KR102625553B1 - The Method And The Computer-Readable Recording Medium To Extract Regulatory Legal Provisions Associated With Inputted Query, And The Computing System Performing That Same - Google Patents

Abstract

The present invention provides a method for deriving regulatory legal provisions related to an inputted query, a computer-readable recording medium, and a computing system for performing the same. When a regulatory-related law or regulation-related search word is input from a user, legal provisions similar to the inputted information are provided. is derived based on the similarity score derived using one or more artificial neural network-based inference models, and among the similar legal provisions derived, legal provisions whose regulatory score derived using the learned regulatory classification model are higher than the preset standard are derived, It relates to a method of deriving regulatory legal provisions related to an inputted query, a computer-readable recording medium, and a computing system for performing the same, which provides the user with a list of regulatory legal provisions related to the regulations requested by the user.

Description

The Method And The Computer-Readable Recording Medium To Extract Regulatory Legal Provisions Associated With Inputted Query, And The Computing System Performing That Same}

The present invention provides a method for deriving regulatory legal provisions related to an inputted query, a computer-readable recording medium, and a computing system for performing the same. When a regulatory-related law or regulation-related search word is input from a user, legal provisions similar to the inputted information are provided. is derived based on the similarity score derived using one or more artificial neural network-based inference models, and among the similar legal provisions derived, legal provisions whose regulatory score derived using the learned regulatory classification model are higher than the preset standard are derived, It relates to a method of deriving regulatory legal provisions related to an inputted query, a computer-readable recording medium, and a computing system for performing the same, which provides the user with a list of regulatory legal provisions related to the regulations requested by the user.

The government introduced the regulatory registration system with the enforcement of the Framework Act on Administrative Regulations in 1998 in order to provide the contents of regulations in an easy-to-understand manner to policy users such as citizens and businesses and at the same time use it as a means of quality management of regulatory reform. More specifically, according to the <Regulatory Reform White Paper> (1998; 2015; 2016.) published by the Regulatory Reform Committee, when the head of a central administrative agency establishes or changes a relevant regulation, he or she shall do so within 30 days from the date of promulgation of the regulatory law, etc. The name, content, and duration of the regulation must be registered with the Regulatory Reform Committee, and the Regulatory Reform Committee must inform the public of the registered regulations in the form of a regulation registration card through the Internet website and submit them to the National Assembly by the end of June every year. . Since then, the regulatory registration system, which was completely revamped in 2015, registers all regulations in trillions and discloses registered regulations in real time to the regulatory information portal and the National Law Information Center, providing information on regulations to the public.

Meanwhile, the legal provisions provided in the aforementioned regulatory information portal or national legal information center are structured to open lower-level laws within higher-level law folders in a classification system listed by legal field and relevant ministries, and are simply listed without relation to the content of the legal field. Therefore, when a business or government tries to find regulatory information on a new industrial field and does not have sufficient relevant knowledge or information, it is not only difficult to select the specific field needed for the user among the numerous fields already classified, but especially for small and medium-sized businesses. When a start-up company provides convergence services and multiple fields and relevant ministries overlap, there is a problem that it is difficult to simultaneously check scattered information on the platform.

In addition, in the case of the 'life laws and regulations by topic' classification provided by the above-mentioned platform, a classification system is being established based on categories consisting of topics related to real life, so small and medium-sized businesses and startup companies can use the classification system provided by the platform based on industrial area. When trying to use it, there are also problems that make it difficult to obtain the desired search results.

In relation to the above-mentioned problem, there is a conventional technology for users to more easily search for legal provisions, such as Republic of Korea Patent Publication No. 10-2022-0122429. However, as described above, regulatory legal provisions required by government ministries or businesses can be effectively searched. There is absolutely no indication as to what can be found. Therefore, there is a need for technology that can effectively find regulatory provisions that users want.

Republic of Korea Patent Publication No. 10-2022-0122429 (2022.09.02.)

The present invention provides a method for deriving regulatory legal provisions related to an inputted query, a computer-readable recording medium, and a computing system for performing the same. When a regulatory-related law or regulation-related search word is input from a user, legal provisions similar to the inputted information are provided. is derived based on the similarity score derived using one or more artificial neural network-based inference models, and among the similar legal provisions derived, legal provisions whose regulatory score derived using the learned regulatory classification model are higher than the preset standard are derived, The purpose is to provide a method for providing users with a list of regulatory legal provisions related to regulations requested by the user, a method for deriving regulatory legal provisions related to an input query, a computer-readable recording medium, and a computing system for performing the same.

In order to solve the above problems, an embodiment of the present invention is a method of deriving regulatory legal provisions related to an input query, which is performed on a computing system including one or more processors and one or more memories, and includes a plurality of laws. A regulatory score derivation step of deriving a regulatory score indicating how related each of a plurality of legal provisions stored in the database is to the regulation using a regulatory classification model learned with the regulatory card information of the legal provisions with regulatory card information among the provisions; A similarity score derivation step of deriving a similarity score according to the similarity between a query input from a user and each of the plurality of legal provisions using one or more artificial neural network-based inference models; and a related regulatory law providing step of classifying and sorting regulatory legal provisions related to the query according to preset standards based on the regulatory score and the similarity score and providing the relevant regulatory legal provisions to the user. provides.

In one embodiment of the present invention, the similarity score derivation step involves inputting a query input from a user and a plurality of legal provisions stored in the database into a first inference model based on an artificial neural network to derive a keyword matching score. Score derivation step; and a cosine similarity score derivation step of deriving a cosine similarity score between each legal provision stored in the database and a query input from the user.

In one embodiment of the present invention, the keyword matching score deriving step includes extracting keywords from a query input from the user; Converting each of the extracted plurality of keywords into a vector having a dimension of 1 or more, and then deriving the converted plurality of vectors into one keyword vector; and entering a legal provision and a keyword vector into a first inference model based on an artificial neural network, thereby deriving a keyword matching score indicating the similarity between the input legal provision and the keyword vector.

In one embodiment of the present invention, the cosine similarity score deriving step includes deriving a query vector for a query input from the user through an artificial neural network-based word embedding model; and deriving a cosine similarity score based on the cosine similarity derived between each of all legal provisions stored in the database and the query vector.

In one embodiment of the present invention, the similarity score derivation step is performed by inputting the legal provisions and the input query derived based on the keyword matching score and the cosine similarity score into a second inference model based on an artificial neural network to obtain the derived query. It further includes a relevance score derivation step of deriving a relevance score indicating the correlation between a legal provision and the input query, and each of the plurality of legal provisions based on the keyword matching score, the cosine similarity score, and the relevance score. A similarity score can be derived.

In one embodiment of the present invention, the relevant regulatory law provision step classifies regulatory legal provisions related to the query based on the similarity score, determines legal provisions higher than the preset regulatory score as regulatory legal provisions, and then determines the regulatory legal provisions. Legal provisions can be provided to users in the form of a list arranged in order according to preset conditions.

In one embodiment of the present invention, the step of providing relevant regulatory laws classifies regulatory legal provisions related to the query based on the similarity score, and classifies regulatory legal provisions related to the query based on the regulatory scores of legal provisions having a preset number of higher similarity scores. After determining the legal provisions above the calculated standard regulatory score as regulatory legal provisions, the determined regulatory legal provisions can be provided to the user in the form of a list arranged in order according to preset conditions.

In order to solve the above problems, an embodiment of the present invention is a computing system that includes one or more processors and one or more memories and performs a method of deriving regulatory legal provisions related to an input query, and includes a plurality of laws. A regulatory score derivation unit that derives a regulatory score indicating how related each of a plurality of legal provisions stored in the database is to the regulation using a regulatory classification model learned with the regulatory card information of the legal provisions with regulatory card information among the provisions; a similarity score deriving unit that uses one or more artificial neural network-based inference models to derive a similarity score according to the similarity between a query input from a user and each of the plurality of legal provisions; and a related regulation law provision unit that classifies and sorts regulatory law provisions related to the query according to preset standards based on the regulation score and the similarity score and provides the information to the user.

In order to solve the above problems, an embodiment of the present invention is a computer for implementing a method of deriving regulatory provisions related to an input query, which is performed on a computing system including one or more processors and one or more memories. - A readable recording medium, wherein the computer-readable recording medium stores instructions for causing the computing system to perform the following steps, the following steps being: a law containing regulatory card information among a plurality of legal provisions; A regulatory score derivation step of deriving a regulatory score indicating how related each of a plurality of legal provisions stored in the database is to the regulation using a regulatory classification model learned with the regulation card information of the provision; A similarity score derivation step of deriving a similarity score according to the similarity between a query input from a user and each of the plurality of legal provisions using one or more artificial neural network-based inference models; and a related regulatory law providing step of classifying and sorting regulatory legal provisions related to the query according to preset standards based on the regulatory score and the similarity score and providing the relevant regulatory law provisions to the user. to provide.

According to one embodiment of the present invention, laws that are not classified as regulations are derived by deriving regulatory legal provisions based on regulatory scores derived through a learned regulatory classification model without providing information on regulatory legal provisions designated by the government as is. This can have the effect of making it easier to find shadow regulation provisions that are recognized as regulations.

According to one embodiment of the present invention, when a company, etc., is reviewing a new business or preparing to start a business, it is possible to easily find regulatory provisions related to the business to be conducted.

According to one embodiment of the present invention, government ministries, legislative bodies, and national research institutes reviewing regulations can easily find legal provisions related to a specific field, which has the effect of easily monitoring legal provisions that require legislation or revision. can be demonstrated.

According to one embodiment of the present invention, users can enter search terms in various formats, allowing users to easily search for regulatory provisions in the field they are looking for even if they do not have sufficient legal knowledge such as legal terms.

According to one embodiment of the present invention, by providing a list of regulatory laws classified by topic based on the contents of actual legal provisions, rather than a list classified by field of law or relevant ministry, legal provisions substantially related to the search word or search statement entered by the user It can be effective in providing .

Figure 1 schematically shows the execution steps of a method for deriving regulatory legal provisions according to an embodiment of the present invention and the configuration of a computing system that performs the same.
Figure 2 schematically shows the process of deriving a list of related regulatory laws in a computing system according to an embodiment of the present invention.
Figure 3 schematically shows the process of learning a regulatory classification model according to an embodiment of the present invention.
Figure 4 schematically shows the process of deriving a regulatory score for each legal provision through a learned regulatory classification model according to an embodiment of the present invention.
Figure 5 schematically shows the execution steps of the similarity score derivation step according to an embodiment of the present invention.
Figures 6 and 7 schematically show the execution process of the keyword matching score derivation step according to several embodiments of the present invention.
Figure 8 schematically shows the process of performing the cosine similarity score derivation step according to an embodiment of the present invention.
Figure 9 schematically shows the process of deriving candidate legal provisions according to an embodiment of the present invention.
Figure 10 schematically shows the process of deriving a similarity score according to an embodiment of the present invention.
Figure 11 schematically shows the process of determining regulatory legal provisions according to an embodiment of the present invention.
Figure 12 schematically shows a list of regulatory laws provided to users according to an embodiment of the present invention.
Figure 13 exemplarily shows the internal configuration of a computing device according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS Various embodiments and/or aspects are now disclosed with reference to the drawings. In the following description, for purposes of explanation, numerous specific details are set forth to facilitate a general understanding of one or more aspects. However, it will also be appreciated by those skilled in the art that this aspect(s) may be practiced without these specific details. The following description and accompanying drawings set forth in detail certain example aspects of one or more aspects. However, these aspects are illustrative and some of the various methods in the principles of the various aspects may be utilized, and the written description is intended to encompass all such aspects and their equivalents.

Additionally, terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but the components are not limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be named a second component, and similarly, the second component may also be named a first component without departing from the scope of the present invention. The term and/or includes any of a plurality of related stated items or a combination of a plurality of related stated items.

In addition, in the embodiments of the present invention, unless otherwise defined, all terms used herein, including technical or scientific terms, are generally understood by those skilled in the art to which the present invention pertains. It has the same meaning as Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and unless clearly defined in the embodiments of the present invention, have an ideal or excessively formal meaning. It is not interpreted as

Meanwhile, the "legal provisions" mentioned below are a concept that includes all laws, enforcement decrees, enforcement regulations, and administrative rules, and according to one embodiment of the present invention, precedents, administrative guidance, supervisory administration, and oral instructions are further included. It can be included.

Figure 1 schematically shows the performance steps of a method for deriving regulatory legal provisions according to an embodiment of the present invention and the configuration of a computing system 1000 that performs this, and Figure 2 shows a computing system 1000 according to an embodiment of the present invention. The process of deriving a list of related regulatory laws from the system 1000 is schematically shown.

As shown in FIG. 1, a method of deriving regulatory legal provisions related to an input query, performed in a computing system 1000 including one or more processors and one or more memories, includes regulatory card information among a plurality of legal provisions. A regulatory score derivation step (S100) that uses the regulatory classification model 2100 learned with the regulation card information of the existing legal provisions to derive a regulatory score indicating how related each of the plurality of legal provisions stored in the database 2000 is to the regulation. ); A similarity score derivation step (S200) of deriving a similarity score according to the similarity between a query input from a user and each of the plurality of legal provisions using one or more artificial neural network-based inference models; and a related regulatory law provision step (S300) in which regulatory legal provisions related to the query are classified and sorted according to preset standards based on the regulatory score and the similarity score and provided to the user.

Schematically, Figure 1 (a) shows the execution steps of the method for deriving regulatory legal provisions, and Figure 1 (b) shows the configuration of the computing system 1000 that performs the method for deriving the regulatory legal provisions. It shows.

Specifically, the regulatory score derivation unit 1100 of the computing system 1000 performs a regulatory score derivation step (S100) of deriving a regulatory score for each of all legal provisions stored in the database 2000. The database 2000 stores a plurality of legal provisions and information corresponding to each legal provision. Information corresponding to the above legal provisions includes the title of the law, the original text of the legal provisions, information on the upper law of the relevant law, relevant department, date of promulgation, date of enforcement, and regulatory information. It is desirable that the above regulatory information corresponds to the information included in the regulatory registration card released by the Regulatory Reform Committee.

The regulatory score derivation unit 1100 derives a regulatory score for each of all legal provisions stored in the database 2000 using a regulatory classification model 2100 that has learned legal provisions containing regulatory information. According to one embodiment of the present invention, the regulatory classification model 2100 is stored in the database 2000. Additional explanation of the regulatory score derivation step (S100) will be described later with reference to FIGS. 3 and 4, and the computing device provides query information (hereinafter referred to as query) input from the user after the regulatory score derivation step (S100). Based on this, the similarity score derivation step (S200) and the related regulatory law provision step (S300) are performed to derive regulatory law provisions related to the query information entered by the user.

As shown in FIG. 2, the user can input a search word or search statement through his or her terminal, and the terminal generates a query based on the input search word or search statement, and sends the generated query to the computing system ( 1000). Or, as another embodiment of the present invention, the terminal may transmit the input search word or search sentence to the computing system 1000, and generate the search word or search sentence received from the computing system 1000 as query information. . The computing system 1000 that has received this may derive regulatory law provisions related to the data included in the query, generate a list of regulatory laws, and provide the generated list to the user.

It is desirable for the user to enter legal provisions in the field he or she desires as the search word or search statement. For example, if the user wants to know the regulatory laws regarding clinical trials, the user enters ‘Drug clinical trial plan approval procedures and standards’ as a search term, or ‘Article 24 of the Rules on the Safety of Pharmaceuticals, etc.’ Alternatively, you can enter the full text of Article 24 of the Rules on the Safety of Medicines, etc. as a search query.

Meanwhile, according to another embodiment of the present invention, the user can input the sentence 'Please tell me the regulatory laws related to clinical trials in the bio healthcare field' as a search sentence, and according to another embodiment of the present invention, Users can enter only key keywords as search terms, such as 'bio healthcare, clinical trials, and regulations'.

The similarity score deriving unit 1200, which has received a query containing data related to such a search term or search statement, uses one or more artificial neural network-based inference models to determine the similarity between the query input from the user and each of the plurality of legal provisions. A similarity score derivation step (S200) is performed to derive a similarity score according to the present invention. Briefly, the similarity score derivation step (S200) includes a keyword matching score derivation step (S210) and a cosine similarity score derivation step (S220), and the similarity score derivation unit 1200 calculates the keyword matching score and cosine derived through this. Based on the similarity score, the similarity score between the input query and the legal provisions is derived. A more detailed description of the similarity score derivation step (S200) will be described later in the description of FIGS. 5 to 10.

When a regulatory score is derived for each legal provision stored in the database (2000) and a similarity score between the input query and each of the legal provisions is derived, the related regulation law provision unit 1300 determines the score based on the similarity score and the regulatory score. A related regulatory law provision step (S300) is performed to derive regulatory legal provisions related to the entered query.

More specifically, relevant regulatory laws refer to regulatory legal provisions related to the query entered by the user, and the relevant regulatory laws include regulatory laws designated by government agencies, etc. In addition, even if they are legal provisions that are not designated as regulatory laws, they are in fact regulatory provisions. It is characterized by the ability to further include legal provisions corresponding to so-called 'shadow regulations' that play a role.

In other words, the present invention applies regulatory laws and shadow regulations designated by the government in the current law in the field desired by the user based on the regulatory score and similarity score derived through the above-described regulatory score derivation step (S100) and similarity score derivation step (S200). The corresponding legal provisions can be derived, and through this, users can easily find regulatory legal provisions related to the field they wish to proceed or review even if they do not have sufficient legal knowledge.

In addition, the relevant regulatory law provision step (S300) provides a list of regulatory laws classified by topic based on the actual legal provisions for the derived related regulatory laws, thereby providing legal provisions substantially related to the search word or search statement entered by the user. It can be effective in providing legal provisions in a way that is structured so that lower-level laws are opened within a higher-level law folder in a classification system that simply lists legal provisions by relevant ministries in conventional technology or related platforms, or in a category composed of topics related to real life. Compared to the method of providing a classification system based on , working-level staff can effectively provide more practical and necessary legal provisions.

Below, the above-described regulatory score derivation step (S100), similarity score derivation step (S200), and related regulatory law provision step (S300) will be described in more detail along with related drawings.

Figure 3 schematically shows the process of learning the regulatory classification model 2100 according to an embodiment of the present invention.

Specifically, the database 2000 stores a plurality of legal provisions and legal information for each of the plurality of legal provisions. The legal information, as described above in the description of Figure 1, includes the title of the law, the original text of the law, information on the higher law of the law, relevant department, date of promulgation, date of enforcement, and regulatory information, and in the [Background of the invention] in this specification As described in [Technology that becomes this], the regulatory information may be stored in the form of a regulatory registration card. The regulatory registration card includes information such as regulatory office name, regulatory summary, promulgation date, enforcement date, sunset information, name of law, main search words, related laws, and name of relevant ministry. In the case of legal provisions other than regulatory laws designated by the government, There is no regulatory information in legal provisions.

As shown in FIG. 3, the regulatory classification model 2100 performs a process of learning about legal provisions for which the regulatory information (regulation card information in FIG. 3) exists among all legal provisions stored in the database 2000. Additionally, the regulatory legal model learns not only the aforementioned regulatory information, but also the entire original text of the legal provisions in which the regulatory information exists. Through this, the regulatory classification model 2100 analyzes the original text of the input legal provision and derives a regulatory score indicating how related the input legal provision is to the regulation.

In other words, even if it is a legal provision that was not classified as a regulatory legal provision from the beginning, in the case of a legal provision in which the rules derived from the learned regulatory legal provisions appear at a certain level or higher, a regulatory score higher than a certain value is performed in the regulatory score derivation step (S100) of the present invention. It is derived through this method, and the technical feature is that the relevant legal provisions are classified as regulatory legal provisions accordingly.

An example of a rule derived from the above regulatory legal provisions is "If it contains words with mandatory or necessary force, such as 'regulation', 'obligation', or 'must', more than a preset number, it is classified as a regulatory legal provision." There may be etc. The process of deriving the regulatory score through the regulatory classification model 2100 learned as above will be described in the explanation of FIG. 4.

Figure 4 schematically shows the process of deriving a regulatory score for each legal provision through the learned regulatory classification model 2100 according to an embodiment of the present invention.

Schematically, Figure 4(a) shows the process of deriving the regulatory score of the corresponding legal provision by inputting an unclassified legal provision into the regulatory classification model 2100, and Figure 4(b) shows the process of deriving the regulatory score of the regulatory score list. One embodiment is shown.

Specifically, with reference to the contents of FIG. 3, as shown in (a) of FIG. 4, in the regulatory score derivation step (S100), when a legal provision is entered into the regulatory classification model (2100), the regulation for the corresponding legal provision is Derive a score. Preferably, the regulatory score derivation step (S100) may be performed on all legal provisions stored in the database 2000 before receiving a query from the user. The regulatory score can be expressed as a real number between 0 and 1, as shown in (b) of FIG. 4, and the closer it is to 1, the closer it is to 1, meaning that the relevant legal provision is closely related to regulation.

As an embodiment of the present invention, the relevant regulatory law provision unit 1300 converts legal provisions with regulatory scores greater than or equal to a preset standard into regulatory laws based on the regulatory score derived through the regulatory score derivation step (S100). You can judge. For example, a legal provision with a regulatory score of 0.800 or higher can be judged as a regulatory law.

In other words, the regulatory score derivation step (S100) of the present invention uses a regulatory classification model (2100) that has learned regulatory laws designated by the government, etc., to determine regulatory laws in the field where the relevant laws are actually applied even though they have not been designated as regulatory laws by the government, etc. It can have the effect of deriving regulatory laws corresponding to shadow regulations that can act as.

Figure 5 schematically shows the execution steps of the similarity score derivation step (S200) according to an embodiment of the present invention.

As shown in Figure 5, the similarity score derivation step (S200) performed by the similarity score derivation unit 1200 includes a keyword matching score derivation step (S210) of deriving a keyword matching score; A cosine similarity score derivation step (S220) of deriving a cosine similarity score; and a relevance score derivation step (S230) for deriving a relevance score indicating how substantially the input query is related to the candidate legal provisions derived through the similarity score derivation step (S200) and the keyword matching score derivation step (S210); Includes.

As an embodiment of the present invention, in FIG. 5, the keyword matching score derivation step (S210) is shown as preceding the cosine similarity score derivation step (S220). However, according to another embodiment of the present invention, the cosine similarity score derivation step (S220) is shown in FIG. Step (S220) may precede the keyword matching score derivation step (S210), and this corresponds to a matter that can be changed in design as much as desired by the inventor of the present invention.

Therefore, the similarity score derivation unit 1200 derives the keyword matching score, the cosine similarity score, and the relevance score to calculate a similarity score between the query input from the user and each of the plurality of legal provisions stored in the database 2000. Perform the similarity score derivation step (S200).

Figures 6 and 7 schematically show the execution process of the keyword matching score derivation step (S210) according to several embodiments of the present invention.

As shown in Figures 6 and 7, the similarity score derivation step (S200) is a query input from a user in the first inference model 2300 based on an artificial neural network and a plurality of legal provisions stored in the database 2000. It includes a keyword matching score derivation step (S210) in which each keyword is entered to derive a keyword matching score.

Additionally, as shown in FIG. 7, the keyword matching score derivation step (S210) includes extracting keywords from a query input from the user; Converting each of the extracted plurality of keywords into a vector having a dimension of 1 or more, and then deriving the converted plurality of vectors into one keyword vector; and entering a legal provision and a keyword vector into the first inference model 2300 based on an artificial neural network, thereby deriving a keyword matching score indicating the similarity between the input legal provision and the keyword vector.

Specifically, Figure 6 schematically shows the process of deriving a keyword matching score as an embodiment of the present invention. As shown in (a) of FIG. 6, the similarity score deriving unit 1200 extracts a plurality of keywords (A , B, C, D) are derived. Figures 6 and 7 show an example of deriving four keywords (A, B, C, D) from one query, but the number of keywords actually derived in the present invention is not limited to those shown in Figures 6 and 7. No.

Afterwards, as shown in (b) of FIG. 6, the similarity score deriving unit 1200 calculates the keyword set {A, B, C, D} of the query entered by the user and each of the legal provisions stored in the database 2000. The keyword matching score is derived by inputting it into the first inference model (2300) based on artificial intelligence.

The keyword matching score determines how many keywords (A, B, C, D) belonging to the keyword set {A, B, C, D} appear in the legal provisions input to the first inference model 2300, or the number of keywords (A, B, C, D) belonging to the keyword set {A, B, C, D}. It is a score that quantifies the occurrence of noun phrases related to keywords (A, B, C, D) belonging to the keyword set {A, B, C, D}. The higher the keyword matching score, the more keywords in the query are included in the legal provisions. Alternatively, it can be understood as a legal provision that appears with high frequency.

Meanwhile, the keyword matching score is another embodiment of the present invention, as shown in FIG. 7, rather than directly inputting the keyword set {A, B, C, D} into the first inference model 2300, After converting the keyword set {A, B, C, D} into a keyword vector, each of the keyword vector and the legal provisions stored in the database (2000) is input into the first inference model (2300) to derive a keyword matching score. You can.

More specifically, the similarity score deriving unit 1200 derives a keyword set {A, B, C, D} through the process shown in (a) of FIG. 6 and then converts the words into a vector with a dimension of 1 or more. By entering each keyword belonging to the keyword set into the conversion matrix, vectors (V _A , V _B , V _C , V _D ) for each keyword can be derived. According to various embodiments of the present invention, the function performed by the matrix may be performed by a model or algorithm that converts text into a corresponding vector, such as W2V. When a vector for each of the keywords (V _A , V _B , V _C , V _D ) is derived, the similarity score deriving unit 1200 converts them into one keyword vector, as shown in (a) of FIG. 7. Derived as (V _A+B+C+D ). For example, the corresponding keyword vector can be derived by adding up the vectors for each of the keywords (V _A , V _B , V _C , and V _D ).

When the keyword vector (V _A+B+C+D ) is derived, the similarity score derivation unit 1200 combines the derived keyword vector (V _A+B+C+D ) and the legal provisions with the first inference model (2300). ), you can derive the keyword matching score between the relevant legal provisions and the query.

Figure 8 schematically shows the process of performing the cosine similarity score derivation step (S220) according to an embodiment of the present invention.

As shown in FIG. 8, the cosine similarity score derivation step (S220) includes deriving a query vector for the query input from the user through an artificial neural network-based word embedding model (2400); and deriving a cosine similarity score based on the cosine similarity derived between each of all legal provisions stored in the database 2000 and the query vector.

Schematically, Figure 8(a) shows the process of deriving a query vector (V query) by inputting a query input from a user into the word embedding model 2400, and Figure 8(b) shows the process of deriving a query vector (V _query ) from the database 2000. Shows the process of deriving a legal vector (V _law ) for each legal provision by inputting each of a plurality of legal provisions stored in the word embedding model 2400, and Figure 8 (c) shows the query vector (V The process of deriving the cosine similarity score between _query ) and the law vector (V _law ) is shown.

Specifically, the similarity score deriving unit 1200 inputs each of the query input from the user and the legal provisions stored in the database 2000 into the word embedding model 2400 shown in FIG. 8 to generate a query vector (V _query ) and a plurality of After deriving a legal vector (V _law ), cosine similarity is derived between the query vector (V _query ) and the law vector (V _law ) to obtain a cosine similarity score for each of the query and the legal provisions stored in the database (2000). Derive . The word embedding model 2400 performs a similar function to the matrix in the keyword matching score derivation step (S210), but preferably corresponds to a different model or algorithm. That is, both the keyword vector (V _A+B+C+D ) and the query vector (V _query ) mean vectors for queries, but they correspond to different vectors.

V _law1 , V _law2 , V _law3 , V _law4 , V _law5 , and V _law6 shown in FIG. 8 are LAW#1, LAW#2, LAW#3, LAW#4, and LAW#, respectively, with reference to FIG. 2. 5, and LAW #6 correspond to a vector with a dimension of 1 or more output by inputting it into the word embedding model (2400), and the process of obtaining the legal vector (V _law ) is based on all legal provisions stored in the database (2000). It is desirable to carry out

In other words, the similarity score deriving unit 1200 derives a query vector (V _query ) and a plurality of law vectors (V _law ) through the word embedding model 2400, as shown in (c) of FIG. 8. As described above, the cosine similarity between each of the query vector (V _query ) and the plurality of law vectors (V _law ) is calculated, and then the cosine similarity score is derived based on the calculated cosine similarity.

As described above, the similarity score deriving unit 1200 derives a similarity score according to the similarity between the query input from the user and the legal provisions based on different scores (keyword matching score and cosine similarity score) derived in different ways. Through this, the similarity score can be derived more accurately than deriving the similarity score using a single method.

Figure 9 schematically shows a process for deriving candidate legal provisions according to an embodiment of the present invention, and Figure 10 schematically shows a process for deriving a similarity score according to an embodiment of the present invention.

As shown in Figures 9 and 10, the similarity score derivation step (S200) is a second inference model based on an artificial neural network based on legal provisions and input queries derived based on the keyword matching score and the cosine similarity score. It further includes a relevance score derivation step (S230) of entering into (2500) and deriving a relevance score indicating the correlation between the derived legal provision and the input query, wherein the keyword matching score, the cosine similarity score, and the Based on the relevance score, a similarity score is derived for each of the plurality of legal provisions above.

Specifically, according to one embodiment of the present invention, the similarity score deriving unit 1200 may derive the similarity score of the relevant legal provision based on the keyword matching score and the cosine similarity score. Meanwhile, preferably, as another embodiment of the present invention, a similarity score for each of the plurality of legal provisions can be derived based on the keyword matching score, the cosine similarity score, and the relevance score.

As shown in (a) of FIG. 9, the similarity score deriving unit 1200 calculates an intermediate score for each legal provision based on the keyword matching score and cosine similarity score for all legal provisions stored in the database 2000. is derived, and as shown in (b) of Figure 9, all the legal provisions are listed in order of the intermediate score.

Afterwards, among all the above legal provisions, some are selected as candidate legal provisions according to pre-established conditions. In Figure 9, an embodiment of selecting legal provisions with the top 100 intermediate scores as candidate legal provisions is shown. However, the configuration of selecting candidate legal provisions of the present invention is not limited to the above-described embodiment and can be selected through various embodiments. A number of candidate legal provisions can be selected.

In this way, each of the plurality of legal provisions selected as candidate legal provisions and the query input from the user are entered into the artificial intelligence-based second inference model 2500 to derive a correlation score between the query and each candidate legal provision. The relevance score derivation step (S230) is performed by the similarity score derivation unit 1200. In a preferred embodiment, the relevance score may be derived based on self-supervised learning using a hard negative mixing method.

The similarity score derivation unit 1200 derives a similarity score different from the input query and the legal provisions based on the keyword matching score, cosine similarity score, and relevance score derived through the above-described process. Thereafter, referring to the explanation of FIG. 1, the relevant regulatory law provision unit 1300 of the computing system 1000 derives regulatory law provisions related to the query entered by the user based on the regulatory score and the similarity score, and , provided to the user.

Figure 11 schematically shows the process of determining regulatory law provisions according to an embodiment of the present invention, and Figure 12 schematically shows a list of regulatory laws provided to users according to an embodiment of the present invention.

As shown in Figures 11 and 12, as an embodiment of the present invention, the relevant regulatory law provision step (S300) classifies regulatory legal provisions related to the query based on the similarity score, and preset regulations After determining the legal provisions higher than the score as regulatory legal provisions, the determined regulatory legal provisions are provided to the user in the form of a list arranged in order according to preset conditions.

Meanwhile, in another embodiment of the present invention, the relevant regulatory law providing step (S300) classifies regulatory legal provisions related to the query based on the similarity score, and classifies the regulatory legal provisions related to the query based on the similarity score, After determining the legal provisions above the standard regulatory score calculated based on the regulatory score as regulatory legal provisions, the determined regulatory legal provisions are provided to the user in the form of a list arranged in order according to preset conditions.

Specifically, as shown in FIG. 11, the relevant regulatory law provision unit 1300 determines whether the relevant legal provision is a relevant regulatory legal provision based on the similarity score and regulatory score for the legal provision, The relevant regulatory law provision step (S300) is performed in which a plurality of legal provisions determined to be relevant regulatory legal provisions are provided in the form of a list.

An example of the relevant regulatory provisions provided in the form of the list is shown in FIG. 12. As shown in Figure 12, with reference to the explanation of Figure 3, the list includes the title of the law, the original text of the law, information on the higher law of the law, relevant department, date of publication, date of enforcement, and regulation information, for convenience of explanation. For this purpose, Figure 12 shows only the title of the law, the text of the legal provision, the similarity score of the relevant legal provision, and the regulatory score of the relevant legal provision. The similarity score can be expressed as a real number between 0 and 1, similar to the regulation score described above in the description of FIG. 4, and the closer it is to 1, the closer the legal provision is to the entered query.

The relevant regulatory law provision step (S300) may generate the list in order of similarity scores for queries entered by the user. At this time, only legal provisions that are higher than a preset regulatory score can be included in the list, or after calculating the standard regulatory score according to a specific rule, only legal provisions that are higher than the standard regulatory score can be included in the list.

As an embodiment of the present invention, the standard regulatory score is based on the regulatory scores of legal provisions with a preset number of higher similarity scores when listing a plurality of legal provisions based only on the pure similarity score, regardless of the regulatory score. can be derived. For example, when a plurality of legal provisions are listed based on the similarity score, the average value of the regulatory scores of the top 50 legal provisions can be determined as the standard regulatory score.

The present invention provides users with a list of related regulatory laws derived through the above-described embodiments, so that users can easily search for regulatory laws and regulations in the field they are looking for by simply entering a search word or search statement. It can be provided in a structured way to open lower-level laws within a higher-level law folder in a classification system that simply lists legal provisions by relevant ministries in the prior art or related platforms, or provides a classification system based on categories made up of topics related to real life. Compared to the provision method, working-level staff can provide more practical and necessary legal provisions more effectively.

FIG. 13 exemplarily illustrates the internal configuration of a computing device 11000 according to an embodiment of the present invention.

The computing system 1000 mentioned in the description of FIG. 1 may include components of the computing device 11000 shown in FIG. 13, which will be described later.

As shown in FIG. 13, the computing device 11000 includes at least one processor 11100, a memory 11200, a peripheral interface 11300, and an input/output subsystem ( It may include at least an I/O subsystem (11400), a power circuit (11500), and a communication circuit (11600).

Specifically, the memory 11200 may be, for example, high-speed random access memory, magnetic disk, SRAM, DRAM, ROM, flash memory, or non-volatile memory. may include. The memory 11200 may include software modules, instruction sets, or other various data necessary for the operation of the computing device 11000.

At this time, access to the memory 11200 from other components such as the processor 11100 or the peripheral device interface 11300 may be controlled by the processor 11100. The processor 11100 may be composed of a single processor or a plurality of processors, and may include GPU and TPU type processors to improve calculation processing speed.

The peripheral device interface 11300 may couple input and/or output peripheral devices of the computing device 11000 to the processor 11100 and the memory 11200. The processor 11100 may execute a software module or set of instructions stored in the memory 11200 to perform various functions for the computing device 11000 and process data.

The input/output subsystem 11400 can couple various input/output peripheral devices to the peripheral device interface 11300. For example, the input/output subsystem 11400 may include a controller for coupling peripheral devices such as a monitor, keyboard, mouse, printer, or, if necessary, a touch screen or sensor to the peripheral device interface 11300. there is. According to another aspect, the input/output peripheral devices may be coupled to the peripheral interface 11300 without going through the input/output subsystem 11400.

The power circuit 11500 may supply power to all or part of the components of the terminal. For example, the power circuit 11500 may include a power management system, one or more power sources such as batteries or alternating current (AC), a charging system, a power failure detection circuit, a power converter or inverter, a power status indicator, or It may include any other components for power generation, management, and distribution.

The communication circuit 11600 may enable communication with another computing device using at least one external port. Alternatively, as described above, if necessary, the communication circuit 11600 may include an RF circuit to transmit and receive RF signals, also known as electromagnetic signals, to enable communication with other computing devices.

This embodiment of FIG. 13 is only an example of the computing device 11000, and the computing device 11000 may omit some components shown in FIG. 13 or further include additional components not shown in FIG. 13. , may have a configuration or arrangement that combines two or more components. For example, a computing device for a communication terminal in a mobile environment may further include a touch screen or a sensor in addition to the components shown in FIG. 13, and the communication circuit 1160 may include various communication methods (Wi-Fi, Circuits for RF communication (3G, LTE, 5G, 6G, Bluetooth, NFC, Zigbee, etc.) may be included. Components that can be included in the computing device 11000 may be implemented as hardware, software, or a combination of both hardware and software, including an integrated circuit specialized for one or more signal processing or applications.

Methods according to embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computing devices and recorded on a computer-readable medium. In particular, the program according to this embodiment may be composed of a PC-based program or a mobile terminal-specific application. The application to which the present invention is applied can be installed on a user terminal through a file provided by a file distribution system. As an example, the file distribution system may include a file transmission unit (not shown) that transmits the file according to a request from the user terminal.

The device described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), etc. , may be implemented using one or more general-purpose or special-purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include. For example, a processing device may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.

Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device. Software and/or data may be used by any type of machine, component, physical device, virtual equipment, computer storage medium or device to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be stored or executed in a standardized manner on a networked computing device. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

According to one embodiment of the present invention, laws that are not classified as regulations are derived by deriving regulatory legal provisions based on regulatory scores derived through a learned regulatory classification model without providing information on regulatory legal provisions designated by the government as is. This can have the effect of making it easier to find shadow regulation provisions that are recognized as regulations.

According to one embodiment of the present invention, when a company, etc., is reviewing a new business or preparing to start a business, it is possible to easily find regulatory provisions related to the business to be conducted.

According to one embodiment of the present invention, government ministries, legislative bodies, and national research institutes reviewing regulations can easily find legal provisions related to a specific field, which has the effect of easily monitoring legal provisions that require legislation or revision. can be demonstrated.

According to one embodiment of the present invention, users can enter search terms in various formats, allowing users to easily search for regulatory provisions in the field they are looking for even if they do not have sufficient legal knowledge such as legal terms.

According to one embodiment of the present invention, by providing a list of regulatory laws classified by subject based on the contents of actual legal provisions rather than a list classified by field of law or relevant ministry, legal provisions substantially related to the search word or search statement entered by the user It can be effective in providing .

As described above, although the embodiments have been described with limited examples and drawings, various modifications and variations can be made by those skilled in the art from the above description. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent. Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

Claims (9)

A method of deriving regulatory provisions related to an input query, performed on a computing system including one or more processors and one or more memories,
Among the plurality of legal provisions, a regulatory score indicating how related each of the plurality of legal provisions stored in the database is to the regulation is derived using a regulatory classification model learned with the regulation card information of the legal provisions with regulatory card information, and the regulation score is derived. A regulatory score derivation step where the score corresponds to a real number between 0 and 1, and the closer it is to 1, the closer the legal provision is to regulation.
One or more artificial neural network-based inference models are used to derive a similarity score based on the similarity between a query containing data related to a search word or search statement entered by the user and each of the plurality of legal provisions, where the similarity score is between 0 and 1. A similarity score derivation step, where the closer to 1 the closer the legal provision is to the entered query, the more similar the legal provision is to the entered query; and
Based on the similarity score, the regulatory legal provisions related to the query are classified, a list listing a plurality of legal provisions is created according to the similarity score, and the regulatory score of the legal provisions with a preset number of higher similarity scores in the list is generated. A related regulatory law provision step of determining legal provisions corresponding to a regulatory score higher than the standard regulatory score calculated based on the regulatory legal provisions as regulatory legal provisions, and then providing the determined regulatory legal provisions to the user in the form of a list arranged in order according to preset conditions; Contains,
The similarity score derivation step is,
A keyword matching score derivation step of deriving a keyword matching score by inputting a query input from a user and a plurality of legal provisions stored in the database into a first inference model based on an artificial neural network;
A cosine similarity score deriving step of deriving a cosine similarity score between each legal provision stored in the database and a query input from a user;
Based on the keyword matching score and cosine similarity score for all legal provisions stored in the database, an intermediate score is derived for each legal provision, all legal provisions are listed in order of the intermediate score, and some legal provisions are selected among all legal provisions. Selecting them as candidate legal provisions according to preset conditions; and
The candidate legal provisions and the input query are entered into a second inference model based on an artificial neural network to derive a correlation score indicating the correlation between the candidate legal provisions and the input query, and the correlation score is calculated using a hard negative mixing method. It includes a correlation score derivation step derived based on self-supervised learning,
A method of deriving regulatory legal provisions, wherein a similarity score for each of the plurality of legal provisions is derived based on the keyword matching score, the cosine similarity score, and the relevance score.
delete In claim 1,
The keyword matching score derivation step is,
extracting keywords from a query input from the user;
Converting each of the extracted plurality of keywords into a vector having a dimension of 1 or more, and then deriving the converted plurality of vectors into one keyword vector; and
A method of deriving regulatory legal provisions, comprising: inputting legal provisions and keyword vectors into a first inference model based on an artificial neural network, thereby deriving a keyword matching score indicating the similarity between the input legal provisions and the keyword vectors.
In claim 1,
The cosine similarity score derivation step is,
Deriving a query vector for the query input from the user through an artificial neural network-based word embedding model; and
A method of deriving regulatory legal provisions comprising: deriving a cosine similarity score based on cosine similarity derived between each of all legal provisions stored in the database and the query vector.
delete delete delete A computing system comprising one or more processors and one or more memories, and performing a method of deriving regulatory provisions related to an input query,
Among the plurality of legal provisions, a regulatory score indicating how related each of the plurality of legal provisions stored in the database is to the regulation is derived using a regulatory classification model learned with the regulation card information of the legal provisions with regulatory card information, and the regulation score is derived. The score corresponds to a real number between 0 and 1, and the closer it is to 1, the closer the legal provision is to regulation, meaning that the legal provision is closely related to the regulation.
One or more artificial neural network-based inference models are used to derive a similarity score based on the similarity between a query containing data related to a search word or search statement entered by the user and each of the plurality of legal provisions, where the similarity score is between 0 and 1. As a real number, the closer to 1, the similarity score derivation section indicates that the corresponding legal provision is similar to the entered query; and
Based on the similarity score, the regulatory legal provisions related to the query are classified, a list listing a plurality of legal provisions is created according to the similarity score, and the regulatory score of the legal provisions with a preset number of higher similarity scores in the list is generated. A related regulatory law provision department that determines the legal provisions corresponding to a regulatory score higher than the standard regulatory score calculated based on the regulatory legal provisions as regulatory legal provisions and provides the determined regulatory legal provisions to the user in the form of a list arranged in order according to preset conditions; Contains,
The similarity score derivation unit,
A keyword matching score derivation step of deriving a keyword matching score by inputting a query input from a user and a plurality of legal provisions stored in the database into a first inference model based on an artificial neural network;
A cosine similarity score deriving step of deriving a cosine similarity score between each legal provision stored in the database and a query input from a user;
Based on the keyword matching score and cosine similarity score for all legal provisions stored in the database, an intermediate score is derived for each legal provision, all legal provisions are listed in order of the intermediate score, and some legal provisions are selected among all legal provisions. Selecting them as candidate legal provisions according to preset conditions; and
The candidate legal provisions and the input query are entered into a second inference model based on an artificial neural network to derive a correlation score indicating the correlation between the candidate legal provisions and the input query, and the correlation score is calculated using a hard negative mixing method. Perform a correlation score derivation step derived based on self-supervised learning,
A computing system that derives a similarity score for each of the plurality of legal provisions based on the keyword matching score, the cosine similarity score, and the relevance score.
A computer-readable recording medium for implementing a method of deriving regulatory legal provisions related to an inputted query, performed on a computing system comprising at least one processor and at least one memory, comprising:
The computer-readable recording medium stores instructions that cause the computing system to perform the following steps,
The steps below are:
Among the plurality of legal provisions, a regulatory score indicating how related each of the plurality of legal provisions stored in the database is to the regulation is derived using a regulatory classification model learned with the regulation card information of the legal provisions with regulatory card information, and the regulation score is derived. A regulatory score derivation step where the score corresponds to a real number between 0 and 1, and the closer it is to 1, the closer the legal provision is to regulation.
One or more artificial neural network-based inference models are used to derive a similarity score based on the similarity between a query containing data related to a search word or search statement entered by the user and each of the plurality of legal provisions, where the similarity score is between 0 and 1. A similarity score derivation step, where the closer to 1 the closer the legal provision is to the entered query, the more similar the legal provision is to the entered query; and
Based on the similarity score, the regulatory legal provisions related to the query are classified, a list listing a plurality of legal provisions is created according to the similarity score, and the regulatory score of the legal provisions with a preset number of higher similarity scores in the list is generated. A related regulatory law provision step of determining legal provisions corresponding to a regulatory score higher than the standard regulatory score calculated based on the regulatory legal provisions as regulatory legal provisions, and then providing the determined regulatory legal provisions to the user in the form of a list arranged in order according to preset conditions; Contains,
The similarity score derivation step is,
A keyword matching score derivation step of deriving a keyword matching score by inputting a query input from a user and a plurality of legal provisions stored in the database into a first inference model based on an artificial neural network;
A cosine similarity score deriving step of deriving a cosine similarity score between each legal provision stored in the database and a query input from a user;
Based on the keyword matching score and cosine similarity score for all legal provisions stored in the database, an intermediate score is derived for each legal provision, all legal provisions are listed in order of the intermediate score, and some legal provisions are selected among all legal provisions. Selecting them as candidate legal provisions according to preset conditions; and
The candidate legal provisions and the input query are entered into a second inference model based on an artificial neural network to derive a correlation score indicating the correlation between the candidate legal provisions and the input query, and the correlation score is calculated using a hard negative mixing method. It includes a correlation score derivation step derived based on self-supervised learning,
A computer-readable recording medium for deriving a similarity score for each of the plurality of legal provisions based on the keyword matching score, the cosine similarity score, and the relevance score.

Images