阅读说明：本技术 基于esg的企业评价执行装置及其运转方法 (ESG-based enterprise evaluation execution device and operation method thereof ) 是由 尹悳灿 德伊勒·瓦森达拉 林枝妍 谢尔盖·列克西科夫 于 2017-11-23 设计创作，主要内容包括：本发明公开了一种基于ESG的企业评价执行装置及其运转方法。本发明多样实施例的ESG企业评价装置从ESG(Environmnet,Social,Governance：环境、社会、治理)角度评价企业,算出分数,可以包括：新闻收集部,所述新闻收集部在互联网上收集多个新闻报导,按日期或按企业分类,通过所述新闻报导间的类似度分析,执行对类似度为基准值以上的新闻报导的分类归并；新闻分类部,所述新闻分类部将所述各个新闻报导分类为与环境、社会或治理结构中哪个话题相关；及评价结果导出部,所述评价结果导出部按群集单位,计算相应群集的ESG风险,以计算的值为基础,算出ESG企业评价分数。(The invention discloses an enterprise evaluation execution device based on an ESG and an operation method thereof. The ESG enterprise evaluation apparatus according to various embodiments of the present invention evaluates an enterprise from an ESG (environmental, Social, Governance) perspective, and calculates a score, which may include: a news collecting unit that collects a plurality of news reports on the internet, classifies the news reports by date or by enterprise, and performs classification and consolidation of news reports having a similarity of a reference value or more by similarity analysis among the news reports; a news classification section that classifies the individual news stories as being related to which topic in an environment, society, or governance structure; and an evaluation result deriving unit that calculates an ESG risk for each cluster, and calculates an ESG enterprise evaluation score based on the calculated value.)

1. An ESG enterprise evaluation device that evaluates enterprises from an ESG perspective and calculates a score, comprising:

a news collecting unit that collects a plurality of news reports on the internet, classifies the news reports by date or by enterprise, and performs classification and consolidation of news reports having a similarity of a reference value or more by similarity analysis among the news reports;

a news classification section that classifies the individual news stories as being related to which topic in an environment, society, or governance structure; and

and an evaluation result deriving unit that calculates an ESG risk for each cluster, and calculates an ESG enterprise evaluation score based on the calculated value.

2. The ESG enterprise rating apparatus of claim 1, wherein,

the news gathering unit performs morpheme analysis using a morpheme analyzer corresponding to the news report production language, and performs vectorization of the respective news reports on the basis of morpheme analysis results, the similarity analysis being performed by cosine similarity analysis of the respective news reports.

3. The ESG enterprise rating apparatus of claim 2, wherein,

the news gathering section performs a method of vectorizing the individual news reports by morpheme analysis results, using word frequency-inverse document frequency values.

4. The ESG enterprise rating apparatus of claim 1, wherein,

the news categorizing section categorizes whether the news story is related to at least one of an environment, a society, or an governance structure in a true-false manner before categorizing the respective news story as related to which topic in the environment, the society, or the governance structure.

5. The ESG enterprise rating apparatus of claim 1, wherein,

the news classification section classifies each of the news reports as being related to which topic in an environment, a society, or a management structure, then classifies each of the topics into a refined catalog, and classifies each of the news reports into the catalog.

6. The ESG enterprise rating apparatus of claim 1, wherein,

the news classification section adopts a specific machine learning algorithm to perform learning by training data before classifying each news report as related to which topic in the environment, the society or the governance structure, thereby improving classification ability.

7. The ESG enterprise rating apparatus of claim 6, wherein,

the machine learning algorithm adopted by the news classification part is one of polynomial Bayes, Bernoulli Bayes, random gradient descent, linear support vector classifier, perceptron or random forest.

8. The ESG enterprise rating apparatus of claim 1, wherein,

the evaluation result deriving unit performs classification of nouns extracted from the news report, and performs order setting or evidence level calculation for each directory item in units of clusters based on the frequency of words included in the directory item.

9. The ESG enterprise rating apparatus of claim 8, wherein,

the evaluation result derivation unit calculates the probability that each cluster belongs to a topic of the environment, society, or treatment structure based on the calculated evidence level value.

10. The ESG enterprise rating apparatus of claim 9, wherein,

the evaluation result deriving unit calculates the ESG enterprise evaluation score,

and utilizing the calculated evidence level value and the probability that each cluster belongs to the environment, the society or the treatment structure.

11. A score calculating method of an ESG enterprise evaluation device, which is a method for evaluating an enterprise and calculating a score from the ESG perspective, comprises the following steps:

collecting a plurality of news reports on the internet, classifying the news reports according to dates or enterprises, and performing classification and merging on the news reports with the similarity higher than a reference value through the similarity analysis among the news reports;

a step of classifying each news story as being related to which topic in an environment, society, or governance structure; and

and calculating the ESG risk of the corresponding cluster according to the cluster unit, and calculating the ESG enterprise evaluation score based on the ESG risk.

Technical Field

Various embodiments of the present invention relate to an enterprise evaluation execution apparatus based on an ESG and an operation method thereof, and more particularly, to an apparatus for collecting and analyzing news reports on the internet, analyzing and scoring risks related to environmental, social, or administrative structural topics owned by an enterprise, and an operation method thereof.

Background

Recently, enterprises have paid more attention to risk management, and simultaneously, have evaluated their own enterprises and other enterprises in multiple angles, and have applied such evaluation results to risk management of investment, co-procurement, production lines, and the like.

Generally, enterprises are generally evaluated based on financial data that can be obtained quantitatively, but recently, methods for evaluating enterprises and analyzing risks based on non-financial data have been earnestly developed. In the financial data disclosed by the enterprise, although the content unfavorable to the corresponding enterprise is not reflected, the reliability of the financial related report provided by the enterprise is questioned. If the reason for the need for enterprise analysis through non-financial data is examined, as SNS activities of people through the Internet are increasingly active, the reputation of an enterprise may be shaken by specific news-disseminated events or the like to the enterprise or products, and in addition, risks to the corresponding enterprise may be increased due to crimes or health-related news of the owner of the enterprise, which are actually difficult to analyze through financial data. Therefore, the enterprise is analyzed not only by the financial data but also by the non-financial data, and thereby the enterprise evaluation can be performed more precisely.

Under the trend, a methodology for dividing non-financial data into three topics of ESG (environmental, Social and Governance) for analysis is silently developed.

Although there are companies that make evaluation reports for companies based on non-financial data such as an ESG, the non-financial data is subjective and the report making speed is relatively slow, and thus there is a disadvantage that it is difficult for companies to utilize the data. In order to improve such a drawback, in terms of collecting ESG-related news reports on the internet by a computer program or the like and automatically analyzing and performing enterprise evaluation, the news reports have many difficulties in automating the classification and evaluation of the news reports due to non-normative data.

Disclosure of Invention

Various embodiments of the present invention have been developed to solve the above-described problems, and an object thereof is to collect news reports on the internet, and based thereon, perform evaluation of businesses based on an ESG.

It is another object of the present invention to categorize news stories for the same event by similarity analysis between the collected news stories.

It is still another object of the present invention to enable an apparatus for performing ESG enterprise evaluation to perform learning using a machine learning algorithm, thereby improving the performance of classifying collected news.

The problems to be solved by the present invention are not limited to the above-mentioned problems, and other problems not mentioned can be understood by those skilled in the art from the following descriptions.

(means for solving the problems)

In order to achieve the above object, an embodiment of the present invention provides an ESG enterprise evaluation apparatus for evaluating an enterprise from an ESG (environmental, Social, government) perspective, and calculating a score, the apparatus including: a news collecting unit that collects a plurality of news reports on the internet, classifies the news reports by date or by enterprise, and performs classification and consolidation of news reports having a similarity of a reference value or more by similarity analysis among the news reports; a news classification section that classifies the individual news stories as being related to which topic in an environment, society, or governance structure; and an evaluation result deriving unit that calculates an ESG risk for each cluster, and calculates an ESG enterprise evaluation score based on the calculated value.

The news gathering part may perform a morpheme analysis using a morpheme analyzer corresponding to the news story production language, and perform vectorization of the respective news stories based on a morpheme analysis result, and the similarity analysis may be performed by cosine similarity analysis of the respective news stories.

The news gathering unit may use a TF-IDF (Term Frequency-Inverse Document Frequency) value in performing vectorization of each news story by a morpheme analysis result.

The news taxonomy may first categorize whether the news story is related to at least one of an environment, a society, or an governance structure in a true-false manner before categorizing the respective news story as related to which topic in the environment, society, or governance structure.

The news classifying unit may classify each of the news reports into a refined list and classify each of the news reports into the list after classifying each of the news reports as being related to a topic in an environment, a society, or a management structure.

The news classification section may employ a specific machine learning algorithm to perform learning by training data before classifying each news story as being related to which topic in an environment, a society, or a management structure, thereby improving classification ability.

The machine learning algorithm adopted by the news classification section may be any one of polynomial Bayes (multinomialebayes), Bernoulli Bayes (Bernoulli Bayes), SGDs (Stochastic Gradient descnd: random Gradient descent), Linear support vector classifiers (Linear SVC), perceptrons (Perceptron), or random forests (randomfort).

The evaluation result deriving unit may classify the nouns extracted from the news article, and may perform the setting of the order or the calculation of the evidence level for each directory item in units of clusters based on the frequency of words included in the directory item.

The evaluation result derivation unit may calculate a probability that each cluster belongs to a topic of the environment, society, or treatment structure based on the calculated evidence level value.

The evaluation result deriving unit may use the calculated evidence level value and the probability of each cluster belonging to an environment, a society, or an administration structure in calculating the ESG enterprise evaluation score.

In order to achieve the above object, according to another embodiment of the present invention, there is provided a score calculating method of an ESG enterprise evaluating apparatus, which is a method of evaluating an enterprise and calculating a score from an ESG (environmental, Social, government) perspective as the ESG enterprise evaluating apparatus, including: collecting a plurality of news reports on the internet, classifying the news reports according to dates or enterprises, and performing classification and merging on the news reports with the similarity higher than a reference value through the similarity analysis among the news reports; a step of classifying each news story as being related to which topic in an environment, society, or governance structure; and calculating the ESG risk of the corresponding cluster according to the cluster unit, and calculating the ESG enterprise evaluation score based on the ESG risk.

Effects of the invention

According to one embodiment of the invention, ESG enterprise evaluation is automatically executed, so that the speed of ESG enterprise evaluation export can be improved.

According to another embodiment of the present invention, an enterprise rating apparatus that continuously improves the performance of classifying news stories by means of machine learning can be provided.

According to yet another embodiment of the present invention, analysis may be performed on news stories produced in multiple languages and the business may be evaluated based thereon.

The effects of the present invention are not limited to the above-mentioned effects, and other effects not mentioned are clearly understood by those of ordinary skill from the following descriptions.

Drawings

Fig. 1 is a conceptual diagram schematically showing a flow of performing evaluation of an ESG enterprise according to an embodiment of the present invention.

FIG. 2 is a block diagram schematically illustrating the construction of an EGS enterprise valuation module in accordance with one embodiment of the present invention.

Fig. 3 is a diagram for explaining a method of classifying nouns extracted from news reports by the evidence level calculation unit.

Fig. 4 is a table for explaining a method in which the proof level calculating unit normalizes the items by setting the order of the items based on the number of words included in the list items, and calculates the proof level based on the normalized numerical value.

Fig. 5 is a diagram showing the result of the ESG probability calculation unit performing the ESG probability calculation for each cluster according to one embodiment of the present invention.

Fig. 6 is a block diagram schematically illustrating a process in which an ESG enterprise rating means derives an ESG enterprise rating score from the process of collecting news reports in accordance with one embodiment of the present invention.

Detailed Description

The terminology used in the description is for the purpose of describing the embodiments and is not intended to be limiting of the invention. In this specification, the singular forms also include the plural forms as long as they are not specifically mentioned in the sentence. The use of "comprising" and/or "comprising" in the specification does not preclude the presence or addition of one or more other components in addition to the recited components. Throughout the specification, the same reference numerals refer to the same constituent elements, and "and/or" includes each or all combinations of one or more of the constituent elements mentioned. Although the terms "first", "second", and the like are used to describe various components, it is needless to say that these components are not limited by these terms. These terms are only used to distinguish one constituent element from other constituent elements. Therefore, the first component described below is within the technical idea of the present invention, and may be a second component.

In the following description, when a part "includes" a certain component, unless otherwise specified, the other component is not excluded, and it means that the other component may be included. In addition, terms such as "… section" and "module" described in the specification mean a unit that processes at least one kind of function or action, and may be implemented by hardware or software, or a combination of hardware and software.

Fig. 1 is a conceptual diagram schematically showing a flow of performing evaluation of an ESG enterprise according to an embodiment of the present invention.

The evaluation of the ESG enterprise disclosed in the present invention can be performed by means of automation of a program embodied in the form of computer software. That is, each method exemplarily illustrated in fig. 1 may be executed by performing an arithmetic process by software loaded in an ESG enterprise evaluation device 100 that performs an ESG enterprise evaluation.

Referring to fig. 1, the ESG enterprise evaluation device 100 may calculate a final enterprise evaluation result through three steps. If fig. 1 (a) is considered, the ESG enterprise rating apparatus 100 may first collect news reports, which are basic materials for performing enterprise rating, from the internet. The ESG enterprise evaluating apparatus 100 determines whether news is related to which enterprise, which topic, and the like, by morpheme analysis, similarity calculation between documents, and the like, in collecting news reports, and can perform classification merging of classification between similar reports for the first time.

If referring to (b) of fig. 1, the ESG enterprise evaluation apparatus 100 may perform a more precise news story classification operation on the basis of the news stories collected and classified and merged for the first time. The ESG enterprise evaluation apparatus 100 first determines whether the collected news is related to the ESG, that is, whether the collected news is related to at least one of the environment, the society, or the administration structure, and then determines which subject of the environment, the society, or the administration structure the collected news is related to and classified in performing the news report classification operation. Finally, the ESG enterprise evaluation apparatus 100 can perform more detailed directory classification for each news report classified into three subjects of environment, society, and governance structure.

Referring to fig. 1 (c), the ESG enterprise rating apparatus 100 derives a final enterprise rating score based on the categorized news stories. In this process, the ESG enterprise evaluation apparatus 100 may calculate an evidence level for a main word included in a news report and a probability of which the news report belongs to an environment, a society, or an administration structure, for each cluster unit.

For convenience of explanation, the ESG enterprise rating device 100 of the present invention is illustrated and described as a case where a final ESG enterprise rating score is derived through three steps, and such steps may be divided or integrated, and may be embodied in a smaller or greater number of steps.

Fig. 2 is a block diagram schematically showing the configuration of the EGS enterprise valuation module 100 according to an embodiment of the present invention.

Referring to fig. 2, the ESG enterprise evaluation apparatus 100 may include a control unit 110, a news collection unit 120, a news classification unit 130, an evaluation result derivation unit 140, a communication unit 150, and a storage unit 160. The news gathering unit 120 may include a morpheme analyzing unit 121, a business and date classifying unit 122, and a news classification merging unit 123, the news classifying unit 130 may include a binary classifying unit 131, an ESG classifying unit 132, and a directory classifying unit 133, and the evaluation result deriving unit 140 may include an evidence rank calculating unit 141, an ESG probability calculating unit 142, and a score calculating unit 143.

For convenience of description, the main body performing each function in the ESG enterprise evaluation device 100 is illustrated as a unit, but each unit may be a subroutine module operating in the ESG enterprise evaluation device 100. Such program modules are, without limitation, concepts of routines, subroutines, programs, objects, components, data structures, etc., that perform various actions or operate on specific abstract data types.

The control unit 110 according to one embodiment may perform a function of controlling data flow among the news collection unit 120, the morpheme analysis unit 121, the business and date classification unit 122, the news classification merging unit 123, the news classification unit 133, the binary classification unit 131, the ESG classification unit 132, the catalog classification unit 133, the evaluation result derivation unit 140, the evidence rank calculation unit 141, the ESG probability calculation unit 142, the score calculation unit 143, the communication unit 150, and the storage unit 160. That is, the control unit 110 of the present invention may control the news gathering unit 120, the morpheme analyzing unit 121, the business and date classifying unit 122, the news classification merging unit 123, the news classifying unit 133, the binary classifying unit 131, the ESG classifying unit 132, the catalog classifying unit 133, the evaluation result deriving unit 140, the evidence rank calculating unit 141, the ESG probability calculating unit 142, the score calculating unit 143, the communication unit 150, and the storage unit 160 to execute their own functions.

News gathering section 120 for one embodiment, as previously described, may include a morpheme analyzing section 121, a business and date classifying section 122, and a news classification merging section 123. The ESG enterprise evaluation apparatus 100 is required to perform enterprise evaluation by collecting only news reports related to the environment, society, or management structure of an enterprise among a plurality of news reports occurring on the internet, and generating no erroneous evaluation due to false news or the like only when obtaining news from a trusted place. The news gathering unit 120 may set a periodic time interval to gather news reports updated on the internet, and may gather date information of each news report distribution, public opinion media information of the corresponding news distribution, and the like together in gathering the news reports. The news report collection of the news gathering part 120 may be implemented by the communication part 150 in the ESG enterprise evaluation device 100.

The morpheme analyzing unit 121 may analyze the entire text of the collected news report in morpheme units, which are the smallest units of significance. When the morpheme analyzing section 121 of one embodiment analyzes a news report produced in korean, one of a variety of korean morpheme analyzers can be selected through which analysis is performed. According to one embodiment, as the korean morpheme analyzer, there are various kinds such as kiwi, HAM, HLX, Mecab, etc., and the morpheme analyzing part 121 may perform morpheme analysis using one of them and use the result thereof. According to another embodiment, the morpheme analyzing part 121 may perform morpheme analysis using a morpheme analyzer corresponding to a corresponding language in analyzing news made in other languages than korean.

The business and date sorting section 122 may sort the collected news stories by business and by date. The business and date classification unit 122 may determine which business the specific report is about by using a business dictionary constructed in a database form by business names. The business and date classification unit 122 determines which business the specific report is about, and then can check the date of the release of the corresponding report and sort the news reports by time. According to one embodiment, the business and date categorizing section 122 may categorize news related to a business by a period or a month period after categorizing news stories by business. Such a categorization by period may be used for the ESG enterprise rating means 100 to calculate in a manner that the latest news reports have a weighted value when later performing enterprise rating.

The news category merge section 123 may group the related news related to the same topic in the collected news reports into one cluster. The news classification merge unit 123 may calculate the similarity between news reports in order to determine whether or not a plurality of news reports are related news related to the same topic.

According to an embodiment, the news classification merge part 123 may calculate the similarity between news reports by vectorizing each document and then calculating the cosine similarity between vectors.

The news category merging unit 123 can vectorize each news report using TF-IDF (Term Frequency-Inverse document Frequency). Tf (term frequency) is a value showing the number of frequencies at which a specific word is present in the document, and the higher the value, the more important the corresponding word can be considered in the document. The Frequency with which a particular word is placed in a collection of news stories can be expressed as DF (Document Frequency), the Inverse of which is IDF (Inverse Document Frequency). If a particular word frequently hits between multiple news stories, the word becomes a long-standing word and cannot become a core word in the news story, thus using IDF instead of DF. TF-IDF is defined as a value by which TF and IDF are multiplied, and the way in which the news classification merge part 123 of an embodiment of the present invention calculates TF-IDF of a news report may use mathematical formula 1.

[ mathematical formula 1]

In the above mathematical formula 1, tf_i.jRepresenting the frequency of the word i's presence in the news story j, df_iRepresenting the number of news stories that contain the word i in the set of news stories.

The news classification merge unit 123 may vectorize each news report based on the above equation 1, and may calculate the similarity between news reports based on the vector value of each news report.

[ mathematical formula 2]

According to one embodiment, the news classification merge section 123 may calculate the similarity between news stories by the mathematical expression 2. In the formula 2, A and B are vectors, A_iAnd B_iRefer to the ith component in the A and B vectors, respectively. The cosine similarity between news stories is calculated as a number between 0 and 1, 0 being the case where the news stories are independent of each other, and 1 being the case where the news stories are identical to each other.

The news classification merge unit 123 may determine whether the news reports are similar based on a reference value set in advance through an experiment after calculating the similarity between the news reports, and may bundle the news reports determined to be similar into a cluster. In the following description, the "cluster" refers to a set of news reports that are judged to be similar to news by the news classification and merge unit 123.

News classification component 130 for one embodiment, as previously described, may include a binary classification component 131, an ESG classification component 132, and a directory classification component 133. The news categorizing section 130 may categorize the news stories collected by the news gathering section 120 according to three criteria. The news classification section 130 may classify news stories for the first time by the binary classification section 131, then classify by the ESG classification section 132, and finally classify by the directory classification section 133. In this process, news stories may be sorted on a more detailed basis.

The binary classifier 131 may determine whether each collected news story is related to an ESG and classify the news story in a True false (True error) manner. That is, the binary classifier 131 may classify the collected news into TRUE (TRUE) if it is determined that the collected news is related to at least one of environment, society, and government structure in determining whether the collected news is data that can be used for evaluation of the ESG enterprise, and may classify the collected news into FALSE (FALSE) otherwise. Therefore, the binary classification unit 131 classifies news reports that are not needed for the ESG enterprise evaluation, such as sports, performances, politics, arts, etc., into FALSE, and thus the amount of news reports that the ESG enterprise evaluation apparatus 100 needs to process can be greatly reduced.

According to one embodiment, the ESG classification section 132 may classify the binary classification section 131 into a TRUE (TRUE) news story, into three topics of environmental, social, and governance structures. The categorized news stories may then be categorized into a more refined category by means of the category categorization section 133. That is, the topics of the environment, society, and management structure are classified into more detailed subordinate catalogs, and one news report belongs to one of the subordinate catalogs by the catalog classification unit 130.

According to one embodiment, the binary classification section 131, the ESG classification section 132 and the directory classification section 133 may perform classification of news stories using a machine learning algorithm.

If the binary classification section 131, the ESG classification section 132 and the directory classification section 133 are examined in a way of performing classification using a machine learning algorithm, first, only nouns can be extracted from news reports through natural language analysis. According to one embodiment, for the binary classification section 131, the ESG classification section 132 and the directory classification section 133 to extract only nouns in a news story composed of korean, it is possible to use KoNLP, which is one of the categories of natural language analysis packets. According to another embodiment, the binary classification unit 131, the ESG classification unit 132 and the directory classification unit 133 may use an analysis tool corresponding to a language in order to extract nouns in news reports composed of different languages.

The binary classification unit 131, the ESG classification unit 132, and the directory classification unit 133 can generate a matrix using the TF-IDF by using the above equation 1 after extracting only nouns from news reports. The matrix generated by using the TF-IDF can be generated in accordance with the clusters classified by the news classification merge section 123. According to one embodiment, in the generated matrix, words (nouns) contained in respective news stories may be listed in a row, with news stories containing respective words (nouns) listed in a column for respective words.

The binary classifier 131, the ESG classifier 132 and the directory classifier 133 may then perform classification in accordance with the role of each classifier by a machine learning algorithm based on the matrix generated by TF-IDF. According to one embodiment, the binary classification unit 131, the ESG classification unit 132 and the directory classification unit 133 adopt a specific machine learning algorithm, and then learn by receiving input of training data previously classified by a user, thereby improving the classification capability. Then, test data may be input to the binary classification unit 131, the ESG classification unit 132, and the catalog classification unit 133, which have been learned to a predetermined degree by the training data, and each classification unit that has passed the test may perform classification of news reports by a corresponding learning algorithm.

According to one embodiment, the binary classification unit 131, the ESG classification unit 132 and the directory classification unit 133 may select and use one of machine learning algorithms such as polynomial Bayes (multinominal Bayes), Bernoulli Bayes (Bernoulli Bayes), SGD (Stochastic Gradient descnd), Linear support vector classifiers (Linear SVC), perceptrons (Perceptron), Random Forest (Random Forest).

The evaluation result deriving part 140 of one embodiment may perform ESG enterprise evaluation on the basis of news reports collected by means of the news collecting part 120 and classified by categories by means of the news classifying part 130. That is, the evaluation result deriving unit 140 may perform a business evaluation by a score by performing a score for each business based on the ESG-related report. As described above, the evaluation result deriving unit 140 may include the proof level calculating unit 141, the ESG probability calculating unit 142, and the score calculating unit 143.

The evidence level calculation unit 141 can classify nouns extracted from the news report into a plurality of word lists. According to one embodiment, the evidence level calculation section 141 may classify nouns into environmental Damage (D _ Env), social stakeholders (S _ Company), General Damage (D _ gen), and the like.

Fig. 3 is a diagram for explaining a method of classifying nouns extracted from news reports by the evidence level calculation unit 141.

Referring to fig. 3, the evidence level calculation section 141 may determine which item in the ESG the extracted noun corresponds to, which elements are related to in the corresponding item, and finally perform classification. The classification of such nouns may be different from the classification performed by the aforementioned catalog classification section 133. That is, the classification performed by the catalog classification unit 133 is a lower catalog that classifies each news story into a topic finer than the topic of the environment, society, and governance structure, and the classification performed by the evidence level calculation unit 141 may be a classification of a noun extracted from the news story. That is, the catalog classification unit 133 classifies news reports, the evidence level calculation unit 141 classifies nouns, and the individuals belonging to the catalog are classified into news reports and nouns, respectively.

According to one embodiment, after the evidence level calculation section 141 performs the classification as described above, it is possible to calculate that each directory item contains several words in accordance with the cluster as a news report set classified by the news classification merging section 123. Such a word may be a noun that is extracted and classified by the evidence level calculation unit 141, as shown in fig. 3.

The evidence level calculation unit 141 may set the order of each item in the list based on the number of words included in the item in the list, normalize the number of words, and calculate the evidence level of each item in the list based on the normalized numerical value.

Fig. 4 is a table for explaining a method in which the proof level calculating unit 141 normalizes the items by setting the order of the items based on the number of words included in the list items, and calculates the proof level based on the normalized numerical value.

Referring to fig. 4, it is possible to identify, for a cluster, which words are included and which words are included by each directory, and the order of each directory is set based on the number of included words. The number of words included in each list is displayed as a normalized numerical value, and the evidence rank is displayed based on the normalized numerical value. The evidence level calculation unit 141 may calculate the evidence level by classifying the normalized numerical value into sections based on a predetermined reference value.

The ESG probability calculation unit 142 may calculate the probability that each cluster, which is a similar news group, belongs to a topic in the environment, society, or treatment structure based on the directory item used by the evidence level calculation unit 141.

The ESG probability calculation unit 142 may normalize the directory entries used by the evidence level calculation unit 141 into a smaller number of sets before calculating the probability that each cluster belongs to a certain topic in the ESG. According to one embodiment, these collections may be categorized into environmental, social, abatement, enterprise risk, and other related issues.

[ mathematical formula 3]

The ESG probability calculation unit 142 may calculate the evidence level of the environment, society, or administrative structural topic of the specific cluster by using the above equation 3. Taking a method of calculating the evidence rating of the environment by the ESG probability calculating unit 142 as an example, in equation 3, C_iRefers to the ith directory in the context-related directory, n refers to the number of context-related directories, E (C)_i-) The evidence level of the i-th directory regarding the environment is calculated by the evidence level calculation unit 141 as described above. That is, the evidence level calculated by the evidence level calculation unit 141 is an evidence level for a catalog used when classifying extracted nouns, and the evidence level calculated by the ESG probability calculation unit 142 is an evidence level for three topics including environment, society, and governance structure.

The ESG probability calculation unit 142 can calculate the probability of each cluster being related to any one of the three categories of environment, society, and treatment structure.

[ mathematical formula 4]

The ESG probability calculation unit 142 may calculate the ESG probability of each cluster using the above equation 4.

Fig. 5 is a diagram showing the result of the ESG probability calculation unit 142 according to one embodiment of the present invention performing the ESG probability calculation for each cluster.

Referring to fig. 5, the evidence level and its ESG probability about three topics of environment, society and governance structure are shown for a plurality of clusters. It can be confirmed that the probabilities of a particular cluster being related to which of the environmental, social, and treatment structure topics are all added to 1.

The score calculating part 143 according to an embodiment may calculate the ESG enterprise evaluation score using the calculated auxiliary index after calculating various auxiliary indexes before calculating the ESG enterprise evaluation score as a final step of the ESG enterprise evaluation. The auxiliary index and the ESG enterprise evaluation score may be calculated by the evidence level and the ESG probability calculated as described above.

According to one embodiment, various auxiliary indicators may be comprised of an ESG risk score, an enterprise risk score, an association score, and the like. These auxiliary indexes are not limited to the three types described above, and may be defined as various numbers and calculation methods.

[ math figure 5]

According to an embodiment, the score calculating part 143 may finally calculate the ESG enterprise evaluation score in the manner as shown in the above equation 5. In equation 5, ESGrisk, company risk, Relevance represent various auxiliary indices. In equation 5, ESGrisk, company risk, Relevance represent various auxiliary indices. ESGrisk may be a value of a total after calculating how much risk a set of news reports classified into a specific cluster has with respect to environment, society, and business structure, respectively, company risk may be a value related to a business risk related set in the set specified by the ESG probability calculation section 142 and a word list classified into the corresponding set, and Relevance may be a value related to a word list classified into other sets and a classification corresponding set in the set specified by the ESG probability calculation section 142.

The communication section 150 of one embodiment enables communication between the ESG enterprise evaluation device 100 and an external device. Specifically, the ESG enterprise evaluation apparatus 100 is allowed to communicate with the user terminal of the corresponding apparatus, so that news distributed by the news gathering part 120 can be collected through internet connection.

The storage 160 of one embodiment may store data required for the operation of the ESG enterprise evaluation device 100. The storage unit 160 may store the collected news reports, classification information for the news reports, assignment information, learning history of the machine learning algorithm, and the like in the form of data.

In the manner as described above, the ESG enterprise rating apparatus 100 may finally derive an ESG enterprise rating score for each cluster as a similar report set.

Fig. 6 is a block diagram schematically illustrating a process of deriving an ESG enterprise rating score from a process of collecting news reports by the ESG enterprise rating device 100 according to one embodiment of the present invention.

If referring to fig. 6, news stories are data that are fundamental in performing business evaluation, the ESG enterprise evaluation device 100 may collect news stories on the internet at regular or irregular intervals S601.

Then, the ESG enterprise evaluation device 100 may classify the collected news stories by enterprise and by date S603, and analyze the corresponding stories using an appropriate morpheme analyzer according to the language in which the collected news stories are produced S605.

The ESG enterprise evaluation apparatus 100 calculates the similarity between news reports through the similarity analysis between vectors after vectorizing the news reports based on the news reports on which the morpheme analysis is completed, collects related news together, and performs classification and merge S607. In this process, the ESG enterprise evaluation apparatus 100 may perform vectorization on each news report using TF-IDF, and may calculate the similarity between each news report by cosine similarity calculation.

The ESG enterprise evaluation apparatus 100 may classify S609 as to which topic in the environment, society, and governance structure each news report collected belongs to, and which in the inventory of topics in the environment, society, and governance structure belongs to. In this process, the ESG enterprise evaluation apparatus 100 may classify each news report into whether or not the report is related to the execution of the ESG enterprise evaluation in a TRUE or FALSE (TRUE or FALSE) form, and in each classification step, may adopt an appropriate machine learning algorithm to generate data for exercise so that the ESG enterprise evaluation apparatus 100 executes learning by the corresponding algorithm. The user of the EGG enterprise evaluation apparatus 100 can perform each classification of step S609 by verifying the ESG enterprise evaluation apparatus 100, which performs learning using the machine learning algorithm and the training data, using the test data.

The ESG enterprise evaluating apparatus 100 may calculate an evidence level value of each ESG topic of the corresponding cluster in each cluster unit classified in step S607, and calculate an ESG probability value of the corresponding cluster S611. In this process, the ESG enterprise rating means 100 may classify nouns extracted from news reports into a plurality of word categories, such as a set of environmental, social, governance structures, enterprise risks, other related issues. Then, the ESG enterprise evaluation apparatus 100 may calculate an evidence level value about the environment, the society, and the treatment structure in the set, and may calculate an ESG probability value.

Finally, the ESG enterprise evaluation device 100 may calculate a final ESG enterprise evaluation score S613 based on the evidence level value and the ESG probability value calculated in step S611. In this process, the ESG enterprise evaluation device 100 may use the ESG probability value calculated in step S611 and the evidence level value classified as the set of enterprise risks.

According to an embodiment of the present invention, the ESG enterprise rating apparatus 100 may include a function of providing a basis for deriving a corresponding score when an individual or a business using an ESG enterprise rating result requires a basis for deriving a rating score for a specific business. That is, when calculating the evaluation score for a specific business report, information that the corresponding evaluation score is greatly affected depending on the frequency number of specific words present in the report may be provided, or a report including the corresponding word may be searched for and provided.

As described above, the ESG enterprise evaluation apparatus 100 according to the embodiment of the present invention performs automated enterprise evaluation, so that when similar news reports about a specific enterprise are released on the internet, it is possible to determine which topic of the environment, society, and governance structure the corresponding news report relates to, and to what degree the risk of the corresponding topic is significant.

On the other hand, the ESG enterprise rating apparatus 100 according to an embodiment of the present invention may also be embodied in a computer readable recording medium as computer readable code. The computer-readable recording medium includes all kinds of recording devices that store data that can be read by means of a computer system.

For example, as the computer-readable recording medium, there are Read Only Memory (ROM), Random Access Memory (RAM), compact disc read only drive (CD-ROM), magnetic tape, hard disk, floppy disk, removable storage device, nonvolatile memory (flash memory), optical data storage device, and the like.

In addition, the computer-readable recording medium can be distributed to computer systems connected via a computer communication network, and stored and operated as codes that can be read in a distributed manner.

While the embodiments of the present invention have been described with reference to the drawings, it will be understood by those skilled in the art that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The embodiments described above are therefore to be considered in all respects as illustrative and not restrictive.