performing RoBERTa encoding on the second labeled data, and then inputting the second LSTM model, and extracting keywords as key third-level indicators in combination with the second-level indicators; Step S6, using synonyms of the key third-level indicators as common third-level indicators, and the set of common third-level indicators and key third-level indicators as the initial third-level indicators; Step S7, obtaining the initial weight of the initial third-level indicator according to the appearance frequency of the initial third-level indicator in the second type of policy documents and the corresponding weight of the second-level indicator.

5. The bank ESG index system according to claim 4, Features: The construction process of the first Snorkel-MeTaL model and the second Snorkel-MeTaL model includes the following steps: Step T1, manually annotating part of the first-category policy documents as a first training data set, and manually annotating part of the second-category policy documents as a second training data set; Step T2, compiling a plurality of labeling rules as labeling functions, respectively inputting the first training data set and the second training data set into Snorkel-MeTaL models initialized with different labeling functions to obtain corresponding training labeling data; Step T3, checking each of the training labeled data according to the voting model, selecting the labeling function corresponding to the training labeled data with the best performance in the first training data set as the first labeling function, and selecting the labeling function corresponding to the training labeled data with the best performance in the second training data set as the second labeling function, The Snorkel-MeTaL model with the first annotation function is the first Snorkel-MeTaL model, and the Snorkel-MeTaL model with the second annotation function is the second Snorkel-MeTaL model.

6. The bank ESG index system according to claim 4, characterized in that: in, The first labeled data and the second labeled data are both phrases with labels, the structure of the phrases is a verb plus a noun, and the labels are positive, negative or neutral.

7. The bank ESG index system according to claim 1, characterized in that: in, The specific process of extracting the simplified sentences from the ESG report is as follows: The content of the ESG report is converted into a well-defined sentence, and then irrelevant punctuation marks in the sentence are removed, and then rules are set to extract grammatically valid sentences, and finally stop words in the grammatically valid sentences are removed to obtain the simplified sentence.

8. The bank ESG index system according to claim 1, characterized in that: in, When the user-side communication unit only sends the ESG report to the cloud server, the indicator and weight updating unit does not update the third-level indicators and the third-level weights in the third-level indicator and weight storage unit, and the third-level indicators and the third-level weights in the third-level indicator and weight storage unit are directly used to calculate the ESG index.

Images