Word embedding similarity-based lexicon dynamic expansion and ESG expression quantitative evaluation method

Abstract

The invention discloses a word embedding similarity-based lexicon dynamic expansion and ESG expression quantitative evaluation method, and relates to the technical field of natural language processing and enterprise social responsibility evaluation. Comprising the steps of preprocessing obtained enterprise annual report unstructured text data; vectorizing the preprocessed data set, and dividing the vectorized data set into training set data and test set data; performing training set data word screening and classification by utilizing a cosine similarity method so as to construct E, S and G word libraries; calculating the similarity between the test set data and the E, S and G word libraries to obtain the ESG score of the enterprise; and constructing an enterprise ESG performance radar map according to the ESG score, and displaying the ESG performance of the enterprise. According to the method, the ESG performance of the enterprise is evaluated by applying the unstructured data, so that a method system for evaluating the ESG performance is greatly enriched, and a brand new analysis view angle is provided for predicting the comprehensive strength of the enterprise.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing and corporate social responsibility assessment, and in particular to a method for dynamic expansion of a vocabulary based on word embedding similarity and quantitative assessment of ESG performance. Background Art

[0002] The environmental, social and corporate governance performance (ESG, where E stands for environment, S stands for society and G stands for governance) of an enterprise is an investment philosophy and corporate evaluation standard that focuses mainly on the environmental, social and governance performance of an enterprise rather than its financial performance. It is a methodology for evaluating the sustainable development of an enterprise. As environmental challenges such as global warming and natural resource shortages have increased the attention of countries around the world to sustainable development, policymakers and regulators in various countries have increasingly attached importance to the implementation of corporate environmental governance and social responsibility. In order to promote sustainable development, the country is increasingly aware of the importance of ESG, and research on it has become a hot topic. However, at present, there is no unified standard for evaluating the ESG performance of an enterprise, and the data types of the ESG performance of an enterprise are complex, with structured and unstructured data merged together, and cannot be uniformly evaluated. For the processing of structured data (such as numerical data, tabular data, etc.), traditional data analysis methods are sufficient to cope with it, but when processing unstructured data such as text, traditional methods are difficult to capture the relationship between information, resulting in inaccurate scoring and affecting investor judgment. In response to the above problems, the present invention introduces a deep learning model to vectorize unstructured data and accurately evaluate the ESG performance of an enterprise.

[0003] In evaluating corporate ESG performance, how to expand the seed word library of ESG information and construct the ESG score of the company based on the similarity between the information disclosed in the company's annual report and the word library has become an important research topic. To address this issue, the integration of Embedding and word vector similarity (semantic closeness index, word embedding similarity index, word vector correlation score) technology provides a new perspective for processing text soft information.

[0004] Embedding technology can map high-dimensional discrete data (text data) to low-dimensional continuous vector space while retaining the structure and semantic information between data. Word vectors can represent the intrinsic connection between words, that is, the word vectors of two words can be calculated to obtain their degree of association. It is particularly suitable for expanding keyword lexicons and has become a mainstream technology in fields such as natural language processing.

[0005] Therefore, it is an urgent problem for those skilled in the art to propose a method for dynamic expansion of the vocabulary and quantitative evaluation of ESG performance based on word embedding similarity to solve the difficulties existing in the prior art. Summary of the invention

[0006] In view of this, the present invention provides a method for dynamic expansion of the vocabulary and quantitative evaluation of ESG performance based on word embedding similarity. By using unstructured data to evaluate the ESG performance of enterprises, it greatly enriches the method system of ESG performance evaluation and provides a new analytical perspective for predicting the comprehensive strength of enterprises.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A method for dynamically expanding a vocabulary and quantitatively evaluating ESG performance based on word embedding similarity includes the following steps:

[0009] S1. Data acquisition: Collect unstructured text data of corporate annual reports;

[0010] S2. Data preprocessing: preprocess the unstructured text data of the annual reports to obtain a preprocessed data set;

[0011] S3. Vectorization processing: vectorize the preprocessed data set, convert each word in the preprocessed data set into a word vector, and divide the vectorized data set into training set data and test set data;

[0012] S4. Construct the vocabulary: Use the cosine similarity method to screen and classify the words in the training set data to construct the E, S, and G vocabulary;

[0013] S5. Calculate the score: Calculate the similarity between the test set data and the E, S, and G vocabulary to obtain the ESG score of the enterprise;

[0014] S6. Data visualization: Build a radar chart of corporate ESG performance based on ESG scores to display the company’s ESG performance.

[0015] Optionally, S1 collects unstructured document text data and report text data of corporate annual reports.

[0016] Optionally, S2 cleans and filters the collected unstructured text data, and uses natural language processing tools to extract and process the content of the unstructured text data.

[0017] Optionally, S3 uses an embedding method based on a word2vec model to perform word embedding on the preprocessed dataset and convert each word into a word vector.

[0018] Optionally, the specific structure of the word2vec model includes one-hot encoding, initialization word vector, input layer and embedding layer;

[0019] One-hot encoding is used to count the number of all words in the corpus and one-hot encode each word in sequence;

[0020] Initialize word vectors to map one-hot encoded neural network training into vectors of fixed dimensions;

[0021] Input layer, which takes the one-hot encoding of the context words as input;

[0022] The embedding layer is used to convert high-dimensional, sparse word vectors into low-dimensional, dense word vectors. There is an embedding matrix in the embedding layer, and the matrix size is N*V, where N is the corpus size and V is the vector dimension size.

[0023] Optionally, in S3., the vectorized data set is divided into training set data and test set data in a ratio of 7:3.

[0024] Optionally, in S4, the cosine similarity method is used to screen and classify the words in the training set data, so as to construct the E, S, and G vocabulary. The specific contents are:

[0025] The degree of association between words is expressed by calculating cosine similarity. Cosine similarity is based on the vector space model and measures the similarity between two vectors by calculating the cosine value of the angle between them. In the word vector space, the smaller the angle between two word vectors, the closer their cosine similarity is to 1, indicating that the two words are more semantically similar.

[0026] Randomly take two word vectors A and B, the word vector of A is represented as [A1, A2, A3…A n ], the word vector of B is represented as [B1,B2,B3…B n ], for word vectors A and B, the cosine similarity calculation formula is as follows:

[0027]

[0028] Words with high similarity are selected from the corpus, and the words are manually selected and classified into the E, S, and G vocabulary, thereby constructing the E, S, and G vocabulary.

[0029] Optionally, S5 calculates the similarity between the test set data and the E, S, and G vocabulary, and obtains the specific content of the enterprise's ESG score:

[0030] The ESG dictionary is used as a quantitative tool to conduct quantitative analysis on the three dimensions of E, S and G, and the three dimensions of E, S and G are labeled for training the Finbert model, where E stands for environment and is labeled 0, S stands for society and is labeled 1, and G stands for governance and is labeled 2;

[0031] First, we calculate the different frequencies of occurrence of words related to each dimension in the annual report of the enterprise, and normalize the frequencies to the size of the entire vocabulary list. The formula for measuring the performance of enterprise i in various aspects of ESG is as follows:

[0032]

[0033] Where c(t) i is the number of E, S, and G related words appearing in the annual report of enterprise i, and c(n) is the total length of the vocabulary;

[0034] Secondly, the company’s annual report is input into the FinBert model to obtain the company’s ESG score, and the company’s ESG score is calculated based on a multi-angle integrated ranking algorithm.

[0035] Optionally, when evaluating the ESG performance of a company, the company is also categorized by labels. The labels are divided into Category 1 and Category 0. Category 1 indicates that the company is in ST status and is marked as 1; Category 0 indicates that the company is in normal status and is marked as 0.

[0036] It can be seen from the above technical solution that, compared with the prior art, the present invention provides a method for dynamic expansion of the vocabulary and quantitative evaluation of ESG performance based on word embedding similarity, which has the following beneficial effects:

[0037] (1) The present invention uses unstructured data to evaluate the ESG performance of enterprises, which greatly enriches the method system of ESG performance evaluation and provides a new analytical perspective for predicting the comprehensive strength of enterprises. Specifically, the present invention breaks through the limitation that traditional ESG evaluation mainly relies on structured data and innovatively introduces unstructured data. Unstructured data often contains rich ESG-related information, but it is difficult to capture and analyze it through traditional quantitative methods. By using advanced natural language processing, text mining and other technologies, the present invention successfully converts these unstructured data into quantifiable ESG indicators, thereby achieving a comprehensive and in-depth evaluation of the ESG performance of enterprises;

[0038] (2) The present invention not only focuses on the specific performance of enterprises in the three dimensions of environment, society and governance, but also helps to more comprehensively evaluate the overall performance of enterprises in sustainable development by comprehensively considering the interactions and impacts among these dimensions, avoiding one-sided emphasis on one dimension and neglecting other important aspects, thereby providing a more comprehensive and accurate analytical perspective for predicting the comprehensive strength of enterprises. It not only helps investors, consumers and other stakeholders to have a more comprehensive understanding of the ESG performance of enterprises, but also provides strong data support for the formulation of sustainable development strategies of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 A flow chart of a method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity provided by the present invention;

[0041] Figure 2 The word2vec model structure diagram provided by the present invention;

[0042] Figure 3 A flowchart for quantitative evaluation of enterprise ESG performance provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Reference Figure 1 As shown, the present invention discloses a method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity, comprising the following steps:

[0045] S1. Data acquisition: Collect unstructured text data of corporate annual reports;

[0046] S2. Data preprocessing: preprocess the unstructured text data of the annual reports to obtain a preprocessed data set;

[0047] S3. Vectorization processing: vectorize the preprocessed data set, convert each word in the preprocessed data set into a word vector, and divide the vectorized data set into training set data and test set data;

[0048] S4. Construct the vocabulary: Use the cosine similarity method to screen and classify the words in the training set data to construct the E, S, and G vocabulary;

[0049] S5. Calculate the score: Calculate the similarity between the test set data and the E, S, and G vocabulary to obtain the ESG score of the enterprise;

[0050] S6. Data visualization: Build a radar chart of corporate ESG performance based on ESG scores to display the company’s ESG performance.

[0051] Furthermore, S1 collects unstructured document text data and report text data of corporate annual reports.

[0052] Furthermore, S2 cleans and filters the collected unstructured text data, and uses natural language processing tools to extract and process the content of the unstructured text data.

[0053] Furthermore, S3 uses the embedding method based on the word2vec model to perform word embedding on the preprocessed dataset and convert each word into a word vector.

[0054] Furthermore, the specific structure of the word2vec model includes one-hot encoding, initialization word vector, input layer and embedding layer;

[0055] One-hot encoding is used to count the number of all words in the corpus and one-hot encode each word in sequence;

[0056] Initialize word vectors to map one-hot encoded neural network training into vectors of fixed dimensions;

[0057] Input layer, which takes the one-hot encoding of the context words as input;

[0058] The embedding layer is used to convert high-dimensional, sparse word vectors into low-dimensional, dense word vectors. There is an embedding matrix in the embedding layer, and the matrix size is N*V, where N is the corpus size and V is the vector dimension size.

[0059] Furthermore, in S3., the vectorized data set is divided into training set data and test set data in a ratio of 7:3.

[0060] Furthermore, in S4, the cosine similarity method is used to screen and classify the words in the training set data, so that the specific contents of constructing the E, S, and G vocabulary are as follows:

[0061] The degree of association between words is expressed by calculating cosine similarity. Cosine similarity is based on the vector space model and measures the similarity between two vectors by calculating the cosine value of the angle between them. In the word vector space, the smaller the angle between two word vectors, the closer their cosine similarity is to 1, indicating that the two words are more semantically similar.

[0062] Randomly take two word vectors A and B, the word vector of A is represented as [A1, A2, A3…A n ], the word vector of B is represented as [B1,B2,B3…Bn ], for word vectors A and B, the cosine similarity calculation formula is as follows:

[0063]

[0064] Words with high similarity are selected from the corpus, and the words are manually selected and classified into the E, S, and G vocabulary, thereby constructing the E, S, and G vocabulary.

[0065] Furthermore, S5 calculates the similarity between the test set data and the E, S, and G word libraries, and obtains the specific content of the company's ESG score:

[0066] The ESG dictionary is used as a quantitative tool to conduct quantitative analysis on the three dimensions of E, S and G, and the three dimensions of E, S and G are labeled for training the Finbert model, where E stands for environment and is labeled 0, S stands for society and is labeled 1, and G stands for governance and is labeled 2;

[0067] First, we calculate the different frequencies of occurrence of words related to each dimension in the annual report of the enterprise, and normalize the frequencies to the size of the entire vocabulary list. The formula for measuring the performance of enterprise i in various aspects of ESG is as follows:

[0068]

[0069] Where c(t) i is the number of E, S, and G related words appearing in the annual report of enterprise i, and c(n) is the total length of the vocabulary;

[0070] Secondly, the company’s annual report is input into the FinBert model to obtain the company’s ESG score, and the company’s ESG score is calculated based on a multi-angle integrated ranking algorithm.

[0071] Specifically, the specific calculation process of the multi-angle integrated ranking algorithm is as follows: first, three different company rankings (three ESG categories) are obtained based on the similarity matching between the company's annual report and the vocabulary, and the three rankings are summarized using the Borda count to obtain a final ranking. This final ranking contains the information of the three ESG rankings, and the score of each company obtained by the Borda count is the ESG score of the company.

[0072] Furthermore, when evaluating the ESG performance of enterprises, the enterprises are also classified by labels. The labels are divided into category 1 and category 0. Category 1 indicates that the enterprise is in ST status and is marked as 1; category 0 indicates that the enterprise is in normal status and is marked as 0.

[0073] In a specific embodiment, Figure 3 As shown, including the following

[0074] S1. Data acquisition: Collect unstructured text data of corporate annual reports;

[0075] S2. Data preprocessing: preprocess the unstructured text data of the annual reports to obtain a preprocessed data set;

[0076] Clean and filter the collected unstructured text data, and use natural language processing (NLP) tools (such as the pdfplumber library in Python) to extract and preprocess the text content in the report. In the process of text processing, the jieba word segmentation tool and Baidu + Harbin Institute of Technology + Machine Intelligence Laboratory stop word library are used to improve the quality and efficiency of unstructured text data analysis;

[0077] First, a function named clean_illegal_chars is defined to remove illegal characters. Next, a load_stopwords function is defined to read the Baidu + Harbin Institute of Technology + Machine Intelligence Laboratory stop word library file, convert it into a Pandas DataFrame object, and extract the stop word column and convert it into a collection. The use of a collection can speed up the check of stop words. Finally, jieba is used for word segmentation to convert continuous text sequences into discrete word sequences. For example, "Continue to promote the company's ESG governance system" is split into "Continue", "Promote", "Company", "ESG", "Governance", and "System". Stop words are removed and word frequency is counted during the word segmentation process, and the word segmentation results and word frequency are saved. In this way, a purified word list is obtained as the final data set;

[0078] S3. Vectorization processing: vectorize the preprocessed data set, convert each word in the preprocessed data set into a word vector, and divide the vectorized data set into training set data and test set data;

[0079] like Figure 2 As shown in the figure, based on the word2vec model, the embedding method is used to perform word embedding processing on the preprocessed data set, and each word is converted into a word vector; the specific structure of the word2vec model includes one-hot encoding, initialization word vector, input layer and embedding layer;

[0080] One-hot encoding is used to count the number of all words in the corpus and one-hot encode each word in sequence;

[0081] Initialize the word vector, which is used to map the one-hot encoded neural network training into a vector of fixed dimension. The present invention selects a fixed dimension of 300;

[0082] The input layer takes the unique hot encoding of the context words as input, and sets the context window length in advance. The window length set in the present invention is 2, that is, the context words are the first two and the last two of the central prediction word, a total of four;

[0083] The embedding layer is used to convert high-dimensional, sparse word vectors into low-dimensional, dense word vectors. There is an embedding matrix in the embedding layer, and the matrix size is N*V, where N is the corpus size and V is the vector dimension size. For the input of a single word, a row is found in the embedding matrix, which is the word vector of the word. The window length selected by the present invention is 2, so the context of the predicted central word includes four words. These four words will be input into the embedding layer at the same time, and each word will be converted into a word vector, namely v1, v2, v3, and v4. Then the output result of the embedding layer is v=(v1+v2+v3+v4) / 4;

[0084] The vectorized data set is divided into training set data and test set data in a ratio of 7:3;

[0085] S4. Construct the vocabulary: Use the cosine similarity method to screen and classify the words in the training set data to construct the E, S, and G vocabulary;

[0086] The degree of association between words is expressed by calculating cosine similarity. Cosine similarity is based on the vector space model and measures the similarity between two vectors by calculating the cosine value of the angle between them. In the word vector space, the smaller the angle between two word vectors, the closer their cosine similarity is to 1, indicating that the two words are more semantically similar.

[0087] Randomly take two word vectors A and B, the word vector of A is represented as [A1, A2, A3…A n ], the word vector of B is represented as [B1,B2,B3…B n ], for word vectors A and B, the cosine similarity calculation formula is as follows:

[0088]

[0089] Select words with high similarity in the corpus, manually select and classify the words into E, S, and G word libraries, and thus construct E, S, and G word libraries;

[0090] S5. Calculate the score: Calculate the similarity between the test set data and the E, S, and G vocabulary to obtain the ESG score of the enterprise;

[0091] The ESG dictionary is used as a quantitative tool to conduct quantitative analysis on the three dimensions of E, S and G, and the three dimensions of E, S and G are labeled for training the Finbert model, where E stands for environment and is labeled 0, S stands for society and is labeled 1, and G stands for governance and is labeled 2;

[0092] First, we calculate the different frequencies of occurrence of words related to each dimension in the annual report of the enterprise, and normalize the frequencies to the size of the entire vocabulary list. The formula for measuring the performance of enterprise i in various aspects of ESG is as follows:

[0093]

[0094] Where c(t) i is the number of E, S, and G related words appearing in the annual report of enterprise i, and c(n) is the total length of the vocabulary;

[0095] Secondly, the company’s annual report is input into the FinBert model to obtain the company’s ESG score, and the company’s ESG score is calculated based on a multi-angle integrated ranking algorithm.

[0096] S6. Data visualization: Build a radar chart of corporate ESG performance based on ESG scores to display the company’s ESG performance;

[0097] Finally, a radar chart of the company's ESG performance is constructed based on the scores to intuitively display the company's ESG performance.

[0098] When evaluating the ESG performance of an enterprise, the present invention also classifies the enterprise by labels, and the labels are divided into category 1 and category 0. Category 1 indicates that the enterprise is in ST status and is marked as 1; category 0 indicates that the enterprise is in normal status and is marked as 0.

[0099] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0100] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity, characterized in that: The following steps are involved: S1. Data acquisition: Collect unstructured text data of corporate annual reports; S2. Data preprocessing: preprocess the unstructured text data of the annual reports to obtain a preprocessed data set; S3. Vectorization processing: vectorize the preprocessed data set, convert each word in the preprocessed data set into a word vector, and divide the vectorized data set into training set data and test set data; S4. Construct the vocabulary: Use the cosine similarity method to screen and classify the words in the training set data to construct the E, S, and G vocabulary; S5. Calculate the score: Calculate the similarity between the test set data and the E, S, and G vocabulary to obtain the ESG score of the enterprise; S6. Data visualization: Build a radar chart of corporate ESG performance based on ESG scores to display the company’s ESG performance.

2. According to claim 1, a method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity is characterized in that: S1 collects unstructured document text data and report text data of corporate annual reports.

3. According to claim 1, a method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity is characterized in that: In S2, the collected unstructured text data is cleaned and screened, and natural language processing tools are used to extract and process the content of the unstructured text data.

4. According to claim 1, a method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity is characterized in that: In S3, based on the word2vec model, the embedding method is used to perform word embedding on the preprocessed data set, converting each word into a word vector.

5. According to claim 4, a method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity is characterized in that: The specific structure of the word2vec model includes one-hot encoding, initialization word vector, input layer and embedding layer; One-hot encoding is used to count the number of all words in the corpus and one-hot encode each word in sequence; Initialize word vectors to map one-hot encoded neural network training into vectors of fixed dimensions; Input layer, which takes the one-hot encoding of the context words as input; The embedding layer is used to convert high-dimensional, sparse word vectors into low-dimensional, dense word vectors. There is an embedding matrix in the embedding layer, and the matrix size is N*V, where N is the corpus size and V is the vector dimension size.

6. The method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity according to claim 1, characterized in that: In S3., the vectorized data set is divided into training set data and test set data in a ratio of 7:

3.

7. The method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity according to claim 1 is characterized in that: In S4, the cosine similarity method is used to screen and classify the words in the training set data, so as to construct the E, S, and G vocabulary. The specific contents are as follows: The degree of association between words is expressed by calculating cosine similarity. Cosine similarity is based on the vector space model and measures the similarity between two vectors by calculating the cosine value of the angle between them. In the word vector space, the smaller the angle between two word vectors, the closer their cosine similarity is to 1, indicating that the two words are more semantically similar. Randomly take two word vectors A and B, the word vector of A is represented as [A1, A2, A3…A n ], the word vector of B is represented as [B1,B2,B3…B n ], for word vectors A and B, the cosine similarity calculation formula is as follows: Words with high similarity are selected from the corpus, and words are manually selected and classified into E, S, and G vocabulary, thereby constructing the E, S, and G vocabulary.

8. The method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity according to claim 1, characterized in that: In S5, the similarity between the test set data and the E, S, and G word libraries is calculated, and the specific content of the ESG score of the enterprise is obtained as follows: The ESG dictionary is used as a quantitative tool to conduct quantitative analysis on the three dimensions of E, S and G, and the three dimensions of E, S and G are labeled for training the Finbert model, where E stands for environment and is labeled 0, S stands for society and is labeled 1, and G stands for governance and is labeled 2; First, we calculate the different frequencies of occurrence of words related to each dimension in the annual report of the enterprise, and normalize the frequencies to the size of the entire vocabulary list. The formula for measuring the performance of enterprise i in various aspects of ESG is as follows: Where c(t) i is the number of E, S, and G related words appearing in the annual report of enterprise i, and c(n) is the total length of the vocabulary; Secondly, the company’s annual report is input into the FinBert model to obtain the company’s ESG score, and the company’s ESG score is calculated based on a multi-angle integrated ranking algorithm.

9. The method for dynamic expansion of word library and quantitative evaluation of ESG performance based on word embedding similarity according to claim 1, characterized in that: When evaluating the ESG performance of a company, the company is also classified by labels. The labels are divided into Category 1 and Category 0. Category 1 means that the company is in ST status and is marked as 1; Category 0 means the enterprise is in normal state and is marked as 0.

Images