KR102445497B1 - Apparatus for generating lexical pattern and training sentence and operating method thereof - Google Patents

Abstract

A learning example sentence and a pattern generating apparatus are disclosed. In one embodiment, the input sentence is divided into a plurality of words by performing morphological analysis, a target word corresponding to an entity name is recognized among the separated words, and the recognized target word is Calculate the weight of the connected word based on one or more set values, calculate the weight of the recognized target word based on the weight of the connected word and the one or more set values, and combine the calculated weights with the input sentence to create one or more lexical patterns.

Description

Learning example sentence and vocabulary pattern generating device and its operation method {APPARATUS FOR GENERATING LEXICAL PATTERN AND TRAINING SENTENCE AND OPERATING METHOD THEREOF}

The following embodiments relate to learning example sentences and generating vocabulary patterns.

Learning example sentences may be used to accurately determine the intention of variously used sentences, such as a chatbot service.

As a related prior art, there is Korean Patent Publication No. 10- 1757222 (Title of the Invention: Method of Generating Paraphrased Sentences for Korean Sentences, Applicant: Korea Advanced Institute of Science and Technology). According to the registered patent publication, real morphemes are classified in Korean sentences, root words classified as real morphemes in Korean sentences are replaced with synonyms based on linguistic qualities, formal morphemes are classified in Korean sentences, and Korean sentences with synonyms are substituted. The content of grammatically transforming at least one of the pronoun and ending vocabularies classified as formal morphemes in

According to one aspect, a method of operating an apparatus for generating a learning example sentence and a pattern includes performing morphological analysis on an input sentence to separate the input sentence into a plurality of words; recognizing a target word corresponding to an entity name from among the separated words; calculating a weight of the connected word based on the recognized target word based on one or more set values, and calculating a weight of the recognized target word based on the weight of the connected word and the at least one set value; and generating one or more vocabulary patterns using the calculated weights and the input sentence.

The calculating may include: when there are a plurality of connected words, determining the number from a first word among the plurality of connected words to the recognized target word; sequentially calculating a weight of each of the plurality of connected words based on the one or more set values and the identified number; and calculating the weight of the recognized target word using the weight of the last word among the plurality of connected words, the one or more set values, and the identified number.

Calculating the weight of the recognized target word by using the weight of the last word, the one or more set values, and the identified number includes: performing multiplication by the first set value and the weight of the last word; performing division on the result of the multiplication and the determined number; and determining a result of adding the second set value to the result of the division as a weight of the recognized target word.

The second set value may be greater than the first set value.

The generating of the vocabulary pattern may include: generating a first sentence by substituting the recognized target word in the input sentence with the entity name, and generating a first vocabulary pattern corresponding to the first sentence; and generating a second sentence by deleting a connection word having a minimum weight among the weights of each of the plurality of connection words from the generated first sentence, and generating a second vocabulary pattern corresponding to the generated second sentence. may include.

In the generating of the vocabulary pattern, a connecting word having the second smallest weight among the weights of each of the plurality of connecting words is deleted from the generated second sentence to generate a third sentence, and corresponding to the generated third sentence. and generating a third lexical pattern to be used.

The generating of the vocabulary pattern may include generating the vocabulary pattern as a regular expression.

The method may further include generating one or more learning example sentences based on the generated vocabulary pattern, meta information about the entity name, and an ending dictionary.

The meta information may include a plurality of words related to the entity name.

The generating of the one or more learning example sentences may include replacing the entity name in the generated vocabulary pattern with one or more words in the meta information and converting the ending of the generated vocabulary pattern using the ending dictionary. have.

As software stored in a computer-readable recording medium, the operating method can be executed.

A learning example sentence and pattern generating apparatus according to one side performs morphological analysis on the input sentence to separate the input sentence into a plurality of words, recognizes a target word corresponding to an entity name from among the separated words, and the recognition Calculate the weight of the connected word based on the target word based on one or more set values, calculate the weight of the recognized target word based on the weight of the connected word and the at least one set value, and a controller for generating one or more lexical patterns by using the weights and the input sentence.

When there are a plurality of connection words, the controller determines the number from the first word among the plurality of connection words to the recognized target word, and based on the at least one set value and the identified number, the plurality of connection words The weight of each word may be sequentially calculated, and the weight of the recognized target word may be calculated using the weight of the last word among the plurality of connected words, the one or more set values, and the identified number.

The controller multiplies the first set value and the weight of the last word, divides the result of the multiplication and the identified number, and recognizes the result of adding the second set value to the result of the division It can be determined by the weight of the target word.

The second set value may be greater than the first set value.

The controller generates a first sentence by substituting the recognized target word in the input sentence with the entity name, generates a first lexical pattern corresponding to the first sentence, and generates a minimum weight of each of the plurality of connected words. A second sentence may be generated by deleting a connection word having a weight from the generated first sentence, and a second vocabulary pattern corresponding to the generated second sentence may be generated.

The controller generates a third sentence by deleting a connection word having a second smallest weight among the weights of each of the plurality of connection words from the generated second sentence, and a third vocabulary pattern corresponding to the generated third sentence. can create

The controller may generate the vocabulary pattern as a regular expression.

The controller may generate one or more learning example sentences based on the generated vocabulary pattern, meta information about the entity name, and an ending dictionary.

The controller may replace the entity name in the generated vocabulary pattern with one or more words in the meta information and convert the ending of the generated vocabulary pattern using the ending dictionary.

Embodiments can easily and efficiently generate example sentences (or learning data) necessary to train various deep learning models.

1 to 2 are diagrams for explaining a chatbot system according to an embodiment.
3 to 6 are diagrams for explaining a learning example sentence and pattern generation according to an embodiment.
7 is a block diagram illustrating a learning example sentence and a pattern generating apparatus according to an embodiment.
8 is a flowchart for explaining a learning example sentence and an operating method of an apparatus for generating a pattern according to an embodiment.

Specific structural or functional descriptions of the embodiments are disclosed for purposes of illustration only, and may be changed and implemented in various forms. Accordingly, the actual implementation form is not limited to the specific embodiments disclosed, and the scope of the present specification includes changes, equivalents, or substitutes included in the technical spirit described in the embodiments.

Although terms such as first or second may be used to describe various elements, these terms should be interpreted only for the purpose of distinguishing one element from another. For example, a first component may be termed a second component, and similarly, a second component may also be termed a first component.

When a component is referred to as being “connected” to another component, it may be directly connected or connected to the other component, but it should be understood that another component may exist in between.

The singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "have" are intended to designate that the described feature, number, step, operation, component, part, or combination thereof exists, and includes one or more other features or numbers, It should be understood that the possibility of the presence or addition of steps, operations, components, parts or combinations thereof is not precluded in advance.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present specification. does not

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. In the description with reference to the accompanying drawings, the same components are assigned the same reference numerals regardless of the reference numerals, and overlapping descriptions thereof will be omitted.

1 to 2 are diagrams for explaining a chatbot system according to an embodiment.

Referring to FIG. 1 , the chatbot system 110 may receive a natural language question from a user, and may feed back an answer to the natural language question to the user. As an example, when the chatbot system 110 receives a natural language question of “How about Hallasan Mountain?” from the user, an intention analysis section (Natural Language Understand; NLU), a dialogue management section (DM) in the chatbot system 110 . ), and the response generation section (Natural Generation), it is possible to feedback to the user the answer of "It is in the center of Jeju Island, and it has a beautiful view."

The intention analysis part of the chatbot system 110 can be driven on a pattern-based and learning-based basis. As in the example shown in FIG. 2 , a new input sentence is set as a threshold ( threshold), and if the new input sentence is greater than or equal to the threshold, intention extraction and prediction results may be provided to the user.

According to an exemplary embodiment, weights of words connected with respect to the entity word in the input sentence may be calculated, and vocabulary patterns related to similar sentences of the input sentence may be generated by using the calculated weights. Accordingly, one embodiment can be actively used in various fields such as conversational systems, machine translation, and text analysis that require various learning examples as well as chatbot services, and a large amount of learning examples required to learn various deep learning models can create This will be described in detail.

3 to 6 are diagrams for explaining a learning example sentence and pattern generation according to an embodiment.

Referring to FIG. 3 , the learning example sentence and pattern generating device 310 may receive the sentence “I miss the beautiful Mt. Halla” as an input.

The learning example sentence and pattern generating apparatus 310 may perform morphological analysis and part-of-speech tagging on the input sentence "I miss the beautiful Hallasan too". In other words, the learning example sentence and pattern generating apparatus 310 may separate the input sentence into morpheme units and determine the part-of-speech. The learning example sentence and pattern generating device 310 can generate "too beautiful Hallasan + I like this + like" the input sentence "I miss Hallasan so beautiful" through morpheme unit separation, and "too/MAG" through part-of-speech tagging , "Beautiful/MM", "Hallasan/NNP", etc. As another example, the learning example sentence and pattern generating device 310 performs morphological analysis and part-of-speech tagging on the input sentence "I went to Jeongdongjin yesterday", so that "yesterday/MAG", "Jeongdongjin/NNP", "e/JKB", A/VV, etc. can be created. This part-of-speech tagging may be performed based on the Sejong part-of-speech tag set.

The learning example sentence and pattern generating apparatus 310 may recognize a target word corresponding to an entity name from among words generated by separating input sentences. In other words, the learning example sentence and pattern generating apparatus 310 may recognize a word having a main meaning. As an example, among the words "too", "beautiful", and "Mt. Halla" in the input sentence "I miss Hallasan so beautiful", "Mt. Halla" corresponds to the entity name location, so the learning example sentence and pattern generating device 310 is the entity name. Through recognition, "Mt. Halla" can be recognized as a target word. In an embodiment, when the learning example sentence and pattern generating apparatus 310 recognizes the entity name "Hallasan", it may identify a connection word connected to "Hallasan" from among previous words of "Hallasan". As an example, the learning example sentence and pattern generating device 310 uses the part-of-speech of “beautiful”, one of the previous words of “Mt. Halla”, as an adjective, and “beautiful” modifies “Mt. Halla” through the part-of-speech of “beautiful”. can recognize Accordingly, the learning example sentence and pattern generating apparatus 310 may identify "beautiful" as a connection word connected to "Mt. Halla". The part-of-speech of "too", one of the previous words of "Hallasan", is a general adverb, and "too" can modify "beautiful" and does not directly modify "Hallasan". In an embodiment, the learning example sentence and pattern generating apparatus 310 may identify "too" as a connecting word (indirectly) connected to "Hallasan". Accordingly, the learning example sentence and pattern generating apparatus 310 may identify each of "too" and "beautiful" in the input sentence "I miss Hallasan, so beautiful" as a connecting word of "Mt. Halla".

The learning example sentence and pattern generating apparatus 310 may calculate the similarity between words based on the entity name. More specifically, when the learning example sentence and pattern generating apparatus 310 recognizes "Mt. Halla" as a target word (or entity name), the weight of "Mt. Halla", the weight of "Beautiful" connected to "Mt. Halla", "Beautiful" The weight of "too" connected to can be calculated through Equation 1 below.

는 단어(w _i )의 이전 인접한 단어들의 가중치의 합으로, 단어(w _i )의 가중치를 나타낼 수 있다. 달리 표현하면,

는 단어(w _i )의 연결 단어들 각각의 가중치의 합을 나타낼 수 있다.

는 개체명(e _i )까지의 인접한 단어들의 집합을 나타낼 수 있다.

는 단어(wj)로부터 개체명(e _i )까지의 모든 단어의 개수를 나타낼 수 있다. 달리 표현하면,

는 개체명(e _i )의 연결 단어들 첫번째 단어부터 개체명(e _i )까지의 단어의 개수를 나타낼 수 있다.

는 단어(w _j )의 가중치와 단어(w _j )의 이전 인접한 단어의 가중치의 합을 나타낼 수 있다. In Equation 1 above,

is the sum of the weights of previous adjacent words of the word w _i , and may represent the weight of the word w _i . In other words,

may represent the sum of weights of each of the connected words of the word w _i .

may represent a set of adjacent words up to the entity name ( e _i ).

may represent the number of all words from the word ( wj ) to the entity name ( e _i ). In other words,

may represent the number of words from the first word of the connection words of the entity name ( e _i ) to the entity name ( e _i ).

may represent the sum of the weight of the word w _j and the weight of the previous adjacent word of the word w _j .

In Equation 1 above, 1 may correspond to the set value of the word w _i and 0.8 may represent the set value of the adjacent word. In Equation 1 above, another constant may be utilized or replaced in place of 1, and another constant may be utilized or replaced in place of 0.8.

={너무, 아름다운}일 수 있다. The learning example sentence and pattern generating apparatus 310 may generate {too, beautiful} a set of connected words of the entity name e _i from the input sentence. In other words,

= can be {too, beautiful}.

)를 파악할 수 있다. 학습 예문 및 패턴 생성 장치(310)는 개체명 "한라산"의 연결 단어들 중 첫번째 단어인 "너무"로부터 "한라산"까지 총 3개의 단어가 있으므로,

를 3으로 파악할 수 있다. The learning example sentence and pattern generating apparatus 310 is the number of words from the first word to the entity name ( e _i ) among the words connected to the entity name ( e _i ) (ie, the number of words (ie, e i )

) can be found. The learning example sentence and pattern generating device 310 has a total of three words from "too" to "Hallasan", which is the first word among the connected words of the entity name "Hallasan",

can be identified as 3.

Referring to the example shown in FIG. 4 , calculation of the weight of “too” (α ₁ ), the weight of “beautiful” (α ₂ ), and the weight of “Mt. Halla” (α ₃ ) will be described.

Referring to FIG. 4 , “the learning example sentence and pattern generating device 310 may calculate the weight α ₁ of “too” as 1 because there is no word connected before “too”. In other words, “the example learning sentence and The pattern generating apparatus 310 may calculate the weight α ₁ of the first connected word “too” among the connected words of “Hallasan” as 1.

The learning example sentence and pattern generating apparatus 310 may calculate the weight α ₂ of “beautiful” as 1.266. In other words, "the learning example sentence and pattern generating device 310 may calculate the weight (α _{2 ) of the last connected word "beautiful" among the connected words of "Mt. Halla" as 1.266. The weight (α 2 )} of "beautiful" can be calculated based on the weight of “too” connected with “beautiful.” The weight of “beautiful” (α ₂ ) can be expressed differently as the weight of connection between “beautiful” and “too” (α ₂ ) .

The learning example sentence and pattern generating apparatus 310 may calculate the weight α ₃ of “Mt. Halla” as 1.337. The weight α ₃ of “Mt. Halla” may be calculated based on the weight of each of the connecting words “too” and “beautiful” based on “Mt. Halla”. The weight (α ₃ ) of “Mt. Halla” may be expressed differently as the weight of connection (α ₃ ) between “Mt. Halla” and the connected words. As described above, since "Mt. Halla" corresponds to the name of an entity, the learning example sentence and pattern generating apparatus 310 may finish calculating the weight. In other words, the learning example sentence and pattern generating apparatus 310 may end the weight calculation when the weight of the entity name is calculated.

The learning example sentence and pattern generating apparatus 310 may generate one or more vocabulary patterns by using the calculated weights and the input sentence. Referring to the example shown in FIG. 5 , the learning example sentence and pattern generating device 310 creates a first sentence 510 by replacing “Mt. Halla” with “name of an entity” in the input sentence “I like Mt. Halla, so beautiful.” can do. As shown in the example shown in FIG. 5 , the learning example sentence and pattern generating device 310 generates the first sentence “I like the beautiful %location” 510 by replacing “Mt. Halla” in the input sentence with “%location”. can do. The learning example sentence and pattern generating apparatus 310 may generate the first vocabulary pattern 511 by marking the regular expression symbol (^) indicating the start of the character string in the first sentence 510 . As in the example shown in FIG. 5 , the learning example sentence and pattern generating apparatus 310 may generate a first vocabulary pattern "^too.beautiful.%location.?good" 511 .

The learning example sentence and pattern generating apparatus 310 may generate a similar vocabulary pattern through a recursive method. As in the example shown in FIG. 5 , the learning example sentence and pattern generating apparatus 310 may generate the second sentence 520 by removing “too” having the minimum weight from the first sentence 510, and a regular expression The second vocabulary pattern 521 may be generated by marking a symbol (?) indicating that there is no or one word immediately in front of the . As shown in the example shown in FIG. 5 , the learning example sentence and pattern generating device 310 may generate the second sentence “I like the beautiful %location” 520, and the second vocabulary pattern “.? beautiful. %location. ?good$" (521). The learning example sentence and pattern generating apparatus 310 may generate a third sentence 530 by removing "beautiful" having a second minimum weight from the second sentence 520, and there is no word immediately preceding in the regular expression or one or more A third vocabulary pattern 531 may be generated by marking a symbol (*) indicating repetition. As shown in the example shown in FIG. 5 , the learning example sentence and pattern generating device 310 may generate the third sentence “I like %location” 530, and the third vocabulary pattern “.*%location.? "(531) can be created.

In the above, the learning example sentence and pattern generating apparatus 310 has been described for generating the first sentence 510 to the third sentence 530, but this is only an example, and the learning example sentence and pattern generating apparatus 310 is At least one of the first sentences 510 to 530 may be transformed to generate a fourth sentence, a fifth sentence, etc., and each lexical pattern such as the generated fourth sentence and the fifth sentence may be generated. can

The learning example sentence and pattern generating apparatus 310 may finish generating the vocabulary pattern since all the words connected to the entity name have been removed from the input sentence.

The learning example sentence and pattern generating apparatus 310 may generate various learning example sentences by using meta information (or meta dictionary) related to the entity name and the ending dictionary. As shown in the example shown in FIG. 6 , the learning example sentence and pattern generating apparatus 310 may generate various learning example sentences by substituting "%location" for one or more of several words in the meta information 610 . Also, the learning example sentence and pattern generating apparatus 310 may generate learning example sentences of various verbs by converting the endings of the verbs using the ending dictionary 620 .

7 is a block diagram illustrating a learning example sentence and a pattern generating apparatus according to an embodiment.

Referring to FIG. 7 , the learning example sentence and pattern generating apparatus 310 includes a memory 710 and a controller 720 .

The memory 710 may store data necessary for an operation of the controller 720 , an operation result of the controller 720 , and the like.

The controller 720 controls the overall operation of the learning example sentence and pattern generating device 310 .

The controller 720 may divide the input sentence into a plurality of words by performing morphological analysis on the input sentence. As an example, the controller 720 may perform a shape analysis on the input sentence "I miss the beautiful Hallasan so much" and divide it into "too", "beautiful", "Mt. Halla", and the like.

The controller 720 may recognize a target word corresponding to the entity name from among the separated words. As an example, the controller 720 may recognize "Mt. Halla" as a target word by performing entity name recognition.

The controller 720 may calculate the weight of the connected word based on the recognized target word based on one or more set values, and may calculate the weight of the target word based on the weight of the corresponding connected word and one or more set values. .

More specifically, when there are a plurality of connected words based on the recognized target word, the controller 720 may determine the number from the first word to the target word among the plurality of connected words, A weight of each of the plurality of connected words may be sequentially calculated based on the number. As an example, there are "too" and "beautiful", which are connecting words based on the target word "Mt. Halla" in "too", "beautiful", and "mountain of halla". The controller 720 may determine that there are a total of three words from the first connection word "too" to the target word "Hallasan". Since there is no word connected before "too", the controller 720 may calculate the weight of "too" as the second set value (eg, 1 as described above). The controller 720 multiplies the weight of “beautiful” by the first set value (eg, 0.8) and the weight “1” of “too”, and divides the result of the multiplication and the identified number “3” may be performed, and the sum of the result of division and the second set value may be determined as a weight value of “beautiful”. In other words, the controller 720 may determine 1+(0.8×the weight of “too”)/3 as the weight of “beautiful”. Since the calculation process of the weight of “beautiful” is shown in FIG. 4 , a detailed description thereof will be omitted. Although the first set value is 0.8 and the second set value is 1 as an example, this is only an example, and each of the first set value and the second set value is not limited to the above example.

The controller 720 may calculate the weight of the target word by using the weight of the last word among the plurality of connected words, one or more set values, and the identified number. As an example, the controller 720 multiplies the weight and the first set value of the last word “beautiful” among the plurality of connected words, and divides the result of the multiplication and the identified number “3”, and the result of the division and the second set value may be determined as the weight of the target word “Hallasan”. In other words, the controller 720 may determine 1+(0.8×the weight of “beautiful”)/3 as the weight of “Mt. Halla”. Since the calculation process of the weight of "Mt. Halla" is shown in FIG. 4, a detailed description thereof will be omitted.

The controller 720 may generate one or more vocabulary patterns using the calculated weights and the input sentence. In this case, the controller 720 may generate the vocabulary pattern as a regular expression. A lexical pattern may be expressed differently as a sentence pattern. As described above, the controller 720 may generate the first sentence 510 by substituting the target word in the input sentence with the entity name and generate the first vocabulary pattern 511 corresponding to the first sentence 510 . can do. The controller 720 may generate the second sentence 520 by deleting the connection word having the minimum weight among the weights of each of the plurality of connection words of the target word from the first sentence 510 , and the second sentence 520 . A second vocabulary pattern 521 corresponding to can be generated. The controller 720 may generate the third sentence 530 by deleting the connection word having the second smallest weight among the weights of each of the plurality of connection words of the target word from the second sentence 520, and the third sentence ( A third vocabulary pattern 531 corresponding to 530 may be generated.

The controller 720 may generate one or more learning example sentences based on one or more vocabulary patterns, meta information about the entity name, and the ending dictionary. More specifically, the controller 720 may replace the entity name in the vocabulary pattern with one or more words in the meta information and convert the ending of the vocabulary pattern using the ending dictionary. Accordingly, the controller 720 may generate various learning example sentences.

Since the matters described with reference to FIGS. 1 to 6 may be applied to the matters described with reference to FIG. 7 , a detailed description thereof will be omitted.

8 is a flowchart for explaining a learning example sentence and an operating method of an apparatus for generating a pattern according to an embodiment.

Referring to FIG. 8 , the learning example sentence and pattern generating apparatus 310 separates the input sentence into a plurality of words by performing morphological analysis on the input sentence ( 810 ).

The learning example sentence and pattern generating apparatus 310 recognizes a target word corresponding to an entity name from among the separated words ( 820 ).

The learning example sentence and pattern generating apparatus 310 calculates the weight of the connected word based on the target word based on one or more set values, and calculates the weight of the target word based on the weight of the connected word and one or more set values. (830).

In operation 830 , when there are a plurality of connected words based on the target word, the learning example sentence and pattern generating apparatus 310 may determine the number from the first word to the target word among the plurality of connected words. The learning example sentence and pattern generating apparatus 310 may sequentially calculate the weight of each of the plurality of connected words based on one or more set values and the identified number. The learning example sentence and pattern generating apparatus 310 may calculate the weight of the target word by using the weight of the last word among the plurality of connected words, one or more set values, and the identified number.

In step 830, when there are a singular number of connected words based on the target word, the learning example sentence and pattern generating apparatus 310 may determine the total number of connected words and target words as 2, and set the weight of the connected word as " 1" can be determined. The learning example sentence and pattern generating apparatus 310 may calculate the weight of the target word by using the weight of the connected word, one or more set values, and the identified number. As an example, the learning example sentence and pattern generating apparatus 310 may calculate 1+(0.8×the weight of the connected word)/3 as the weight of the target word.

The learning example sentence and pattern generating apparatus 310 generates one or more vocabulary patterns by using the calculated weights and the input sentence ( 840 ).

Since the matters described with reference to FIGS. 1 to 7 may be applied to the matters described with reference to FIG. 8 , a detailed description thereof will be omitted.

The embodiments described above may be implemented by a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the apparatus, methods and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate (FPGA). array), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions, may be implemented using a general purpose computer or special purpose computer. The processing device may execute an operating system (OS) and a software application running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes 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 one processor and one controller. Other processing configurations are also possible, such as parallel processors.

The software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or device, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored in a computer-readable recording medium.

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 in a computer-readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination, and the program instructions recorded on the medium are specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. may be Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

The hardware devices described above may be configured to operate as one or a plurality of software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with reference to the limited drawings, a person skilled in the art may apply various technical modifications and variations based thereon. For example, the described techniques are performed in a different order than the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

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

Claims (19)

In the operating method of the learning example sentence and the pattern generating apparatus,
separating the input sentence into a plurality of words by performing morphological analysis on the input sentence;
recognizing a target word corresponding to an entity name from among the separated words;
calculating a weight of the connected word based on the recognized target word based on one or more set values, and calculating a weight of the recognized target word based on the weight of the connected word and the at least one set value; and
generating one or more lexical patterns using the calculated weights and the input sentence
containing,
Learning example sentences and the operation method of the pattern generating device.
According to claim 1,
The calculating step is
determining the number from a first word among the plurality of connected words to the recognized target word when there are a plurality of connected words;
sequentially calculating a weight of each of the plurality of connected words based on the one or more set values and the identified number; and
calculating the weight of the recognized target word using the weight of the last word among the plurality of connected words, the one or more set values, and the identified number;
containing,
Learning example sentences and the operation method of the pattern generating device.
3. The method of claim 2,
Calculating the weight of the recognized target word using the weight of the last word, the one or more set values, and the identified number includes:
performing multiplication on the first set value and the weight of the last word;
performing division on the result of the multiplication and the determined number; and
determining a result of adding a second set value to the result of the division as a weight of the recognized target word
containing,
Learning example sentences and the operation method of the pattern generating device.
4. The method of claim 3,
the second set value is greater than the first set value,
Learning example sentences and the operation method of the pattern generating device.
3. The method of claim 2,
The step of generating the vocabulary pattern comprises:
generating a first sentence by substituting the recognized target word in the input sentence with the entity name and generating a first vocabulary pattern corresponding to the first sentence; and
generating a second sentence by deleting a connecting word having a minimum weight among the weights of each of the plurality of connecting words from the generated first sentence, and generating a second vocabulary pattern corresponding to the generated second sentence;
containing,
Learning example sentences and the operation method of the pattern generating device.
6. The method of claim 5,
The step of generating the vocabulary pattern comprises:
A third sentence is generated by deleting a connection word having a second smallest weight among the weights of each of the plurality of connection words from the generated second sentence, and a third vocabulary pattern corresponding to the generated third sentence is generated. step
containing,
Learning example sentences and the operation method of the pattern generating device.
According to claim 1,
The step of generating the vocabulary pattern comprises:
Generating the vocabulary pattern as a regular expression (regular expression)
containing,
Learning example sentences and the operation method of the pattern generating device.
According to claim 1,
Generating one or more learning example sentences based on the generated vocabulary pattern, meta information about the entity name, and a suffix dictionary
further comprising,
Learning example sentences and the operation method of the pattern generating device.
9. The method of claim 8,
The meta information includes a plurality of words about the entity name,
The step of generating the one or more learning example sentences,
replacing the entity name in the generated vocabulary pattern with one or more words in the meta information and converting the ending of the generated vocabulary pattern using the ending dictionary
containing,
Learning example sentences and the operation method of the pattern generating device.
As software stored in a computer-readable recording medium, for executing any one of claims 1 to 9, the software stored in the computer-readable recording medium.
In the learning example sentence and pattern generating apparatus,
Separating the input sentence into a plurality of words by performing morphological analysis on the input sentence, recognizing a target word corresponding to an entity name from among the separated words, and weighting the connected word based on the recognized target word is calculated based on one or more set values, a weight of the connected word and a weight of the recognized target word are calculated based on the one or more set values, and one or more weights are calculated using the calculated weights and the input sentence. A controller that generates a lexical pattern
containing,
Learning example sentences and pattern generating device.
12. The method of claim 11,
The controller is
When there are a plurality of connection words, the number from the first word of the plurality of connection words to the recognized target word is determined, and based on the one or more setting values and the number of the connection words, each of the plurality of connection words is determined. calculating weights sequentially, calculating the weight of the recognized target word using the weight of the last word among the plurality of connected words, the one or more set values, and the identified number,
Learning example sentences and pattern generating device.
13. The method of claim 12,
The controller is
The first set value and the weight of the last word are multiplied, the result of the multiplication and the determined number are divided, and the result of adding the second set value to the result of the division of the recognized target word determined by weight,
Learning example sentences and pattern generating device.
14. The method of claim 13,
the second set value is greater than the first set value,
Learning example sentences and pattern generating device.
13. The method of claim 12,
The controller is
A first sentence is generated by substituting the recognized target word in the input sentence with the entity name, a first vocabulary pattern corresponding to the first sentence is generated, and a minimum weight among the weights of each of the plurality of connected words is generated. generating a second sentence by deleting a connection word from the generated first sentence, and generating a second vocabulary pattern corresponding to the generated second sentence;
Learning example sentences and pattern generating device.
16. The method of claim 15,
The controller is
A third sentence is generated by deleting a connection word having a second smallest weight among the weights of each of the plurality of connection words from the generated second sentence, and a third vocabulary pattern corresponding to the generated third sentence is generated. ,
Learning example sentences and pattern generating device.
12. The method of claim 11,
The controller is
Generating the lexical pattern as a regular expression (regular expression),
Learning example sentences and pattern generating device.
12. The method of claim 11,
The controller is
generating one or more learning example sentences based on the generated vocabulary pattern, meta information about the entity name, and a suffix dictionary,
Learning example sentences and pattern generating device.
19. The method of claim 18,
The meta information includes a plurality of words about the entity name,
The controller is
replacing the entity name in the generated vocabulary pattern with one or more words in the meta information and converting the ending of the generated vocabulary pattern using the ending dictionary,
Learning example sentences and pattern generating device.

Images