KR102642012B1 - Electronic apparatus for performing pre-processing regarding analysis of text constituting electronic medical record - Google Patents

Abstract

An electronic device is disclosed. The electronic device includes a memory storing a word dictionary indicating a matching relationship between words of a first language and words of a second language, and at least one deep learning model for processing text of the second language, and a processor connected to the memory. Includes. The processor identifies, within the entire text, a first text corresponding to the first language and a second text corresponding to the second language, and, based on the word dictionary, converts the identified first text into a third text corresponding to the second language. Convert it to text, process the integrated text including the second text and the third text into natural language to obtain one or more vectors, and train a deep learning model based on the obtained vectors.

Description

An electronic device that performs preprocessing related to the analysis of text constituting an electronic medical record { ELECTRONIC APPARATUS FOR PERFORMING PRE-PROCESSING REGARDING ANALYSIS OF TEXT CONSTITUTING ELECTRONIC MEDICAL RECORD }

The present disclosure relates to an electronic device that performs preprocessing for text analysis. More specifically, the present disclosure relates to an electronic device that performs translation for a specific language based on a word dictionary specialized in a specialized field, effectively translating the entire text in which multiple languages are mixed. It is about electronic devices that can process

When using electronic medical records of patients with various diseases, such as cerebral infarction, it was not efficient to replace accurate medical terms into English using general machine translation.

In particular, words that are frequently used in general conversations or articles are quite far from medical terminology, and when using this general text preprocessing method, the meaning of the text due to unclear conversion of Korean to English cannot be mechanically and accurately recognized. does not exist.

Compared to artificial languages, in the case of natural languages, especially Hangul, the same words are used with different meanings in fields such as medicine and art, so when mechanical translation is applied, conversion to incorrect vectors occurs, making it difficult to implement efficient text data. It is impossible.

Figure 1A is a schematic diagram for explaining the process of conventional English natural language processing, and Figure 1B is a schematic diagram for explaining the conventional Korean natural language processing process.

Referring to Figure 1a, English text preprocessed through Tokenization, Stemming, Stop Word Removal, Part-Of-Speech Tagging, etc. can be used as an English text vector.

And, referring to Figure 1b, in the case of Korean text, it can be used as a vector through processes such as Entity Detection and Relation Detection in addition to Tokenization and POS Tagging.

However, in cases where Korean and English are mixed, Korean and English tokens must be produced separately. In this case, the order of Korean and English is mixed, so accurate interpretation of the context cannot be expected.

Registered Patent Publication No. 10-233144 (Text recognition system and method using neural network)

In the present disclosure, when preprocessing text containing multiple languages (e.g. Korean, English, etc.), all of them are converted into a specific language using a word dictionary specialized in the specialized field, and then a deep learning model is built using the integrated text. Provides an electronic device and control method.

The objects of the present disclosure are not limited to the purposes mentioned above, and other objects and advantages of the present disclosure that are not mentioned can be understood by the following description and will be more clearly understood by the examples of the present disclosure. Additionally, it will be readily apparent that the objects and advantages of the present disclosure can be realized by the means and combinations thereof indicated in the patent claims.

An electronic device according to an embodiment of the present disclosure stores a word dictionary indicating a matching relationship between words of a first language and words of a second language, and at least one deep learning model for processing text of the second language. It includes a memory and a processor connected to the memory. The processor identifies a first text corresponding to the first language and a second text corresponding to the second language within the entire text, and, based on the word dictionary, converts the identified first text into the second text. Convert to a third text corresponding to the language, process the integrated text including the second text and the third text into natural language to obtain one or more vectors, and train the deep learning model based on the obtained vectors. .

The processor identifies, within the first text, at least one first keyword included in the word dictionary, and, based on the word dictionary, converts the first keyword included in the first text into the second language. and, with the second keyword set as one entity, the first text can be converted into the third text corresponding to the second language.

In this case, the processor compares the accuracy before and after training of the deep learning model trained based on the obtained vector, and when the accuracy of the deep learning model is lowered as a result of comparing the accuracy before and after training, the first keyword The word dictionary may be updated so that keywords of the second language that match are changed.

In addition, when a plurality of second keywords matching the first keyword are identified according to the word dictionary, the processor acquires a plurality of third texts by independently reflecting each of the plurality of second keywords, and If the difference between the outputs of the deep learning model as the vector matching each of the third texts is independently input is less than a threshold, the word dictionary is maintained, and the vector matching each of the plurality of third texts is independently input. If the difference between the outputs of the deep learning model is greater than or equal to the threshold, the word dictionary may be updated.

The processor may obtain one or more vectors by performing tokenization, part-of-speech tagging, and word embedding on the integrated text.

Meanwhile, the memory may include a first word dictionary for a first specialty field, a second word dictionary for a second specialty field, and a third word dictionary for a third specialty field. In this case, when the entire text is set to the first specialized field, the processor may convert the first text into a third text corresponding to the second language using the first word dictionary.

Here, when a keyword included in the first word dictionary does not exist in the first text, the processor uses at least one of the second word dictionary and the third word dictionary to convert the first text into the first word dictionary. 2 It can also be converted to a third text corresponding to the language.

In this case, the processor selects at least one word dictionary related to a keyword included in the first text among the second word dictionary and the third word dictionary, and uses the selected word dictionary to create the first text may be converted into a third text corresponding to the second language, and the first word dictionary may be updated based on the keyword of the second language that matches the keyword in the selected word dictionary.

According to an embodiment of the present disclosure, an electronic device including a word dictionary indicating a matching relationship between words of a first language and words of a second language, and at least one deep learning model for processing text of the second language. The control method includes identifying a first text corresponding to the first language and a second text corresponding to the second language within the entire text, and based on the word dictionary, dividing the identified first text into the second text. Converting into a third text corresponding to a second language, processing the integrated text including the second text and the third text into natural language to obtain one or more vectors, and forming the deep learning model based on the obtained vector. It includes training steps.

An electronic device according to an embodiment of the present disclosure includes a memory storing a word dictionary indicating a matching relationship between words of a first language and words of a second language, and a processor connected to the memory. The processor identifies a first text corresponding to the first language and a second text corresponding to the second language within the entire text, and, based on the word dictionary, converts the identified first text into the second text. One or more vectors can be obtained by converting the text into a third text corresponding to the language and processing the integrated text including the second text and the third text into natural language.

The electronic device and control method according to the present disclosure can unify the types of languages by utilizing a word dictionary optimized for the specialized field when handling the entire text in which multiple languages are mixed, and as a result, the types of languages can be unified, and as a result, the context is unified and precise. By performing natural processing, there is an effect of improving the training environment of deep learning models.

When used in the medical field, the electronic device according to the present disclosure creates a knowledge-based dictionary that can accurately convey the meaning of medical terms, applies it to the electronic medical record, and then translates it into Korean-English through an integrated pre-processing method through additional machine translation. The efficiency of natural language processing of mixed electronic medical records can be secured. In this case, in addition to the individual meanings of the words, the meaning and context of the sentence can be reflected as much as possible. The electronic device according to the present disclosure performs a preprocessing process that can be applied to various types of diseases, including brain diseases such as cerebral infarction.

Figure 1a is a schematic diagram for explaining the process of conventional English natural language processing;
Figure 1b is a schematic diagram for explaining the process of conventional Hangul natural language processing;
2 is a block diagram illustrating the configuration of an electronic device including a word dictionary according to an embodiment of the present disclosure;
3 is a flowchart for explaining the operation of an electronic device according to an embodiment of the present disclosure;
4 is a diagram illustrating an operation in which an electronic device performs Korean-English conversion using a word dictionary according to an embodiment of the present disclosure; and
FIG. 5 is an algorithm for explaining an operation in which an electronic device selectively utilizes a word dictionary classified by specialized field according to an embodiment of the present disclosure.

Before explaining the present disclosure in detail, the description method of the present specification and drawings will be explained.

First, the terms used in the specification and claims are general terms selected in consideration of their functions in various embodiments of the present disclosure. However, these terms may vary depending on the intention of technicians working in the relevant technical field, legal or technical interpretation, and the emergence of new technologies. Additionally, some terms are arbitrarily selected by the applicant. These terms may be interpreted as defined in this specification, and if there is no specific term definition, they may be interpreted based on the overall content of this specification and common technical knowledge in the relevant technical field.

In addition, the same reference numbers or symbols in each drawing attached to this specification indicate parts or components that perform substantially the same function. For convenience of explanation and understanding, the same reference numerals or symbols are used in different embodiments. That is, even if all components having the same reference number are shown in multiple drawings, the multiple drawings do not represent one embodiment.

Additionally, in this specification and claims, terms including ordinal numbers such as “first”, “second”, etc. may be used to distinguish between components. These ordinal numbers are used to distinguish identical or similar components from each other, and the meaning of the term should not be interpreted limitedly due to the use of these ordinal numbers. For example, the order of use or arrangement of components combined with these ordinal numbers should not be limited by the number. If necessary, each ordinal number may be used interchangeably.

In this specification, singular expressions include plural expressions, unless the context clearly dictates otherwise. In this application, terms such as “comprise” or “consist of” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are intended to indicate the presence of one or more other It should be understood that this does not exclude in advance the presence or addition of features, numbers, steps, operations, components, parts, or combinations thereof.

In embodiments of the present disclosure, terms such as “module”, “unit”, “part”, etc. are terms to refer to components that perform at least one function or operation, and these components are either hardware or software. It may be implemented or may be implemented through a combination of hardware and software. In addition, a plurality of "modules", "units", "parts", etc. are integrated into at least one module or chip, except in cases where each needs to be implemented with individual specific hardware, and is integrated into at least one processor. It can be implemented as:

Additionally, in an embodiment of the present disclosure, when a part is connected to another part, this includes not only direct connection but also indirect connection through other media. In addition, the meaning that a part includes a certain component does not mean that other components are excluded, but that it may further include other components, unless specifically stated to the contrary.

FIG. 2 is a block diagram illustrating the configuration of an electronic device including a word dictionary according to an embodiment of the present disclosure.

Referring to FIG. 2 , the electronic device 100 may include a memory 110 and a processor 120.

The electronic device 100 may correspond to a server and may correspond to various terminal devices such as a smartphone, smart speaker, desktop PC, laptop PC, and tablet PC.

Additionally, the electronic device 100 may be implemented as a system including one or more computers. For example, the electronic device 100 may correspond to a system that manages electronic medical records operated by at least one hospital, medical institution, or public institution, but is not limited to this.

The memory 110 is configured to store an operating system (OS) for controlling the overall operation of the components of the electronic device 100 and at least one instruction or data related to the components of the electronic device 100. .

The memory 110 may include non-volatile memory, such as ROM or flash memory, and may include volatile memory, such as DRAM. Additionally, the memory 110 may include a hard disk, solid state drive (SSD), etc.

Referring to FIG. 2, the memory 110 may include at least one word dictionary 111 and at least one deep learning model 112.

The word dictionary 111 may include information about matching relationships between words in different languages. For example, the word dictionary 111 may include words in a first language (eg, Korean) and words in a second language (eg, English) that match the corresponding words.

The word dictionary 111 may be provided for each specialized field. For example, the memory 110 may separately include a word dictionary containing words related to brain disease and a word dictionary containing words related to heart disease.

The deep learning model 112 may correspond to a network model for processing text in a specific language. For example, if the word dictionary 111 includes matching relationships between Korean words and English words, the deep learning model 112 may correspond to a network model for processing text composed of English.

The deep learning model 112 may be designed to process text for various purposes, such as text classification, text summary, text recognition/conversion, and text-related dialogue generation.

The deep learning model 112 may be learned through the electronic device 100 or a separate server/system through various learning algorithms.

Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, and the learning algorithm in the present disclosure is specified. Except, it is not limited to the examples described above.

The deep learning model 112 is a network model (neural network model) based on a neural network and may include a plurality of network nodes with weights. Multiple network nodes can form a connection relationship based on weights between nodes of different layers.

The processor 120 is configured to overall control the electronic device 100. Specifically, the processor 120 may perform operations according to various embodiments of the present disclosure by being connected to the memory 110 and executing at least one instruction stored in the memory 110.

The processor 120 may include a general-purpose processor such as a CPU, AP, or DSP (Digital Signal Processor), a graphics-specific processor such as a GPU, a VPU (Vision Processing Unit), or an artificial intelligence-specific processor such as an NPU. An artificial intelligence-specific processor may be designed with a hardware structure specialized for training or use of a specific artificial intelligence model.

Figure 3 is a flowchart for explaining the operation of an electronic device according to an embodiment of the present disclosure.

Referring to FIG. 3, the electronic device 100 may identify a first text corresponding to the first language and a second text corresponding to the second language within the entire text (S310).

For example, the entire text may be a mixed text of Korean and English, the first language may be Korean, and the second language may be English, but the text is not limited to this and various types of languages can be identified.

Additionally, the electronic device 100 may convert the first text into a third text corresponding to the second language based on the above-described word dictionary 111 (S320).

Specifically, the electronic device 100 may identify at least one first keyword included in the word dictionary 111 within the first text. And, based on the word dictionary 111, the electronic device 100 may convert the first keyword included in the first text into a second keyword corresponding to the second language.

At this time, with the second keyword set as one entity, the electronic device 100 may convert the first text into a third text corresponding to the second language.

In relation to this, FIG. 4 is a diagram illustrating an operation in which an electronic device performs Korean-English conversion using a word dictionary according to an embodiment of the present disclosure.

Referring to FIG. 4, the word dictionary 405 includes words related to brain diseases, and specifically includes matching relationships between Korean words and English words.

The list 410 includes texts corresponding to Korean, and the list 410' includes the results of converting the texts of the list 410 into English.

Referring to FIG. 4, each text included in the list 410 may be converted into each text included in the list 410' according to a matching relationship defined in the word dictionary 405.

For example, if the above-described first text includes the keyword “vital symptoms,” the electronic device 100 may convert “vital symptoms” into “vital signs” using the word dictionary 405.

Here, the electronic device 100 may convert the first text into English, with “vital sign” set as one entity in the first text in which “vital symptoms” is replaced with “vital sign.” In this way, if translation is performed with the “vital sign” itself secured according to the word dictionary 405 fixed as a single entity, the accuracy of translation may increase.

At this time, the electronic device 100 may utilize at least one artificial intelligence model to perform Korean-English translation in addition to the word dictionaries 111 and 405 described above (ex. conventional machine translation or statistical-based translation).

According to the above-described embodiment, when the first text of the first language is converted into the third text (: second language), the electronic device 100 converts the integrated text (: second language) including the second text and the third text. ) can be processed into natural language to obtain one or more vectors (S330).

Specifically, the electronic device 100 may perform tokenization and part-of-speech tagging (POS Tagging) on the integrated text and then perform word embedding to obtain one or more vectors.

Before vector conversion through word embedding, the electronic device 100 may perform Stop Word Removal, N-gram-based word selection, or combination judgment between words on the integrated text.

And, when the vector converted into the integrated text is obtained as described above, the electronic device 100 can train the deep learning model 112 based on the obtained vector (S340).

Compared to the case of obtaining vectors at once for the entire text in which the first language and the second language are mixed, the control method of an electronic device according to the present disclosure (FIG. 3) is a method for controlling words with expertise in word conversion. The advantage is that the possibility of incorrect natural language processing before vector conversion is reduced in that vectors are obtained after unification into one language using a dictionary (111). These advantages can lead to an increase in the training performance of deep learning models that are trained using text-converted vectors.

Meanwhile, in one embodiment, the electronic device 100 may verify the effect of training by comparing the accuracy before and after training of the deep learning model 112 trained through the vector of the above-described integrated text.

For example, the electronic device 100 may input vectors of various texts provided for verification of the deep learning model 112 into the deep learning model 112, and the accuracy according to the output of the deep learning model 112 can be identified.

If the accuracy of the deep learning model 112 is lowered as a result of training using the vector of the integrated text described above, the electronic device 100 may update the word dictionary 111.

Specifically, as a result of converting the first keyword (ex. vital sign) included in the first text corresponding to the first language into a second keyword (ex. vital sign) of the second language according to the word dictionary 111, It is assumed that an integrated text (: second language) is obtained.

Here, when the accuracy of the deep learning model 112 trained according to the vector of the integrated text is rather low, the electronic device 100 details the keyword of the second language that matches the first keyword in the word dictionary 111. You can change it to a keyword other than the second keyword.

At this time, the keyword matching the first keyword may be designated according to the user input of one or more experts.

For example, the electronic device 100 may transmit information about the first keyword to a plurality of expert terminals. Additionally, the electronic device 100 may receive a keyword converted from the first keyword into the second language from each expert terminal. At this time, the electronic device 100 may update the word dictionary 111 according to the keywords received from the largest number of expert terminals.

Meanwhile, according to the word dictionary 111, there may be a plurality of second keywords (: second language) matching the first keyword (: first language).

As an example, assume that in step S320 described above, a first keyword corresponding to the first language is identified, and a plurality of second keywords matching the first keyword are identified according to the word dictionary 111.

In this case, the electronic device 100 may acquire a plurality of third texts by independently reflecting each of the plurality of second keywords. Here, each of the plurality of third texts may be combined with the second text corresponding to the second language to obtain a plurality of integrated texts.

In this case, a plurality of integrated texts including different third texts may be separately converted into vectors, and the electronic device 100 may independently input the vectors into which each integrated text is converted into the deep learning model 112. You can.

Here, if the difference between the output of the deep learning model 112 for the vector into which each integrated text is converted is less than the threshold, the word dictionary 111 may remain the same as before. That is, within the word dictionary 111, a plurality of second keywords set to match the first keyword may be maintained as is.

On the other hand, if the difference between the output of the deep learning model 112 for the vector into which each integrated text is converted is greater than or equal to a threshold, the word dictionary 111 may be updated.

As a specific example, the electronic device 100 may select the integrated text that caused the output of the deep learning model 112 that is closest to the correct answer among a plurality of integrated texts previously derived according to different second keywords.

At this time, the electronic device 100 may identify a (low-quality) integrated text resulting from an output that differs by more than a threshold from the above-described output closest to the correct answer among the remaining integrated texts excluding the selected integrated text. Additionally, the electronic device 100 may delete the second keyword related to the identified (low-quality) integrated text from the keyword list that matches the first keyword in the word dictionary 111.

Meanwhile, in one embodiment, the memory 110 may include a plurality of word dictionaries divided by specialized field.

For example, memory 110 may include a first word dictionary for a first specialty, a second word dictionary for a second specialty, and a third word dictionary for a third specialty.

As a specific example, the first specialty may be brain disease, the second specialty may be heart disease, and the third specialty may be mental disease. However, it goes without saying that the specialty field may include various other medical specialties, and may also include specialties in fields completely different from the medical specialty field.

In this way, when word dictionaries corresponding to various specialized fields are pre-stored, the electronic device 100 preferentially utilizes the word dictionary appropriate for the specialized field to which the entire text (a mixture of the first language and the second language) belongs to create an integrated text. can be obtained. However, depending on the situation, word dictionaries in other specialized fields may also be used, and related embodiments will be described with reference to FIG. 5 below.

FIG. 5 is an algorithm for explaining an operation in which an electronic device selectively utilizes a word dictionary classified by specialized field according to an embodiment of the present disclosure.

Referring to FIG. 5, the electronic device 100 can set a specialized field for the entire text (: first language and second language mixed) (S510). At this time, the specialized field may be set according to user input, or may be preset according to metadata stored with the entire text.

Alternatively, the electronic device 100 may identify the specialty field by inputting the entire text described above into at least one classifier model trained to select/classify the specialty field.

And, the electronic device 100 can identify the first text in the first language included in the entire text (S520). This process may be included in the S310 process described above.

Here, the electronic device 100 can convert the first text into a third text in the second language, and in this process, a dictionary of words suitable for the previously set specialized field can be selected and used (S530).

For example, when the entire text is set to the first specialty, the electronic device 100 may select a first word dictionary suitable for the first specialty, and use the selected first word dictionary to convert the first text into the second specialty. It can be converted into a third text of the language.

Specifically, when a keyword from the first word dictionary exists in the first text (S540 - Y), the electronic device 100 converts the first text into the second language by converting the keyword using the first word dictionary. You can do it (S550).

On the other hand, the keyword of the selected first word dictionary may not exist in the first text (S540 - N).

In this case, the electronic device 100 may select at least one word dictionary corresponding to another specialized field (S560).

Then, the electronic device 100 may convert the first text into the second language based on the selected word dictionary (S550).

As an example, the electronic device 100 may select a word dictionary that includes at least one of keywords included in the first text. As a specific example, when a keyword included in the first text is included in a second word dictionary corresponding to a second specialized field, the electronic device 100 converts the first text into the second language by using the second word dictionary. Integrated text can be obtained.

Additionally, as an example, the electronic device 100 may select at least one specialized field with a high degree of relevance to the first specialized field set for the entire text and select a word dictionary of the selected specialized field.

The degree of correlation between specialized fields can be calculated based on the similarity between the contents of the word dictionary matching each specialized distribution.

Specifically, the electronic device 100 compares each of the keywords included in the first word dictionary of the first specialized field with each of the keywords included in the second word dictionary of the second specialized field, thereby averaging the similarity between keywords (or Other statistics) can be calculated.

Additionally, the electronic device 100 may set the degree of association between the first specialized field and the second specialized field according to the average similarity between keywords. For example, the greater the average similarity between keywords, the greater the correlation between the first and second specialized fields can be calculated.

As an example, assume that the degree of correlation between the first and second specialized fields is greater than the degree of correlation between the first and third specialized fields.

Here, as described above, if the keyword of the first text included in the entire text does not exist in the first word dictionary of the first specialty field selected previously, the electronic device 100 selects the keyword with the highest degree of association with the first specialty field. You can select a second dictionary of words in a second specialty.

At this time, the electronic device 100 may convert the first text into the second language using the selected second word dictionary.

However, there may be cases where the second word dictionary also does not include keywords in the first text. In this case, the electronic device 100 may select a third word dictionary in the third specialty field that has the second highest degree of correlation with the first specialty field. Additionally, the electronic device 100 may convert the first text into the second language using a third word dictionary.

On the other hand, as described above, if keyword conversion is not performed through a word dictionary (ex. first word dictionary) selected (for the first time) according to the specialty of the entire text, but is performed through a word dictionary of another specialty (ex. S540 - N -> S560 -> S550), the electronic device 100 first selects a word dictionary (ex. first word dictionary) based on the finally selected word dictionary (ex. second word dictionary and/or third word dictionary). word dictionary) can be updated.

Specifically, within the word dictionary (e.g., second word dictionary and/or third word dictionary) of another specialized field finally selected, the electronic device 100 may generate a second word matching the keyword included in the above-described first text. Can identify language keywords.

Additionally, the electronic device 100 may add a matching relationship between keywords included in the above-described first text and keywords of the second language (matched through another word dictionary) into the initially selected word dictionary.

As a result, the word dictionary (ex. first word dictionary) matching the specialty field (ex. first specialty field) of the entire text can be continuously supplemented.

Meanwhile, the various embodiments described above may be implemented by combining two or more embodiments as long as they do not conflict or contradict each other.

Meanwhile, the various embodiments described above may be implemented in a recording medium that can be read by a computer or similar device using software, hardware, or a combination thereof.

According to hardware implementation, embodiments described in this disclosure include application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), and field programmable gate arrays (FPGAs). ), processors, controllers, micro-controllers, microprocessors, and other electrical units for performing functions.

In some cases, embodiments described herein may be implemented in the processor itself. According to software implementation, embodiments such as procedures and functions described in this specification may be implemented as separate software modules. Each of the software modules described above may perform one or more functions and operations described herein.

Meanwhile, computer instructions or computer programs for performing processing operations in the electronic device 100 according to various embodiments of the present disclosure described above are non-transitory computer-readable medium. It can be saved in . Computer instructions or computer programs stored in such non-transitory computer-readable media, when executed by a processor of a specific device, cause the specific device to perform processing operations in the electronic device 100 according to the various embodiments described above.

A non-transitory computer-readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as registers, caches, and memories. Specific examples of non-transitory computer-readable media may include CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM, etc.

In the above, preferred embodiments of the present disclosure have been shown and described, but the present disclosure is not limited to the specific embodiments described above, and may be used in the technical field pertaining to the disclosure without departing from the gist of the disclosure as claimed in the claims. Of course, various modifications can be made by those skilled in the art, and these modifications should not be understood individually from the technical ideas or perspectives of the present disclosure.

100: electronic device 110: memory
111: Dictionary of words 120: Processor

Claims (11)

In electronic devices,
a memory storing a word dictionary indicating a matching relationship between words of a first language and words of a second language, and at least one deep learning model for processing text of the second language; and
Including a processor connected to the memory,
The processor,
Identifying a first text corresponding to the first language and a second text corresponding to the second language within the entire text,
Based on the word dictionary, converting the identified first text into a third text corresponding to the second language,
Obtaining one or more vectors by natural language processing the integrated text including the second text and the third text,
Train the deep learning model based on the obtained vector,
The processor,
Within the first text, identify at least one first keyword included in the word dictionary,
Based on the word dictionary, converting the first keyword included in the first text into a second keyword corresponding to the second language,
With the second keyword set as one entity, converting the first text into the third text corresponding to the second language,
The processor,
Compare the accuracy before and after training of the deep learning model trained based on the obtained vector,
If the accuracy of the deep learning model is lowered as a result of comparing the accuracy before and after training, updating the word dictionary so that the keyword of the second language matching the first keyword is changed,
The processor,
When a plurality of second keywords matching the first keyword are identified according to the word dictionary, for each of the plurality of second keywords, the first keyword is converted according to each second keyword and set as an entity. Converting the first text to the second language to obtain a plurality of third texts,
If the difference between the outputs of the deep learning model according to the input of vectors matching each of the plurality of third texts is less than a threshold, maintaining the word dictionary,
An electronic device that updates the word dictionary when the difference between the outputs of the deep learning model according to the input of vectors matching each of the plurality of third texts is greater than or equal to the threshold. delete delete delete According to paragraph 1,
The processor,
An electronic device that obtains one or more vectors by performing tokenization, part-of-speech tagging, and word embedding on the integrated text. According to paragraph 1,
The memory is,
a first word dictionary for the first specialty, a second word dictionary for the second specialty, and a third word dictionary for the third specialty;
The processor,
When the entire text is set to the first specialized field, the electronic device converts the first text into a third text corresponding to the second language by utilizing the first word dictionary. According to clause 6,
The processor,
If there is no keyword included in the first word dictionary in the first text, at least one of the second word dictionary and the third word dictionary is used to convert the first text into a word corresponding to the second language. 3 An electronic device that converts text into text. In clause 7,
The processor,
Selecting at least one word dictionary related to a keyword included in the first text from the second word dictionary and the third word dictionary,
Converting the first text into a third text corresponding to the second language using the selected word dictionary,
An electronic device that updates the first word dictionary based on a keyword of the second language that matches the keyword in the selected word dictionary. In a method of controlling an electronic device,
The electronic device is,
A word dictionary representing a matching relationship between words of a first language and words of a second language, and at least one deep learning model for processing text of the second language,
The control method is,
identifying a first text corresponding to the first language and a second text corresponding to the second language within the entire text;
converting the identified first text into a third text corresponding to the second language, based on the word dictionary;
Obtaining one or more vectors by natural language processing the integrated text including the second text and the third text; and
Comprising: training the deep learning model based on the obtained vector,
Converting the identified first text into a third text corresponding to the second language,
Within the first text, identify at least one first keyword included in the word dictionary,
Based on the word dictionary, converting the first keyword included in the first text into a second keyword corresponding to the second language,
With the second keyword set as one entity, converting the first text into the third text corresponding to the second language,
The control method of the electronic device is,
Comparing accuracy before and after training of the deep learning model trained based on the obtained vector; and
When the accuracy of the deep learning model is lowered as a result of comparing the accuracy before and after training, updating the word dictionary so that the keyword of the second language matching the first keyword is changed,
The control method of the electronic device is,
When a plurality of second keywords matching the first keyword are identified according to the word dictionary, for each of the plurality of second keywords, the first keyword is converted according to each second keyword and set as an entity. converting the first text into the second language to obtain a plurality of third texts;
maintaining the word dictionary when a difference between outputs of the deep learning model according to input vectors matching each of the plurality of third texts is less than a threshold; and
A control method of an electronic device comprising: updating the word dictionary when a difference between outputs of the deep learning model according to input of vectors matching each of the plurality of third texts is greater than or equal to the threshold. In a computer program stored on a computer-readable medium,
A computer program stored in a computer-readable medium, which is executed by a processor of an electronic device and causes the electronic device to perform the control method of claim 9. delete