KR102091633B1 - Searching Method for Related Law - Google Patents

Abstract

Disclosed is a method for providing related laws. The method of providing the related laws includes searching for at least one judgment sentence from a search server for an input keyword, extracting reference sentence data included in the searched at least one judgment sentence, and distinguishing the extracted reference sentence data. Parsing by reference, and vectorizing the parsed reference sentence data to generate word vectors for each of the legal terms included in the reference sentence data.

Description

Searching Method for Related Law}

The present invention relates to a method for providing a related law, and more particularly, to a method for providing a related law using a word embedding based Law2Vec model.

Legal information retrieval is a method that is used to find ground rules necessary for legal judgment and interpretation, ranging from judges, prosecutors, and investigators of individuals, corporations, and state agencies. Currently, there are about 5,000 laws and regulations currently in the National Legal Information Center, and as the number of precedents interpreting the law increases and becomes more complex, the demand for efficient information retrieval is increasing.

On the other hand, referring to Non-Patent Document 1 for a method of providing domestic legislation information using text analysis technology, a method for converting domestic legislation into an ontology is disclosed. However, the invention based on ontology has achieved meaningful results by constructing it in the road traffic related laws and regulations in the railway sector, but there is a problem that must be built manually, and recent research on the automatic construction method also has the ontology in the sentence of the statute. There is a limitation that it is difficult to convert when the pattern cannot be clearly applied.

In addition, the non-patent document 2 discloses a point in which a living term corresponds to a legal term, and the non-patent document 3 discloses a point in which a precedent is classified using topic modeling, but the former searches for a living term and a legal term. In order to do so, thesaurus must be searched and executed, so the thesaurus, which is built with an ontology in advance, is required, and the precedent classification method of topic modeling, such as the latter, has a limitation in that it can be classified as only one topic. There is a problem that it is difficult to apply to legal information.

 Jang In-hwan, “A study on the construction and evaluation of an ontology-based legal search system,” Journal of the Korean Society for Library and Information Science, Vol. 2, No. 2, 2011.5, 345-366  Kim Ji-hyun, “Methodology for Searching Correspondence between Living and Legal Terms for Searching Legal Information,” Intelligent Information Research Vol. 18, No. 3, 2012.9, 137-152  Shim Joon-sik, “Methods for Searching and Classifying Cases Using LDA Topic Modeling,” Journal of the Institute of Electronics Engineers of Korea, Vol. 54, No. 9 (passage No. 478), 2017.9, 67-75

An object of the present invention is to provide a method for providing related laws that can efficiently and accurately search for legal information related to a search word even if the exact contents of the legal knowledge are not known.

The method for providing an association law according to an embodiment of the present invention includes searching for at least one judgment sentence from a search server for an input keyword, extracting reference sentence data included in the searched at least one judgment sentence, and extracting the extracted keyword The method may include parsing the reference text data, based on a separator, and vectorizing the parsed reference text data to generate word vectors for each of the legal terms included in the reference text data.

According to an embodiment of the present invention, even if the exact contents of the legal knowledge are not known, the related laws can be searched to increase the efficiency of time and money.

In addition, unlike the existing statute search method only provides results that exactly match the keyword entered to search the search terms and conditions, the present invention provides a search result based on reference statute data that is highly related to the entered keyword. It can provide accurate and meaningful search results.

1 illustrates a system for providing a related statute according to an embodiment of the present invention.
2 is a block diagram of an input unit included in an apparatus for providing related laws and regulations according to an embodiment of the present invention.
3 illustrates a hierarchical diagram for explaining the Word2Vec algorithm that can be applied to the present invention.
4 is a diagram illustrating an example of a screen through which a related statute is output through a system and method for providing a related statute according to an embodiment of the present invention.
5 is a flowchart of a method for providing a related law according to an embodiment of the present invention.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are exemplified only for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention These can be implemented in various forms and are not limited to the embodiments described herein.

Embodiments according to the concept of the present invention can be applied to various changes and can have various forms, so that the embodiments will be illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosure forms, and includes changes, equivalents, or substitutes included in the spirit and scope of the present invention.

Terms such as first or second may be used to describe various components, but the components should not be limited by the terms. The above terms are only for the purpose of distinguishing one component from other components, for example, without departing from the scope of rights according to the concept of the present invention, the first component may be referred to as the second component, Similarly, the second component may also be referred to as the first component.

When an element is said to be "connected" or "connected" to another component, it is understood that other components may be directly connected to or connected to the other component, but other components may exist in the middle. It should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that no other component exists in the middle. Expressions describing the relationship between the elements, for example, "between" and "immediately between" or "directly neighboring to" should also be interpreted.

The terminology used herein is only used to describe specific embodiments and is not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, the terms "include" or "have" are intended to designate the presence of a feature, number, step, action, component, part, or combination thereof as described, one or more other features or numbers, It should be understood that the existence or addition possibilities of steps, actions, components, parts or combinations thereof are not excluded in advance.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains. Terms such as those defined in a commonly used dictionary should be interpreted as having meanings consistent with meanings in the context of related technologies, and should not be interpreted as ideal or excessively formal meanings unless explicitly defined herein. Does not. Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

1 illustrates a system for providing a related statute according to an embodiment of the present invention.

Referring to FIG. 1, a system 1 for providing a related law according to an embodiment of the present invention includes a user terminal 10, a search server 20, and a device 30 for providing related laws.

The user terminal 10 is configured to input a search word into the search server 20 to be described later. Here, the search word may be a statute for searching a keyword or a related statute for a specific subject for searching for judgments in the search server 20. Keywords may include legal terms for searching for rulings, such as extinction aging, damages, and infringement, for example.

On the other hand, the user terminal 10 is a smart phone (smartphone), tablet PC (tablet personal computer), mobile phone (mobile phone), desktop PC (desktop personal computer), laptop PC (laptop personal computer), netbook computer (netbook computer) ).

The search server 20 provides a search result to the user terminal 10 in response to a search query received from the user terminal 10. The search server 20 may select content to be provided to the user terminal 10 according to the search terms and search conditions included in the search query, and may provide search results in which content is listed according to the ranking most suitable for the search terms. .

The search server 20 according to an embodiment may be a server for searching for judgments on keywords entered by the user through the user terminal 10. That is, when the user inputs a keyword such as legal terminology, the search server 20 lists the judgment sentences according to the ranking most suitable for the keyword accordingly.

The search server 20 according to another embodiment may be a server for searching a related statute for a statute entered by the user through the user terminal 10. The search server 20 may provide and output, to the user terminal 10, the relevant laws and regulations provided from the apparatus 30 for providing related laws to be described later according to the user's input.

Meanwhile, the search server 20 may be in the form of separate or combined embodiments described above. In addition, the search server 20 is a portal site server that provides various content services such as cafes, mail, blogs, shopping, maps, dictionaries, news, securities, real estate, movies, music, bulletin boards, in addition to providing search engine services. Can be That is, the search server 20 may have a form in which a search engine and a portal site are combined.

The association law providing device 30 is a device for providing an association law related to a law entered by the user through the user terminal 10 and includes an input unit 310, a learning unit 330, and a provision unit 350.

The input unit 310 constructs a context by searching for at least one judgment sentence from the search server 20 and extracting reference sentence data from the found judgment sentences. Referring to FIG. 2, which shows a block diagram of an input unit included in an apparatus for providing related laws and regulations according to an embodiment of the present invention, the input unit 310 includes a search unit 311, an extraction unit 313, and a parsing unit 315. And a pre-processing unit 317.

The search unit 311 receives a keyword and searches and collects at least one judgment sentence from the search server 20. Keywords may include legal terms for searching for precedents related to a particular subject, as described above. Decisions collected from the online search server 20 of the Supreme Court of the Republic of Korea using specific keywords are shown in Table 1 below.

keyword category Number of documents Corpus A ruling court cause Extinction aging Supreme Court of the Republic of Korea Civil 911 Yes Detective 11 No lyrics 6 No Tax 93 No General administration 58 No Compensation for damages Supreme Court of the Republic of Korea Civil 6,988 Yes Detective 141 No lyrics 35 No Tax 67 No General administration 137 No

Referring to Table 1, it is possible to collect judgments according to keywords (extinction prescription, damages) from the online search server 20 of the Supreme Court of the Republic of Korea, and use documents (judgments) for litigation cases related to the entered keywords as a corpus. .

As described above, the search unit 311 may search for and collect judgments according to keywords, thereby minimizing time and resources required for learning, and use precedents including reference texts having a more relevant reference for learning. .

The extraction unit 313 extracts reference sentence data from the judgments searched through the search unit 311. The judgment document is a documented precedent, and includes a reasonable interpretation of a specific case based on the law, and includes contents corresponding to standardized identification items such as judgment, summary, reference, and judgment. At this time, the reference sentence included in the judgment is the legal sentence referenced in making a judgment on a specific case, and it is the most basic and important information that contains the jurisdiction set forth in the judgment. In the case of the Korean ruling, for example, the reference sentence is written together with the reference item [reference sentence] included in the judgment.

Reference clauses can be repeated in the same reference clauses in cases with similar issues, or multiple reference clauses can appear at the same time.The characteristics of these reference clauses are distributional in which words appearing in the same context share the same semantics. It can be said to be similar to hypothesis. Accordingly, the extracting unit 313 according to an embodiment of the present invention extracts only reference sentence data from the searched judgment sentences, so that all words included in the judgment sentence are not used for learning, but only reference sentence data is used for learning. .

The extracting unit 313 may extract reference sentence data through various embodiments according to the form of the judgment sentence. For example, in the case of the Korean ruling sentence, the extracting unit 313 may extract the strings listed in the next paragraph of the [reference sentence], which are the identification items included in the ruling sentence, as reference sentence data as described above. On the other hand, in the case of a judgment that does not include an identification item related to a reference clause, unlike a Korean judgment, reference clause data can be extracted using symbols or words that indicate legal provisions, such as § (paragraph) or Art.

The reference sentence data extracted through the extraction unit 313 may refer to a string in which legal provisions for each of the reference sentences included in the judgment sentences are listed. Here, the legal clause is a unit for distinguishing each legal clause constituting the statute, and for example, Korean statutes are classified in the order of Article, Clause, Item, and Order. That is, the extracting unit 313 extracts only the legal provisions for the reference provisions as reference provision data, not the entire legal provisions for the reference provisions indicated in the judgment. At this time, as an example of the reference sentence data, there is "Article ○ Article ○ ○" or "Legal Name Article ○".

As described above, the extraction unit 313 extracts only legal terms, not whole legal sentences, into data used for learning, thereby solving the difficulties of complex Hangul preprocessing and improving the search performance of related laws.

The parsing unit 315 constructs a context by parsing reference query data extracted through the extracting unit 313 based on a separator. As described above, when the reference provision data is a string in which legal provisions are listed, the parsing unit 315 parses the reference provision data in units of legal provisions based on the set separator.

Here, the pre-processing unit 317 may be further included. The preprocessing unit 317 preprocesses the parsed reference query data. Here, pre-processing may mean unifying the units of legal provisions included in the reference provision data. For example, when the unit of legal provisions is to be unified in units of trillions, the pre-processing unit 317 removes the terms, items, and titles corresponding to units less than trillion units from the legal provisions included in the reference provision data.

The learning unit 330 vectorizes the reference query data processed through the input unit 310 through a language processing algorithm. Here, the language processing algorithm may mean a word embedding algorithm used in the field of natural language processing. The word embedding is for expressing words in a corpus as a vector having a continuous value, and the vectorized word vector is learned by an algorithm so that other words and contextual meanings are preserved as much as possible.

The word embedding algorithm may include NNLM (Neural Network Language Model), RNNLM (Recurrent Neural Net Language Model), and Word2Vec, but the present invention is not limited thereto, and various algorithms may be applied to perform word embedding.

Hereinafter, an embodiment based on the Word2Vec algorithm will be described. Word2Vec is a word embedding algorithm that utilizes a language model based on a feedforward neural netword. It compensates for the problem of NNLM, which takes a lot of time in computation in the hidden layer and the output layer. It shows excellent performance for vector expression.

Word2Vec learns the meaning of a word using the relationship between the word in the corpus sentence and the adjacent word, assuming that the word with the same context has a close meaning. Words of related meanings are likely to appear close to each other in the document, so in the course of repeating learning, the two words may gradually have a close vector value.

There are two models of Word2Vec, CBOW (Continuous Bag-Of-Words) and skip-gram model. CBOW is a model that predicts a central word when a surrounding word is given, and skip-gram is a model that receives a central word as opposed to CBOW and predicts a surrounding word for it.

Referring to FIG. 3, which shows a hierarchical diagram for explaining the Word2Vec algorithm that can be applied to the present invention, the hierarchical diagram shown in FIG. 3 shows a skip-gram model in Word2Vec, and a reference sentence to be vectorized in the input layer. Words corresponding to legal provisions included in the data are vectorized into one-hot-encoding. In the Hiddeng layer, the data vectorized by one-hot-encoding is multiplied by the weighting matrix W and W` and exported to the Output layer. In the output layer, softmax calculation is performed, and the result of the calculation is backpropagated by comparing the result of calculation with a real word.

Here, the softmax calculation is performed by Equation 1 below.

는 중심 법령(o)이 주어졌을 때 주변 법령(c)이 등장할 조건부 확률,

는 중심 법령에 대한 벡터값,

는 주변 법령에 대한 벡터값을 의미한다.here,

Is the conditional probability that the surrounding statute (c) will appear when the central statute (o) is given,

Is the vector value for the central statute,

Denotes a vector value for surrounding laws and regulations.

을 최대화하는 방향으로 가중치 행렬 W,W`를 업데이트하면서 학습을 수행하여 서로 관련이 있는 법조항에 해당하는 벡터값들은 N차원 공간에서 서로 비슷한 곳에 위치하도록 한다.That is, Word2Vec is a conditional probability according to Equation 1 above.

By updating the weighting matrix W, W` in the direction of maximizing, the vector values corresponding to the legal provisions related to each other are located in the N-dimensional space similar to each other.

The learning unit 330 generates word vectors for each of the legal terms by vectorizing the legal terms included in the reference sentence data parsed through learning as described above, and provides them to the providing unit 350.

The providing unit 350 calculates similarity between word vectors generated through the learning unit 330, and provides the search server 20 with a legal provision corresponding to the word vector when the calculated similarity is highest. Similarity is a measure of semantic association between words, and in the present invention, it may mean a degree of association between legal provisions for learning.

The providing unit 350 according to an embodiment may calculate cosine similarity for arbitrary word vectors a and b as shown in Equation 2 below.

는 코사인 유사도를 의미한다.here,

Means cosine similarity.

That is, the provision unit 350 calculates the cosine similarity between all word vectors generated through the learning unit 330 as shown in Equation 2 above. On the other hand, the providing unit 350 according to an embodiment does not calculate the similarity among all word vectors, but among the generated word vectors, the word vector corresponding to the search target legal provision received by the search server 20 and the search Similarity may be calculated only among word vectors other than the word vector corresponding to the target legal provision.

As described above, the providing unit 350 may provide the search server 20 with a legal provision corresponding to the word vector when the similarity calculated using the similarity calculated between the word vectors is highest.

On the other hand, referring to FIG. 4 showing an example of a screen in which the relevant laws are output through the system and method for providing relevant laws according to an embodiment of the present invention, the providing unit 350 provides the search server 20 Legal provisions may be output at the bottom of the search window through the user terminal 10 connected to the search server 20.

5 is a flowchart of a method for providing a related law according to an embodiment of the present invention. Hereinafter, detailed descriptions of parts overlapping with those described above will be omitted.

Referring to FIG. 5, step S410 is a step of searching for at least one judgment sentence from the search server 20 for a keyword input through the search unit 311. In this case, the keyword may include legal terms for searching for precedents related to a specific subject.

Step S420 is a step of extracting reference sentence data included in at least one judgment sentence searched through step S410. Meanwhile, the extracted reference clause data may mean a string in which legal provisions for each of the reference clauses included in the judgments are listed. Here, the legal clause is a unit for distinguishing each legal clause constituting the statute, and for example, Korean statutes are classified in the order of Article, Clause, Item, and Order.

In step S430, pre-processing of the extracted reference query data is performed. Step S430 may be omitted when the reference query data is extracted in a 'tuple' unit in step S420.

In step S440, parsing is performed in units of legal clauses based on a separator for reference sentence data that has been extracted and preprocessed through steps S420 to S430.

In step S450, the reference sentence data parsed through step S440 is vectorized to generate word vectors for each of the legal terms included in the reference sentence data. As described above, step S450 may be performed through word embedding algorithms such as NNLM, RNNLM, and Word2Vec.

In step S460, among the word vectors generated through step S450, similarity is calculated between the word vectors corresponding to the search target legal terms input by the search server 20 and the remaining word vectors except the word vectors corresponding to the search target legal terms. . Here, the similarity may be, for example, cosine similarity.

In step S470, the search server 20 provides legal provisions corresponding to the word vector when the similarity calculated through step S460 is highest.

Hereinafter, a system and method for providing relevant laws and regulations according to an embodiment of the present invention described above will be compared with an existing method for searching and providing related laws and regulations. Table 2 below shows the relevant laws and regulations that are searched and provided through the system and method of providing relevant laws and regulations according to an embodiment of the present invention when the legal provisions to be searched are Article 168 of the Civil Code, and Table 3 is the legal provisions to be searched for Article 755 of the Civil Code. It shows the related laws that are searched and provided through existing related laws search and provision techniques.

Referring to Tables 2 and 3 above, in the case of the existing technique, only the legal provisions that included the words of Article 168 of the Civil Code and Article 755 of the Civil Code, which are the legal terms to be searched, were searched.

On the other hand, in the case of the present invention, when Article 168 of the Civil Code is a legal clause to be searched for, the relevant laws and regulations are derived from Article 169 of the Civil Code, Article 77 of the Speech Act and Article 430 of the Civil Code. In other words, through the information on the word vector of Article 168 of the Civil Code, a legal statement dealing with the effect of the cessation of prescription (Article 169), which can be regarded as having a semantic similarity to the 'Suspension of Extinctive Prescription', Article 168 of the Civil Code The legal provisions dealing with 'Minor's Responsibility' related to 'Supervisor's Responsibility' through the information on the word vector of Article 755 of the Civil Code (Article 753 of the Civil Code) or the legal provisions dealing with the 'estimation of proxy for business executives' Article 709 of the Civil Code) was derived.

Accordingly, the system and method for providing relevant laws and regulations according to an embodiment of the present invention are different from existing techniques for simply searching for laws and regulations that include the legal clause to be searched in the legal clause.

On the other hand, as described above, as well as performance evaluation through comparison with an existing technique, performance evaluation of a system and method for providing relevant laws and regulations according to an embodiment of the present invention may be performed through calculation of precision and reproducibility. Here, the precision and the reproducibility may be defined by Equations 3 and 4 below as a measure of how many keywords related to the keyword are included in the searched related laws.

Here, Precision is the precision, tp is the number of relevant laws among the related laws found, and fp is the number of non-conformance laws.

Here, Recall is the reproducibility, tn is the number of fits found, and fn is the number of fits not found.

In the case of learning and vectorizing the reference sentence data extracted from the judgments collected by searching and collecting keywords by using the equations 3 and 4 as 'extinction aging', the result of calculating the precision and reproducibility of the searched relevant statute is as follows: Table 4 below.

Search provisions Rank Precision [%] Reproducibility [%] Article 766 of the Civil Code 10 60 55 20 45 50 30 57 63 Article 166 of the Civil Code 10 70 70 20 63 60 30 64 60 Article 168 of the Civil Code 10 50 63 20 55 65 30 50 68 Average 20 57 62

In Table 4, Rank means the rank of the result of the related laws. Referring to Table 4 above, there is a higher probability that the related statute ranked above is a similar statute containing a key word. In addition, the more relevant laws are extracted, the more laws are derived with positive relationships, and the more laws with key words are derived.

As described above, the system and method for providing an association law according to an embodiment of the present invention provide results having a semantic correlation with the provisions of the search term and the provisions of the related laws that are searched and provided through balancing not only precision but also reproducibility. You can see that

The device described above may be implemented with hardware components, software components, and / or combinations of hardware components and software components. For example, the devices and components described in the embodiments include, for example, a processor, controller, arithmetic logic unit (ALU), digital signal processor (micro signal processor), microcomputer, field programmable array (FPA), It may be implemented using one or more general purpose computers 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 run an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, a processing device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and / or a plurality of types of processing elements. It can be seen that may include. For example, the processing device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible.

The software may include a computer program, code, instruction, or a combination of one or more of these, and configure the processing device to operate as desired, or process independently or collectively You can command the device. Software and / or data may be interpreted by a processing device, or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodied in the transmitted signal wave. The software may be distributed on networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in 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, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiments or may be known and usable by those skilled in 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, DVDs, and magnetic media such as floptical disks. -Hardware devices specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the embodiments have been described by the limited drawings as described above, a person skilled in the art can make various modifications and variations from the above description. For example, the described techniques are performed in a different order than the described method, and / or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (6)

A method for providing a related law performed by an apparatus for providing related laws,
Searching for judgments from the search server for the input keyword;
Extracting reference sentence data included in the judgments;
Parsing the reference sentence data based on a separator;
Vectorizing parsed reference sentence data to generate word vectors for each of the legal terms included in the reference sentence data;
Calculating similarities between word vectors corresponding to the search target legal terms received from the search server and word vectors other than the word vectors corresponding to the search target legal terms; And
Providing the search server with a legal provision corresponding to the word vector from which the highest similarity value is derived,
The parsing step further includes a pre-processing step of unifying units of legal provisions included in the reference sentence data into predetermined units,
The similarity (

) Is cosine similarity defined by Equation 1,
Equation 1 is

ego,
The a is a word vector corresponding to the search term law clause,
The b is any one of the word vectors other than the word vector corresponding to the search target legal clause among the word vectors,
How to provide related laws.
According to claim 1,
The vectorizing step is a method of providing an association method for vectorizing the parsed reference sentence data using a word-to-vec (word2vec) algorithm.
According to claim 2,
The word-to-vec algorithm is a method of providing an association method based on a skip-gram model.
delete delete delete

Images