Enterprise bleaching detection method for structured information extraction and related equipment

Abstract

The invention relates to the technical field of greening detection, and provides an enterprise greening detection method for structured information extraction and related equipment, the method comprises the following steps: obtaining an ESG report text of a target enterprise, and extracting a plurality of statements from the ESG report text; performing semantic recognition on each statement to obtain a structured evidence object of each statement; for each statement, calculating the greening confidence coefficient of the statement based on the structured evidence object of the statement; calculating the environment floating-green confidence coefficient of the target enterprise based on the floating-green confidence coefficients corresponding to all the environment statements, calculating the social floating-green confidence coefficient of the target enterprise based on the floating-green confidence coefficients corresponding to all the social statements, and calculating the governance floating-green confidence coefficient of the target enterprise based on the floating-green confidence coefficients corresponding to all the governance statements; and generating an enterprise greening detection result of the target enterprise according to the environment greening confidence coefficient, the social greening confidence coefficient and the governance greening confidence coefficient. According to the method, the credibility of enterprise bleaching detection can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of greenwashing detection, and particularly relates to an enterprise greenwashing detection method based on structured information extraction and related equipment. BACKGROUND

[0002] With the disclosure of environmental, social and governance (ESG) information becoming the focus of capital markets and regulatory agencies, enterprises have the problem of greenwashing in reports, that is, creating a "green image" in a way of ambiguous expression, excessive commitment, and deviation from facts.

[0003] Traditional text detection methods mostly stay at the level of keyword matching or sentiment analysis, and it is difficult to reveal the logical relationship between deep features such as commitment subject, time constraint, quantitative basis and expression strength, resulting in an unreliable enterprise greenwashing detection result. SUMMARY

[0004] The embodiments of the present application provide an enterprise greenwashing detection method based on structured information extraction and related equipment, which can solve the problem of unreliable enterprise greenwashing detection result.

[0005] In a first aspect, the present application provides an enterprise greenwashing detection method based on structured information extraction, which comprises: obtaining an ESG report text of a target enterprise, and extracting a plurality of sentences from the ESG report text; the sentence is an environmental sentence, a social sentence or a governance sentence; respectively for each sentence, performing keyword extraction, responsibility subject extraction, time adverbial and modification component detection, and qualitative modifier analysis on the sentence to obtain a structured evidence object of the sentence; the structured evidence object comprises a core keyword, a responsibility subject, a time state label and an implementation degree index of the sentence; respectively for each sentence, calculating an initial greenwashing confidence of the sentence based on the implementation degree index of the sentence, and determining a greenwashing confidence of the sentence based on the initial greenwashing confidence and the core keyword, the responsibility subject and the time state label; calculating an environmental greenwashing confidence of the target enterprise based on the greenwashing confidence corresponding to all environmental sentences, calculating a social greenwashing confidence of the target enterprise based on the greenwashing confidence corresponding to all social sentences, and calculating a governance greenwashing confidence of the target enterprise based on the greenwashing confidence corresponding to all governance sentences; generating an enterprise greenwashing detection result of the target enterprise according to the environmental greenwashing confidence, the social greenwashing confidence and the governance greenwashing confidence.

[0006] Optionally, keyword extraction, responsibility subject extraction, time adverbial and modification component detection, qualitative modifier analysis are performed on the sentence to obtain the structured evidence object of the sentence, including: identifying a plurality of core keywords of the sentence to form a keyword set; extracting a plurality of responsibility subjects involved in the sentence to form a subject set; detecting the time adverbial and modification component in the sentence to obtain a time state label; establishing an implementation degree index of the sentence according to the qualitative modifier in the sentence; integrating the keyword set, the subject set, the time state label and the implementation degree index into one data to obtain the structured evidence object of the sentence.

[0007] Optionally, the initial green drift confidence of the sentence is calculated based on the implementation degree index of the sentence, and the green drift confidence of the sentence is determined based on the initial green drift confidence and the core keywords, the responsibility subjects and the time state label, including: calculating a consistency index according to the implementation degree index in the structured evidence object of the sentence; iteratively updating the consistency index to obtain a final consistency index; calculating the initial green drift confidence of the sentence based on the final consistency index; judging whether the initial green drift confidence is less than a first confidence and greater than or equal to a second confidence; the first confidence is less than the second confidence; if yes, the initial green drift confidence is taken as the green drift confidence of the sentence; otherwise, re-evaluating using the keyword set, the subject set and the time state label in the structured evidence object of the sentence to obtain the green drift confidence of the sentence.

[0008] Optionally, the consistency index is calculated according to the implementation degree index in the structured evidence object of the sentence, including: the consistency index is calculated by the formula: ; ; ; ; wherein, the weighted geometric mean value is represented by, the weighted arithmetic mean value is represented by, , , all are weight coefficients, the entity credibility is represented by, the time certainty is represented by, the implementation degree index is represented by.​ The consistency index is iteratively updated to obtain a final consistency index, comprising: updating the implementation degree index based on the consistency index; determining whether the updated implementation degree index meets a stability condition; if yes, taking the consistency index as the final consistency index; otherwise, taking the updated implementation degree index as the implementation degree index, and returning to the step of calculating the consistency index according to the implementation degree index in the structured evidence object of the sentence.

[0009] Optionally, updating the implementation degree index based on the consistency index comprises: updating the implementation degree index by the formula: ; ; wherein, represents a balance constant.

[0010] Optionally, calculating the initial green drift confidence of the sentence based on the final consistency index comprises: calculating the initial green drift confidence by the formula: ; ; wherein, represents the final consistency index; re-evaluating the sentence based on the keyword set, the subject set and the time state label in the structured evidence object of the sentence to obtain the green drift confidence of the sentence, comprising: generating a search formula based on the keyword set, the subject set and the time state label of the sentence; searching a plurality of candidate evidence sentences from a sentence library based on the search formula; calculating the similarity between the structured evidence object of the sentence and the structured evidence object of each candidate evidence sentence, and taking the candidate evidence sentence with the largest similarity as a target evidence sentence; calculating the green drift confidence of the sentence based on the target evidence sentence.

[0011] Optionally, calculating the green drift confidence of the sentence based on the target evidence sentence comprises: calculating the green drift confidence by the formula: ; ; wherein, represents the consistency index based on the target evidence sentence: ;​​​ ; ; wherein, represents a weighted geometric mean value based on the target evidence sentence, represents a weighted arithmetic mean value based on the target evidence sentence, represents a stability term, 、 、 are weight coefficients, represents an entity credibility of the target evidence sentence, represents a time certainty of the target evidence sentence, represents an implementation degree index of the target evidence sentence.

[0012] Optionally, the environmental greenwashing confidence of the target enterprise is calculated based on the greenwashing confidence corresponding to all the environmental sentences, including: Through the formula: ; ; ; Calculate the environmental greenwashing confidence ; wherein, represents the weight of the th environmental sentence, represents a set of numbers of the environmental sentences, represents the greenwashing confidence of the th environmental sentence, represents the initial weight of the th environmental sentence, represents the initial weight of the th environmental sentence, represents the classification confidence of the th environmental sentence, represents the structured extraction confidence of the th environmental sentence, represents the external evidence similarity of the th environmental sentence.

[0013] In a second aspect, the present application provides an enterprise greenwashing detection device for structured information extraction, comprising: an acquisition module configured to acquire an ESG report text of a target enterprise, and extract a plurality of sentences from the ESG report text; the sentences are environmental sentences, social sentences or governance sentences; The semantic recognition module is used to extract keywords, identify responsible parties, detect time adverbs and modifiers, and analyze qualitative modifiers for each statement to obtain structured evidence objects. The structured evidence objects include the core keywords, responsible parties, time status labels, and achievement level indicators of the statement. The first calculation module is used to calculate the initial greenwash confidence of each statement based on the statement's implementation level index, and to determine the statement's greenwash confidence based on the initial greenwash confidence, core keywords, responsible parties, and time status labels. The second calculation module is used to calculate the environmental greenwashing confidence of the target company based on the greenwashing confidence of all environmental statements, calculate the social greenwashing confidence of the target company based on the greenwashing confidence of all social statements, and calculate the governance greenwashing confidence of the target company based on the greenwashing confidence of all governance statements. The generation module is used to generate enterprise greenwashing detection results for target companies based on environmental greenwashing confidence, social greenwashing confidence, and governance greenwashing confidence.

[0014] Thirdly, embodiments of this application provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described enterprise greenwashing detection method for structured information extraction.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described enterprise greenwashing detection method for structured information extraction.

[0016] The above-mentioned solution in this application has the following beneficial effects: In the embodiments of this application, the ESG report text of the target enterprise is obtained, and multiple statements are extracted from the ESG report text. Then, for each statement, keyword extraction, responsible entity extraction, time adverbial and modifying component detection, and qualitative modifier analysis are performed to obtain the structured evidence object of the statement. Then, for each statement, the initial greenwashing confidence score of the statement is calculated based on the statement's realization degree index. Based on the initial greenwashing confidence score, core keywords, responsible entities, and time status labels, the greenwashing confidence score of the statement is determined. Then, the environmental greenwashing confidence score of the target enterprise is calculated based on the greenwashing confidence scores corresponding to all environmental statements, the social greenwashing confidence score of the target enterprise is calculated based on the greenwashing confidence scores corresponding to all social statements, and the governance greenwashing confidence score of the target enterprise is calculated based on the greenwashing confidence scores corresponding to all governance statements. Finally, the enterprise greenwashing detection result of the target enterprise is generated based on the environmental greenwashing confidence score, the social greenwashing confidence score, and the governance greenwashing confidence score. Among them, semantic recognition of statements can effectively analyze the semantic information of ESG report text, revealing deeper features such as subject, time state, and degree of implementation in the text, ensuring information richness and depth. Based on the information in the structured evidence object with information richness and depth, the initial greenwash confidence is calculated and the final greenwash confidence is determined, effectively improving the accuracy of greenwash confidence and thus effectively improving the credibility of greenwash detection for enterprises. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a structured information extraction method for detecting green bleaching in enterprises, provided as an embodiment of this application; Figure 2 A schematic diagram of a structured information extraction detection device for enterprises in accordance with an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0020] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0021] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0023] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0024] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0025] To address the unreliability of existing greenwashing detection results for enterprises, this application provides a structured information extraction method for enterprise greenwashing detection. This method obtains the ESG report text of the target enterprise and extracts multiple statements from it. For each statement, keyword extraction, responsible entity extraction, detection of time adverbs and modifiers, and qualitative modifier analysis are performed to obtain structured evidence objects. For each statement, an initial greenwashing confidence score is calculated based on the statement's realization level index. Based on the initial greenwashing confidence score, core keywords, responsible entities, and time status tags, the greenwashing confidence score of the statement is determined. Then, the environmental greenwashing confidence score of the target enterprise is calculated based on the greenwashing confidence scores corresponding to all environmental statements, the social greenwashing confidence score based on the greenwashing confidence scores corresponding to all social statements, and the governance greenwashing confidence score based on the greenwashing confidence scores corresponding to all governance statements. Finally, the enterprise greenwashing detection result for the target enterprise is generated based on the environmental, social, and governance greenwashing confidence scores. Among them, semantic recognition of statements can effectively analyze the semantic information of ESG report text, revealing deeper features such as subject, time state, and degree of implementation in the text, ensuring information richness and depth. Based on the information in the structured evidence object with information richness and depth, the initial greenwash confidence is calculated and the final greenwash confidence is determined, effectively improving the accuracy of greenwash confidence and thus effectively improving the credibility of greenwash detection for enterprises.

[0026] The following is an illustrative example of the enterprise greenwashing detection method for structured information extraction provided in this application.

[0027] like Figure 1 As shown, the enterprise greenwashing detection method for structured information extraction provided in this application includes the following steps: Step 11: Obtain the ESG report text of the target company and extract multiple statements from the ESG report text.

[0028] The target companies mentioned above are those required to undergo greenwashing testing. The statements used are environmental, social, or governance statements. The ESG report text is a disclosure report by the target companies regarding their non-financial performance in environmental, social, and governance aspects. Environmental statements are those related to the environment in the ESG report text (e.g., "The company's greenhouse gas emissions have decreased by 8% compared to the previous year"); social statements are those related to society in the ESG report text (e.g., "Donated 2 million yuan to fund the construction of educational facilities in schools in remote areas and launched 10 environmental volunteer projects"); and governance statements are those related to governance in the ESG report text (e.g., "Holds regular shareholders' meetings and ensures that shareholders' voices can directly influence company decisions").

[0029] In some embodiments of this application, ESG report text can be obtained by accessing the company's official website, etc. Multiple sentences can be extracted from the ESG report text using models such as Sentence-Bidirectional Encoder Representations from Transformers (Sentence-BERT). Each sentence can be classified into environmental, social, or governance sentences using the BERT model for Environmental, Social, and Governance (FinBERT-ESG).

[0030] For example, after obtaining the ESG report text, regular expressions and natural language processing tools (such as spaCy) can be used to preprocess the ESG report text, removing irrelevant formatting symbols, headers and footers, general statements and background descriptions, and retaining the core content relevant to the target company.

[0031] Step 12: For each statement, perform keyword extraction, responsible party extraction, time adverbial and modifying component detection, and qualitative modifier analysis to obtain the structured evidence object of the statement.

[0032] Structured evidence includes core keywords, responsible parties, time status labels, and achievement indicators. Core keywords refer to ESG-related terms, such as carbon emissions, energy consumption, and employee performance. Responsible parties refer to ESG-related entities and organizations, such as the company itself, subsidiaries, and supply chain links. Time status labels describe the certainty and uncertainty of ESG-related events, such as achieved, in progress, future, and ambiguous. Achievement indicators describe the degree to which ESG-related events have been achieved, with values ​​ranging from 0 to 1. The higher the value, the higher the degree of achievement; for example, "significant improvement" corresponds to an achievement indicator of 0.8.

[0033] In some embodiments of this application, the steps described above for each statement—including keyword extraction, responsible party extraction, detection of time adverbs and modifiers, and qualitative modifier analysis—to obtain the structured evidence object of the statement include: The first step is to identify multiple core keywords in the statement and form a keyword set.

[0034] For example, a BERT-based keyword extraction model (KeyBERT) can be used to identify multiple core keywords in a statement and integrate all core keywords into a set to obtain a keyword set.

[0035] The second step is to extract the multiple responsible parties involved in the statement and form a subject set.

[0036] For example, the spaCy named entity recognition model can be used to extract multiple responsible entities involved in a statement and integrate all responsible entities into a set to obtain the entity set.

[0037] The third step is to detect time adverbs and modifiers in the sentence to obtain time status labels.

[0038] For example, event expression recognition tools (such as HeidelTime) can be used to detect time adverbs and modifiers in a statement to obtain time status labels.

[0039] The fourth step is to establish an indicator of the degree of realization of the statement based on the qualitative modifiers in the statement.

[0040] For example, a model for analyzing semantic sentiment intensity (RoBERTa, Robustlyoptimized BERT approach) can be used to establish an achievement index for a statement based on qualitative modifiers (such as significant improvement, industry leadership, etc.). For instance, the statement is input into the RoBERTa model, which performs sentiment intensity analysis on the qualitative modifiers in the statement and outputs the corresponding achievement index.

[0041] The fifth step is to integrate the keyword set, subject set, time status label, and achievement level indicator into a single data set to obtain the structured evidence object of the statement.

[0042] Step 13: For each statement, calculate the initial greenwash confidence score based on the statement's implementation level index, and determine the statement's greenwash confidence score based on the initial greenwash confidence score, core keywords, responsible parties, and time status labels.

[0043] The above confidence level for greenwashing describes the probability of a statement being greenwashed; the higher the value, the higher the probability of greenwashing.

[0044] In some embodiments of this application, the steps of calculating the initial greenwash confidence of a statement based on the statement's implementation level index, and determining the statement's greenwash confidence based on the initial greenwash confidence, core keywords, responsible parties, and time status tags, include: The first step is to calculate the consistency index based on the implementation degree index in the structured evidence object of the statement.

[0045] Specifically, through the formula: ; ; ; Calculate the consistency index .

[0046] in, Represents the weighted geometric mean. This represents the weighted arithmetic mean. , , All are weighting coefficients. Indicates the credibility of an entity. Indicates time determinism, Indicators representing the degree of achievement.

[0047] For a company to make a credible statement, its three dimensions—E, T, and D—should be consistent and mutually supportive. If one dimension (such as the D value for "industry leadership") is extremely high, while other dimensions (such as the T value for "future" or the E value for "supply chain") are extremely low, then there is an internal contradiction in the statement, and the risk of greenwashing is high.

[0048] For the indicator part, considering the weighted geometric mean ( It is extremely sensitive to low values; any "weakness" in any dimension will significantly lower the value. Value; while the weighted arithmetic mean ( This reflects the overall level. The ratio between the two perfectly depicts this "coordination" or "consistency." When all three are balanced and score highly, Approximately equal to Consistency indicators →1; When there is a significant weakness. much smaller , →0.

[0049] The design of the aforementioned consistency indicators can capture nonlinear synergistic effects, assessing not only whether each dimension is "good" or "bad," but also whether they are "compatible." For example, a "grand promise" (high D value) to the "future" (low T value) made by the "supply chain" (low E value) will have a high consistency indicator. It will be very low, because The low values ​​of E and T significantly lower the risk profile, thus revealing its high risk. Linear weighting, on the other hand, may give a moderate score due to a high D value, failing to reveal this inherent risk structure.

[0050] For example, weight parameters can be calculated using algorithms such as sensitivity analysis and fixed-point iteration. , , Specifically, at the reference point To ensure that the standardized elasticities in the three dimensions are equal, the nonlinear constraint equation can be obtained: ; in, λ is a constant. Representing characteristic components, Indicates the weighted components, This represents the dot product of the weight component and the feature component.

[0051] Construct an iterative function based on the above conditions: ; in, >0 indicates a smoothing term. Indicates unnormalized weights , Indicates unnormalized weights , , This represents the new weights after normalization.

[0052] The formula for weight iteration and update is: ; in, This represents the weight vector for the next iteration. Indicates the damping factor. This represents the weight vector for the current iteration. This represents the weight update direction vector.

[0053] When the arithmetic mean has a positive lower bound and a reasonable smoothing term is chosen, the above iteration is a compression mapping. According to Banach's fixed-point theorem, it can be proven that the iteration converges to a unique fixed point from any initial value. ,in This represents the final set of weight coefficients. , , All of these are the final weighting coefficients.

[0054] Considering that fixed weights cannot adapt to the inherent characteristics of different statements, the principle of "equal elasticity" from economics is adopted. Therefore, the model design uses weight allocation to ensure that the "standardized elasticity" of each dimension is equal, meaning that under optimal weights, the relative change of any dimension has a balanced impact on the overall system (weighted average). This ensures that the weight allocation is not subjective but driven by the structure of the data itself. Meanwhile, since the nonlinear constraint equations in the formula have no analytical solution, this fixed-point iterative algorithm is designed for numerical solution. (Smoothing term) To ensure numerical stability, damping factor The convergence speed is controlled by dynamically calculating the (E,T,D) values ​​based on each specific statement, achieving personalized evaluation "tailored to the statement". Most importantly, considering theoretical rigor, the optimization objective is based on "equal elasticity", and the uniqueness and stability of iterative convergence are guaranteed by Banach's fixed-point theorem, avoiding the arbitrariness of heuristic weighting.

[0055] The second step is to iteratively update the consistency index to obtain the final consistency index.

[0056] First, the achievement level indicators are updated based on the consistency indicators.

[0057] Then, determine whether the updated implementation level index meets the stability condition.

[0058] For example, the stability condition mentioned above can be that the difference between the implementation level indicators before and after the update is less than a preset difference.

[0059] If so, the consistency metric will be used as the final consistency metric.

[0060] Otherwise, use the updated implementation metrics as the implementation metrics and return to the steps of calculating the consistency metrics based on the implementation metrics in the structured evidence object of the statement.

[0061] Specifically, through the formula: ; Calculate the updated implementation metrics .

[0062] in, This represents the equilibrium constant.

[0063] Considering the initial value of "degree of implementation" in the statement Analysis based solely on qualitative modifiers is static and isolated; it requires dynamic adjustments based on its compatibility with the E and T dimensions. Therefore, we should consider... The value is transformed from a static extraction result into a dynamically evolving state variable under consistent logical constraints, enabling the evaluation system to learn and correct itself. This achieves deep information integration across the three dimensions of E, T, and D, rather than simply listing them side-by-side.

[0064] New achievement indicators Historical value A weighted average with new evidence. As a new source of evidence, among which This reflects the consistency of the current (E,T,D) combination. It is a certainty of time. This means that no matter how beautifully a statement is described ( (High), if it is inconsistent with entity and time information ( If the time complexity is low (T is low) or the time ambiguity is low (T is low), then its true "realizability" should be lowered. A mild nonlinear function—the square root function—is introduced to prevent this. Or a small change in T causes To mitigate drastic fluctuations and ensure iterative stability.

[0065] The third step is to calculate the initial greenwash confidence level of the statement based on the eventual consistency index.

[0066] Specifically, through the formula: ; Calculate the initial confidence level of green bleaching .

[0067] in, This represents the eventual consistency metric.

[0068] The fourth step is to determine whether the initial confidence level for green bleaching is less than the first confidence level but greater than or equal to the second confidence level.

[0069] The first confidence level is lower than the second confidence level. For example, if the first confidence level is 0.4, the second confidence level is 0.6.

[0070] If so, the initial greenwash confidence score will be used as the greenwash confidence score of the statement.

[0071] Otherwise, the statement's greenwash confidence level is obtained by re-evaluating the keyword set, subject set, and time status label in the structured evidence object.

[0072] The re-evaluation steps described above are as follows: generating a search query based on the keyword set, subject set, and time status label of the statement; retrieving multiple candidate evidence statements from the statement database based on the search query; calculating the similarity between the structured evidence object of the statement and the structured evidence object of each candidate evidence statement, and taking the candidate evidence statement with the highest similarity as the target evidence statement; and calculating the greenwash confidence of the statement based on the target evidence statement.

[0073] The above search query is: ; in, Indicates the search query. Represents a set of keywords. Represents a set of subjects. Indicates a time status label.

[0074] The aforementioned statement database contains a large number of statements from ESG report texts. A search is conducted within this database using the defined search terms, and the retrieved statements are selected as candidate evidence statements. Similarity calculation methods, such as cosine similarity, can be used to calculate the similarity between the structured evidence object of each statement and the structured evidence object of each candidate evidence statement.

[0075] The steps for calculating the greenwash confidence of a statement based on the target evidence statement are as follows: Through the formula: ; Calculate the confidence level of green bleaching .

[0076] in, Indicators of consistency based on the target evidence statement: ; ; ; in, This represents the weighted geometric mean of the statements based on the target evidence. This represents the weighted arithmetic mean of the statements based on the target evidence. Indicates a stable term. , , All are weighting coefficients. The entity credibility of the target evidence statement. This indicates the temporal certainty of the target evidence statement. Indicators representing the degree to which the target evidence statement is achieved.

[0077] For example, if the reassessed greenwash confidence level still does not meet the condition of being less than the first confidence level and greater than or equal to the second confidence level, it is marked as "requiring further audit", the corresponding target evidence statement is retained, and it is reassessed manually or by other means.

[0078] This formula is used to introduce external target evidence statements for secondary evaluation when the initial judgment is in the "hesitation zone." Different evaluation benchmarks are used. For the main entity evaluation, it is based on the statements in the company's own ESG report, calculating their internal consistency. Then, a re-evaluation is conducted based on verified or recognized evidence statements retrieved from external databases. At this point, the evaluation objective changes from the company's own statements to the credibility of the evidence.

[0079] Furthermore, considering the diverse sources of external evidence databases, their structured extraction results (E', T', D') may contain noise or have boundary conditions, so a stability term is introduced. To prevent numerical instability caused by an excessively small denominator and ensure computational robustness, when the credibility of external evidence itself is low (AM' is very small), the presence of ε will significantly reduce the consistency index C', thereby leading to an increase in the greenwash confidence S'.

[0080] Current traditional methods, when introducing external evidence, often calculate text similarity and then directly apply external labels. Therefore, this paper considers a further in-depth credibility deconstruction of the external evidence itself, i.e., deriving C', and using this deconstruction result as the basis for the final judgment, achieving a deeper level of logical reasoning. This ultimately forms a dual verification system of "internal consistency check" + "external credibility benchmarking." Only when the internal statement is logically consistent and there is high-quality supporting external evidence can the statement be considered low-risk, which significantly improves the accuracy of detection.

[0081] Step 14: Calculate the environmental greenwashing confidence of the target company based on the greenwashing confidence of all environmental statements, calculate the social greenwashing confidence of the target company based on the greenwashing confidence of all social statements, and calculate the governance greenwashing confidence of the target company based on the greenwashing confidence of all governance statements.

[0082] The aforementioned environmental greenwashing confidence level is used to describe the probability of greenwashing of environmental information in ESG report texts; the aforementioned social greenwashing confidence level is used to describe the probability of greenwashing of social information in ESG report texts; and the aforementioned governance greenwashing confidence level is used to describe the probability of greenwashing of governance information in ESG report texts.

[0083] Specifically, through the formula: ; ; ; Calculate the confidence level of environmental greening ; in, Indicates the first The weight of each environment statement. This represents the set of numbers for environment statements. Indicates the first Greenwash confidence level of an environmental statement Indicates the first The initial weight of each environment statement. Indicates the first The initial weight of each environment statement. Indicates the first Classification confidence of each environment statement, Indicates the first Structured extraction confidence of each environment statement Indicates the first Similarity of external evidence for each context statement.

[0084] It should be noted that the formula for calculating the social greenwashing confidence level is the same as the formula for calculating the environmental greenwashing confidence level. That is, the weights and greenwashing confidence levels of all social statements are substituted into the above formula to obtain the social greenwashing confidence level. Similarly, the formula for calculating the governance greenwashing confidence level is the same as the formula for calculating the environmental greenwashing confidence level. That is, the weights and greenwashing confidence levels of all governance statements are substituted into the above formula to obtain the governance greenwashing confidence level.

[0085] Step 15: Generate the target company's greenwashing detection results based on environmental greenwashing confidence, social greenwashing confidence, and governance greenwashing confidence.

[0086] The above-mentioned greenwashing test results are used to describe the probability of a target company's greenwashing in three aspects: environmental, social, and governance. This may include confidence levels for environmental greenwashing, social greenwashing, and governance greenwashing.

[0087] Specifically, environmental greenwashing confidence level, social greenwashing confidence level, and governance greenwashing confidence level are used as basic information to integrate and obtain the greenwashing detection results for the target company. Furthermore, corresponding consistency indicators and relevant target evidence statements can be recorded in the greenwashing detection results to improve the information richness and reliability of the results.

[0088] It is worth mentioning that semantic recognition of statements can effectively analyze the semantic information of ESG report text, revealing deeper features such as the subject, time status, and degree of implementation in the text. Based on structured evidence objects, greenwashing confidence is calculated, ensuring information richness and depth, improving the accuracy of greenwashing confidence, and thus effectively improving the credibility of greenwashing detection for enterprises.

[0089] In addition, this application introduces a mechanism for supplementing external evidence to improve the credibility and auditability of greenwashing detection when the judgment is ambiguous or the evidence is insufficient.

[0090] The following is an exemplary description of the enterprise greenwashing detection device for structured information extraction provided in this application.

[0091] like Figure 2 As shown, this application embodiment provides a structured information extraction device for detecting corporate greenwashing. The structured information extraction device 200 for detecting corporate greenwashing includes: The acquisition module 201 is used to acquire the ESG report text of the target enterprise and extract multiple statements from the ESG report text; the statements are environmental statements, social statements, or governance statements. The semantic recognition module 202 is used to extract keywords, extract responsible parties, detect time adverbs and modifiers, and analyze qualitative modifiers for each statement to obtain the structured evidence object of the statement. The structured evidence object includes the core keywords, responsible parties, time status labels, and achievement level indicators of the statement. The first calculation module 203 is used to calculate the initial greenwash confidence of each statement based on the statement's implementation degree index, and to determine the greenwash confidence of the statement based on the initial greenwash confidence, core keywords, responsible parties, and time status labels. The second calculation module 204 is used to calculate the environmental greenwashing confidence of the target enterprise based on the greenwashing confidence of all environmental statements, calculate the social greenwashing confidence of the target enterprise based on the greenwashing confidence of all social statements, and calculate the governance greenwashing confidence of the target enterprise based on the greenwashing confidence of all governance statements. The generation module 205 is used to generate the target company's greenwashing detection results based on environmental greenwashing confidence, social greenwashing confidence, and governance greenwashing confidence.

[0092] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0093] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0094] like Figure 3 As shown, an embodiment of this application provides a terminal device, wherein the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 3 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0095] Specifically, when the processor D100 executes the computer program D102, it acquires the ESG report text of the target company and extracts multiple statements from the ESG report text. Then, for each statement, it performs keyword extraction, responsibility subject extraction, time adverbial and modifying component detection, and qualitative modifier analysis to obtain the structured evidence object of the statement. Then, for each statement, it calculates the initial greenwashing confidence score based on the statement's realization degree index. Based on the initial greenwashing confidence score, core keywords, responsibility subjects, and time status labels, it determines the greenwashing confidence score of the statement. Then, it calculates the environmental greenwashing confidence score of the target company based on the greenwashing confidence scores corresponding to all environmental statements, calculates the social greenwashing confidence score of the target company based on the greenwashing confidence scores corresponding to all social statements, and calculates the governance greenwashing confidence score of the target company based on the greenwashing confidence scores corresponding to all governance statements. Finally, it generates the target company's enterprise greenwashing detection result based on the environmental greenwashing confidence score, social greenwashing confidence score, and governance greenwashing confidence score. Among them, semantic recognition of statements can effectively analyze the semantic information of ESG report text, revealing deeper features such as subject, time state, and degree of implementation in the text, ensuring information richness and depth. Based on the information in the structured evidence object with information richness and depth, the initial greenwash confidence is calculated and the final greenwash confidence is determined, effectively improving the accuracy of greenwash confidence and thus effectively improving the credibility of greenwash detection for enterprises.

[0096] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0097] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0098] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0099] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0100] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to the structured information extraction enterprise greenwashing detection method device / terminal device, recording media, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0101] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0103] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention.

Claims

1. A method for structured information extraction of enterprise greenwashing detection, characterized in that, The method comprises the following steps: obtaining an ESG report text of a target enterprise, and extracting a plurality of sentences from the ESG report text; the sentences are environmental sentences, social sentences or governance sentences; respectively for each of the sentences, performing keyword extraction, responsibility subject extraction, time adverbial and modification component detection, and qualitative modifier analysis on the sentences to obtain a structured evidence object of the sentences; the structured evidence object comprises a core keyword, a responsibility subject, a time state label and an implementation degree index of the sentence; respectively for each of the sentences, calculating an initial greenwashing confidence of the sentence based on the implementation degree index of the sentence, and determining a greenwashing confidence of the sentence based on the initial greenwashing confidence and the core keyword, the responsibility subject and the time state label; calculating an environmental greenwashing confidence of the target enterprise based on the greenwashing confidences corresponding to all the environmental sentences, calculating a social greenwashing confidence of the target enterprise based on the greenwashing confidences corresponding to all the social sentences, and calculating a governance greenwashing confidence of the target enterprise based on the greenwashing confidences corresponding to all the governance sentences; generating an enterprise greenwashing detection result of the target enterprise according to the environmental greenwashing confidence, the social greenwashing confidence and the governance greenwashing confidence.

2. The method of claim 1, wherein, The keyword extraction, responsibility subject extraction, time adverbial and modification component detection, and qualitative modifier analysis on the sentences to obtain the structured evidence object of the sentences comprise: identifying a plurality of core keywords of the sentence to form a keyword set; extracting a plurality of responsibility subjects involved in the sentence to form a subject set; detecting the time adverbial and modification component in the sentence to obtain a time state label; establishing an implementation degree index of the sentence according to the qualitative modifier in the sentence; integrating the keyword set, the subject set, the time state label and the implementation degree index into one data to obtain the structured evidence object of the sentence.

3. The method of claim 2, wherein, The calculation of the initial greenwashing confidence of the sentence based on the implementation degree index and the determination of the greenwashing confidence of the sentence based on the initial greenwashing confidence and the core keyword, the responsibility subject and the time state label comprise: calculating a consistency index according to the implementation degree index in the structured evidence object of the sentence; iteratively updating the consistency index to obtain a final consistency index; calculating the initial greenwashing confidence of the sentence based on the final consistency index; judging whether the initial greenwashing confidence is less than a first confidence and greater than or equal to a second confidence; the first confidence is less than the second confidence; if yes, taking the initial greenwashing confidence as the greenwashing confidence of the sentence; otherwise, reevaluating the greenwashing confidence of the sentence by using the keyword set, the subject set and the time state label in the structured evidence object of the sentence.

4. The method of claim 3, wherein, The calculation of the consistency index according to the implementation degree index in the structured evidence object of the sentence comprises: by the formula: ； ； ； Computing a consistency index ; wherein, denotes a weighted geometric mean, denotes a weighted arithmetic mean, , , are weight coefficients, denotes an entity trustworthiness, denotes a temporal certainty, denotes an implementation degree indicator; The iterative updating of the consistency index to obtain the final consistency index comprises: updating the implementation degree index based on the consistency index; determining whether the updated implementation degree index satisfies a stability condition; if yes, taking the consistency index as a final consistency index; otherwise, taking the updated implementation degree index as the implementation degree index, and returning to the step of calculating a consistency index according to an implementation degree index in a structured evidence object of the sentence.

5. The method of claim 4, wherein, The updating of the implementation degree index based on the consistency index comprises: by the formula: ； calculating an updated achievement indicator ; wherein represents the equilibrium constant.

6. The method of detecting a greenwashing of a business according to claim 3, wherein, The calculation of the initial green drift confidence of the sentence based on the final consistency index comprises: by the formula: ； Computing initial green-up confidence ; wherein, represents the final consistency index; The re-evaluation using the keyword set, the subject set and the time state label in the structured evidence object of the sentence to obtain the green drift confidence of the sentence comprises: generating a search formula based on the keyword set, the subject set and the time state label of the sentence; searching for a plurality of candidate evidence sentences from a sentence library based on the search formula; calculating the similarity between the structured evidence object of the sentence and the structured evidence object of each candidate evidence sentence, and taking the candidate evidence sentence with the largest similarity as a target evidence sentence; calculating the green drift confidence of the sentence based on the target evidence sentence.

7. The method of detecting a greenwashing of a business according to claim 6, wherein, The calculation of the green drift confidence of the sentence based on the target evidence sentence comprises: by the formula: ； Computing green shift confidence ; wherein, represents a consistency indicator based on the target evidence sentence: ； ； ； wherein, denotes a weighted geometric mean based on the target evidence sentence, denotes a weighted arithmetic mean based on the target evidence sentence, denotes a stability term, , , are weight coefficients, denotes an entity credibility of the target evidence sentence, denotes a time certainty of the target evidence sentence, denotes an implementation degree indicator of the target evidence sentence.

8. The method of detecting a greenwashing of a business according to claim 7, wherein, The calculation of the environmental green drift confidence of the target enterprise based on the green drift confidence corresponding to all environmental sentences comprises: by the formula: ； ； ； Computing environment drift confidence ; wherein, represents a weight of the th environment statement, represents a set of numbers of environment statements, represents a drift confidence of the th environment statement, represents an initial weight of the th environment statement, represents an initial weight of the th environment statement, represents a classification confidence of the th environment statement, represents a structured extraction confidence of the th environment statement, represents an external evidence similarity of the th environment statement.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the structured information extraction enterprise green drift detection method of any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the structured information extraction enterprise green drift detection method of any one of claims 1 to 8.

Images