KR20220121182A - Method and system for predicting cognitive impairment based on document classification model and fluency tagging - Google Patents

Abstract

Provided is a method for predicting a cognitive impairment based on a document classification model technique and fluency tagging. The method comprises: a step of receiving utterance data of a subject for a document for which is a subject for evaluation; a step of transcribing the utterance data of the subject; a step of performing tagging for a predetermined non-fluency feature among the transcribed utterance data; a step of tokenizing for the tagged non-fluency feature and each word; a step of performing embedding based on the word and document targeting the tokenized utterance data; and a step of outputting a prediction result value corresponding to a parameter affecting cognitive impairment by inputting the embedded utterance data to a pre-learned prediction model.

Description

Method and system for predicting cognitive impairment based on document classification model technique and fluency tagging

The present invention relates to a method and system for predicting cognitive impairment based on a document classification model technique and fluency tagging.

Cognitive impairment refers to abnormalities in the intellectual abilities of the brain, such as memory, language ability, temporal and spatial analysis and construction ability, attention, concentration, judgment, and reasoning ability. The causes of cognitive impairment are very diverse, ranging from physical damage to the brain to senile dementia, infectious diseases, brain tumors, depression, and dozens of others. Once such cognitive impairment occurs, it has a significant negative impact on daily life and social life.

Prior to diagnosing cognitive impairment, simple means for predicting cognitive impairment include the Mini-Mental State Examination (MMSE) and the Boston Naming Test (BNT). This lowers the threshold for diagnosis of cognitive impairment by enabling a simple method to predict the presence or absence of a cognitive impairment before diagnosing cognitive impairment using specialized medical equipment, time, and money.

If these predictions are frequently and easily carried out, leading to early diagnosis, they will play an important role in early detection of larger subsequent diseases such as dementia and promptly taking appropriate measures, such as treatment.

Laid-open Patent Publication No. 10-2019-0021896 (2019.03.06)

The problem to be solved by the present invention is to model the non-fluency characteristics of spontaneous utterances using a document classification model technique, and based on this, predict values for parameters that can affect cognitive impairment for utterances can be derived. It is to provide a method and system for predicting cognitive impairment based on a document classification model technique and fluency tagging.

However, the problems to be solved by the present invention are not limited to the problems described above, and other problems may exist.

A method for predicting cognitive impairment based on a document classification model technique and fluency tagging according to a first aspect of the present invention for solving the above-described problems comprises the steps of: receiving utterance data of a subject for a document to be evaluated; transcribing the subject's utterance data; performing tagging on a predetermined non-fluency characteristic among the transcribed utterance data; tokenizing for each word and the tagged non-fluency feature; performing word- and document-based embedding on the tokenized utterance data; and inputting the embedded speech data into a pre-trained prediction model and outputting a prediction result corresponding to a parameter affecting cognitive impairment.

In some embodiments of the present invention, the step of tagging a predetermined non-fluency characteristic among the transcribed utterance data includes the hesitant behavior characteristic of being silent for a predetermined period of time among the transcribed utterance data, and heavy vowels not related to meaning transfer. , Characteristics of insertion behavior in which words and phrases are included in utterances, content delivered during utterance, characteristics of corrective actions by changing grammatical form or pronunciation of words, characteristics of incomplete behavior in which words or utterances are not completed without completion, multiple completed words Characteristic of repeated phrase repetition, repeated word repetition characteristic of whole words, repeated syllable repetition characteristic, sound repetition characteristic that only part of a phoneme or diphthong is repeated, extended behavior characteristic of one element of a phoneme or vowel being repeated for a long time , and a non-fluency feature including at least one of a clogging behavior feature in which a sound is stopped due to a time difference that occurs when a phoneme is started or a plosive sound is uttered may be tagged.

In some embodiments of the present invention, the tokenizing for each word and the tagged non-fluency characteristic comprises: deleting a content portion of the content portion and a tag portion corresponding to the tagged non-fluency characteristic; and performing tokenization on each word and the non-fluent feature from which the content portion is deleted.

In some embodiments of the present invention, performing word- and document-based embedding on the tokenized utterance data includes: performing word embedding of vectorizing all words constituting the tokenized utterance data; and performing document embedding of adding a vector representing characteristic information of the document to the word embedding result.

In some embodiments of the present invention, the step of performing word- and document-based embedding on the tokenized utterance data may include: among the tokens constituting the tokenized utterance data, not learned in the pre-trained prediction model. When a new token exists, the embedding may be performed.

Some embodiments of the present invention may further include updating the prediction model based on a result of embedding performed based on the new token.

Some embodiments of the present invention, the steps of setting the learning data for learning the predictive model to be input to the input terminal of the predictive model; setting a preset prediction result value corresponding to a parameter affecting the cognitive impairment to be output to an output terminal of the prediction model; and learning the predictive model in which the input terminal and the output terminal are set, wherein the training data is transcribed for predetermined utterance data, and after tagging of a predetermined non-fluency characteristic among the transcribed utterance data is performed, The tagged utterance data may be tokenized and embedded.

In some embodiments of the present invention, the learning data may be tokenized after the content portion of the content portion and the tagged portion corresponding to the tagged non-fluency characteristic are deleted.

In some embodiments of the present invention, the learning data represents feature information of a document constituting the learning data with respect to a word embedding for vectorizing all words constituting the tokenized learning data and a result of the word embedding being performed. Document embedding that adds a vector may be performed and vectorized.

In addition, the cognitive impairment prediction system based on the document classification model technique and fluency tagging according to the second aspect of the present invention includes a communication module that receives the subject's speech data for the document to be evaluated, and a prediction model learned in advance based on the speech data. and a memory in which a program for outputting a predicted result value corresponding to a parameter affecting cognitive impairment is stored through the memory and a processor for executing the program stored in the memory. In this case, as the processor executes the program stored in the memory, the utterance data of the subject is transcribed, and a predetermined non-fluency characteristic is tagged among the transcribed utterance data, and the tagged non-fluency characteristic and After tokenization of each word is performed, word and document-based embedding is performed on the tokenized speech data, the embedded speech data is input to the prediction model to output a prediction result.

A computer program according to another aspect of the present invention for solving the above problems is combined with a computer that is hardware to execute a program for a cognitive impairment prediction method based on the document classification model technique and fluency tagging, and a computer-readable recording medium is stored in

Other specific details of the invention are included in the detailed description and drawings.

According to the present invention described above, by learning and applying a predictive model through tagging, tokenization, and embedding processes for non-fluency features, it is possible to accurately derive prediction results for parameters that can affect cognitive impairment. There are advantages.

Effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a flowchart illustrating a process of learning a predictive model according to an embodiment of the present invention.
2 is a flowchart for explaining the contents of the learning data.
3 is a diagram for explaining a tag defined for a non-fluidity characteristic.
4 is a flowchart of a method for predicting cognitive impairment according to an embodiment of the present invention.
5 is a block diagram of a cognitive impairment prediction system according to an embodiment of the present invention.

Advantages and features of the present invention and methods of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, and only these embodiments allow the disclosure of the present invention to be complete, and those of ordinary skill in the art to which the present invention pertains. It is provided to fully inform those skilled in the art of the scope of the present invention, and the present invention is only defined by the scope of the claims.

The terminology used herein is for the purpose of describing the embodiments and is not intended to limit the present invention. In this specification, the singular also includes the plural, unless specifically stated otherwise in the phrase. As used herein, “comprises” and/or “comprising” does not exclude the presence or addition of one or more other components in addition to the stated components. Like reference numerals refer to like elements throughout, and "and/or" includes each and every combination of one or more of the recited elements. Although "first", "second", etc. are used to describe various elements, these elements are not limited by these terms, of course. These terms are only used to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may be the second component within the spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used herein will have the meaning commonly understood by those of ordinary skill in the art to which this invention belongs. In addition, terms defined in a commonly used dictionary are not to be interpreted ideally or excessively unless clearly specifically defined.

The present invention relates to a method and system (100) for predicting cognitive impairment based on a document classification model technique and fluency tagging.

In order to further simplify and automate the predictive means that can predict cognitive impairment simply and quickly, models that predict spontaneous speech in real time or close to it by analyzing speech and language have been continuously studied.

One embodiment of the present invention models, as part of such a prediction means, the transcribed content of spontaneous speech and the non-fluency characteristics of the content using a document classification model technique of NLP (Natural Language Processing), and presents a predictive model using the same. The present invention is not diagnosing cognitive impairment, but is characterized by deriving predictive values for at least one parameter that may affect cognitive impairment.

Hereinafter, a method for predicting cognitive impairment based on a document classification model technique and fluency tagging (hereinafter, a method for predicting cognitive impairment) according to an embodiment of the present invention will be described with reference to FIGS. 1 to 4 .

1 is a flowchart illustrating a process of learning a predictive model according to an embodiment of the present invention. 2 is a flowchart for explaining the contents of the learning data. 3 is a diagram for explaining a tag defined for a non-fluidity characteristic. 4 is a flowchart of a method for predicting cognitive impairment according to an embodiment of the present invention.

Meanwhile, each of the steps shown in FIGS. 1, 2 and 4 may be understood to be performed by the cognitive impairment prediction system 100 of FIG. 5 to be described later, but is not necessarily limited thereto.

First, the process of learning the predictive model applied to the present invention will be described with reference to FIGS. 1 to 3, and then, the process of predicting cognitive impairment based on the learned predictive model will be described with reference to FIG. 4 .

First, the training data prepared for learning the predictive model is set to be input to the input terminal of the predictive model (S110).

To this end, learning data composed of predetermined speech data is prepared (S210).

In one embodiment, the learning data in the present invention may be predetermined speech data. Speech data is acquired through a 1:1 conversation between a subject and a subject, and a tag indicating the non-fluency characteristic is automatically or manually added to the transcript of the utterance corresponding to the subject during the conversation between the subject and the subject. is one document. Here, the target may be a predetermined person or may be an AI-based AI chatbot.

At this time, the input learning data may be composed of one sentence, but generally composed of paragraphs composed of a plurality of sentences, and in the present invention, such a unit is referred to as a document for convenience. In addition, one document per subject may be provided as one unit of learning data, and in some cases, a plurality of documents may be provided.

It is preferable that the content of the conversation between the subject and the subject be provided as content that can predict the consistent use of words, such as the existing MMSE course or answers to a series of questions prepared in advance, but is not necessarily limited thereto.

In addition, in order to enable semantic analysis of language, it may be better to record the learning data after removing one-word sentences or meaningless chuimsae, but it is not necessarily limited thereto and may be configured differently depending on the purpose of predicting cognitive impairment. Of course.

In addition, in an embodiment of the present invention, an automatic transcription and tagging method to which a predetermined voice recognition technology is applied may be applied in the transcription or tagging method, or a manual transcription and tagging method may be applied. It is recommended to manually refine even created data to increase accuracy.

On the other hand, according to an embodiment of the present invention, after the utterance data is transcribed to form the learning data (S220), tagging is performed on a predetermined non-fluency characteristic among the transcribed utterance data (S230).

For example, the disfluency characteristic may be divided into types as shown in FIG. 3 , and each is divided by a predetermined tag. Specifically, ① hesitation behavior characteristic is an action of silence for a predetermined time (for example, 1 second or more) among utterance data, and a {H} tag is assigned. ② Interjection The behavior characteristic is that a middle vowel, a word, or a phrase that is not related to the delivery of meaning is included in an utterance, and the {I} tag is assigned. ③ Revision The characteristic of the action is to change the content, grammatical form, or pronunciation of words when speaking, and the {R} tag is assigned. ④ The unfinished behavior feature is an action in which a word or utterance is not completed and is usually followed by a correction behavior feature, and a {U} tag is assigned. ⑤ Phrase repetition characteristic is that the {RP} tag is assigned as a repetitive action in a plurality of completed words. ⑥ Word repetition characteristic is the repetition of the entire word, including the repetition of a single syllable word, and a {RW} tag is assigned. ⑦ Syllable repetition (Syllable repetition) is an activity that repeats only syllables in the middle between sound repetition and word repetition, and a {RL} tag is assigned. ⑧ Sound repetition characteristic is that only a part of one phoneme or diphthong is repeated, and the {RS} tag is assigned. ⑨ The prolongation behavior characteristic is the behavior in which one element of a phoneme or vowel lasts for an inappropriately long time, and is assigned a {P} tag. ⑩ The block behavior characteristic is a blocking behavior in which the sound is stopped due to an inappropriate time difference that occurs when starting a phonology or uttering a plosive, and is assigned a {B} tag. At this time, in the case of the blocking behavior characteristic, it is used when the blocking is clearly distinguished, but in other cases, the hesitant behavior characteristic tag is used.

After tagging is performed on the transcribed speech data, tokenizing is performed on the tagged non-fluency characteristics and each word ( S240 ).

Since the document is composed of the sum of sentences and the sentence is composed of a plurality of words, the document can be said to be a list of certain words. In this case, the word, which is the smallest unit constituting the document, may be referred to as a token. Separately referred to as tokens because these units may not correspond to units of words spoken in the language and may contain non-verbal expressions.

In general, one word can be viewed as one token, and in the present invention, one tag corresponding to the non-fluency characteristic is also treated as one specific token.

Examples of tagging shoes against the non-fluidity feature include:

- Input: So I went to school yesterday...

- Tagging: I {so:I} {uh:RL} went to school yesterday{U}

- Tokenization: I+{I}+{RL}+yesterday+school+go+{U}

In this case, according to an embodiment of the present invention, the content part of the tag part and the content part corresponding to the tagged non-fluency feature may be deleted, and tokenization may be performed for each word and the non-fluency feature from which the content part is deleted.

In the above example, since the {I}, {RL} parts in tagging are not related to the meaning of the sentence, only the tag part is changed to a token after the content part is deleted. Also, the fact that 'go' becomes 'got' at the end is applying morpheme separation to tokenization.

On the other hand, in an embodiment of the present invention, various methods may be applied to the method of dividing a sentence for tokenization, and an appropriate method may be selected and applied according to the type, size, and purpose of the model.

When the non-fluency characteristics and tokenization of each word are completed for the training data, word and document-based embedding is performed on the tokenized utterance data (S250).

In this case, according to an embodiment of the present invention, word embedding for vectorizing all words constituting tokenized utterance data and document embedding for adding a vector indicating document characteristic information to the word embedding result may be performed.

In natural language processing, word embedding refers to a sentence or document made up of words or any method of representing the word in a way that a computer can understand. The most useful case is to convert all the words (tokens) that make up all documents into vectors with appropriate values.

A word in word embedding refers to a basic unit constituting a sentence, and the concept is the same as the token described above. Hereinafter, words and tokens are regarded as having the same meaning.

In order to perform embedding, Word2Vec, Doc2Vec, etc., which are general methods in which words constituting each document well represent a semantic positional relationship with each other, may be used. At this time, an embodiment of the present invention applies Doc2Vec as a basis, and in addition to converting all words constituting sentences and documents into vectors in Word2Vec, a method of adding a vector representing a document to a word through Doc2Vec is applied. characterized in that This is to give information about the entire document to the vector representing the document after transforming the vector of each token in the same way as Word2Vec.

An example of Word2Vec is as follows.

Document: That Sky is Blue Sky → Low(w1) + Sky(w2) + Silver(w3) + Blue(w4) + Sky(w2)

Here, w1, w2, w3, and w4 corresponding to words (tokens) mean arbitrary vectors, respectively.

At this time, when Word2Vec is given as w1, w2, w3, X, and w2 as inputs, the process of learning while changing the value (vector) of each word in a direction to increase the probability that X becomes w4 is performed for all words in all documents. , is the process of repeating calculations until all words have an appropriate vector value. Through this process, each word has a mutual relationship and an appropriate position in each document in the vector space formed by the words to express meaning.

An example of Doc2Vec is as follows.

Document: that sky is blue sky → D1 = low(w1) + sky(w2) + silver(w3) + blue(w4) + sky(w2) + p1

Here, w1, w2, w3, w4, and p1 each mean an arbitrary vector.

Doc2Vec is a method that calculates all vectors through the same process as Word2Vec by adding one vector for the document. That is, when the input is given as w1, w2, w3, X, w2, P1, learning in the direction of increasing the probability that X becomes w4 is the same as in Word2Vec, and in this process, p1 representing the characteristics of the document is additionally determined. will lose

Referring back to FIG. 1, the training data prepared in this way is set to be input to the input terminal of the predictive model (S110), and the preset prediction result value corresponding to the parameter affecting cognitive impairment is set to be output to the output terminal of the predictive model do (S120).

In other words, by taking the embedding of all documents in the learning data as the input value X, and setting the preset prediction result values corresponding to parameters affecting cognitive impairment such as cognitive impairment, severity, MMSE score, etc. learn The predictive model may be a binary or polynomial classification model or a linear regression model depending on the setting of the prediction result. On the other hand, as a method of generating a predictive model, existing machine learning methods such as SVM or DNN can be applied, and it can be appropriately selected according to the scale characteristics of the input data.

The prediction result of the prediction model has the same shape as the input data, and may be provided as original data through transcription and tagging processes similarly.

In this way, after the input terminal and the output terminal are set, the prediction model is trained (S130).

If the size of the pre-formed model is sufficiently large, the token of the prediction result value is already included in the embedding result, so that it can be a prediction result value having high accuracy almost as it is. However, even if the size of the model is quite large, if a new token exists, only the embedding is renewed, a fast prediction is performed, and the prediction model is updated based on the changed embedding result.

Hereinafter, a process of predicting cognitive impairment with respect to the subject's speech data based on the learned prediction model will be described with reference to FIG. 4 . In this case, content that overlaps with the content described in the predictive model learning process in FIGS. 1 to 3 will be omitted.

First, utterance data of a subject for a document to be evaluated is received ( S310 ).

Next, the utterance data of the subject is transcribed (S320), and tagging is performed on a predetermined non-fluency characteristic among the transcribed utterance data (S330).

In this case, among the tagging applied to an embodiment of the present invention, the tagging corresponding to the non-fluency characteristic is a hesitant behavior characteristic, an insertion behavior characteristic, a correction behavior characteristic, an incomplete behavior characteristic, a word repetition characteristic, and a syllable repetition characteristic, as described in FIG. , a sound repetition characteristic, an extended behavior characteristic, and a clogging behavior characteristic may be tagged with respect to at least one non-fluency characteristic.

Next, tokenization is performed on the tagged non-fluency characteristics and each word (S340), and word- and document-based embedding is performed on the tokenized speech data (S350).

In an embodiment, the tokenization may be to delete a content part of a tag part and a content part corresponding to the tagged disfluency characteristic, and then perform tokenization on each word and the non-fluency characteristic from which the content part is deleted. .

In this way, word embedding of vectorizing all words constituting the tokenized utterance data is performed, and document embedding of adding a vector representing document characteristic information to the word embedding result is performed.

On the other hand, the embedding process may be performed when there is a new token that has not been learned in the pre-trained prediction model among the tokens constituting the tokenized utterance data. And the prediction model can be updated based on the embedding result performed based on the new token.

In other words, when the size of the trained model is quite large, if there is a new token, it is also possible to use a method of updating the model with the changed embeddings after performing a quick prediction after only new embeddings.

Next, by inputting the embedded speech data to the pre-trained prediction model, the prediction result corresponding to the parameter affecting the cognitive impairment is output (S360).

On the other hand, if the prediction model is trained with enough data or there are few new tokens, the prediction result can be provided first with the existing trained model, and the method of deriving an accurate prediction value by updating the embedding and model later is also applied. can do.

Meanwhile, in the above description, steps S110 to S360 may be further divided into additional steps or combined into fewer steps according to an embodiment of the present invention. In addition, some steps may be omitted if necessary, and the order between the steps may be changed. On the other hand, the contents of FIGS. 1 to 4 are also applied to the cognitive impairment prediction system 100 of FIG. 5 even if other contents are omitted.

5 is a block diagram of a cognitive impairment prediction system 100 according to an embodiment of the present invention.

A cognitive impairment prediction method according to an embodiment of the present invention includes a communication module 110 , a memory 120 , and a processor 130 .

The communication module 110 receives the subject's speech data for the document to be evaluated. Such a communication module 110 may include both a wired communication module and a wireless communication module. The wired communication module may be implemented as a power line communication device, a telephone line communication device, a cable home (MoCA), Ethernet, IEEE1294, an integrated wired home network, and an RS-485 control device. In addition, the wireless communication module may be implemented by wireless LAN (WLAN), Bluetooth, HDR WPAN, UWB, ZigBee, Impulse Radio, 60GHz WPAN, Binary-CDMA, wireless USB technology, wireless HDMI technology, and the like.

The memory 120 stores a program for outputting a prediction result value corresponding to a parameter affecting cognitive impairment through a prediction model learned in advance based on the speech data. Here, the memory 120 collectively refers to a non-volatile storage device and a volatile storage device that continuously maintain stored information even when power is not supplied.

For example, the memory 120 may include a compact flash (CF) card, a secure digital (SD) card, a memory stick, a solid-state drive (SSD), and a micro SD card. NAND flash memory such as cards, magnetic computer storage devices such as hard disk drives (HDDs), and optical disc drives such as CD-ROMs and DVD-ROMs. can

In addition, the program stored in the memory 120 may be implemented in the form of software or hardware such as a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC), and may perform predetermined roles.

The processor 130 executes a program stored in the memory 120 . Specifically, the processor 130 transcribes the subject's utterance data, performs tagging on a predetermined non-fluency characteristic among the transcribed utterance data, and tokenizes the tagged non-fluency characteristic and each word, and a token After performing word- and document-based embedding on the uttered utterance data, the embedded utterance data is input to the prediction model to output a prediction result.

The embodiment of the present invention described above may be implemented as a program (or application) to be executed in combination with a computer, which is hardware, and stored in a medium.

The above-mentioned program, in order for the computer to read the program and execute the methods implemented as a program, C, C++, JAVA, Ruby, which the processor (CPU) of the computer can read through the device interface of the computer; It may include code coded in a computer language such as machine language. Such code may include functional code related to a function defining functions necessary for executing the methods, etc. can do. In addition, the code may further include additional information necessary for the processor of the computer to execute the functions or code related to memory reference for which location (address address) in the internal or external memory of the computer should be referenced. have. In addition, when the processor of the computer needs to communicate with any other computer or server located remotely in order to execute the functions, the code uses the communication module of the computer to determine how to communicate with any other computer or server remotely. It may further include a communication-related code for whether to communicate and what information or media to transmit and receive during communication.

The storage medium is not a medium that stores data for a short moment, such as a register, a cache, a memory, etc., but a medium that stores data semi-permanently and can be read by a device. Specifically, examples of the storage medium include, but are not limited to, ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. That is, the program may be stored in various recording media on various servers accessible by the computer or in various recording media on the computer of the user. In addition, the medium may be distributed in a computer system connected by a network, and computer-readable codes may be stored in a distributed manner.

The foregoing description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and likewise components described as distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention. do.

100: cognitive impairment prediction system
110: communication module
120: memory
130: processor

Claims (10)

A method performed by a computer comprising:
receiving utterance data of a subject for a document to be evaluated;
transcribing the subject's utterance data;
performing tagging on a predetermined non-fluency characteristic among the transcribed utterance data;
tokenizing for each word and the tagged non-fluency feature;
performing word- and document-based embedding on the tokenized utterance data; and
Comprising the step of inputting the embedded speech data into a pre-learned prediction model and outputting a prediction result value corresponding to a parameter affecting cognitive impairment,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
According to claim 1,
The step of tagging a predetermined non-fluency characteristic among the transcribed utterance data includes:
Among the transcribed speech data, the characteristic of hesitant behavior for a predetermined period of time, the characteristic of intervening vowels, words, and phrases that are not related to meaning transfer, the characteristics of the insertion behavior that is included in the utterance, the content delivered during utterance, and correction of speaking by changing the grammatical form or pronunciation of the word Behavioral characteristics, incomplete behavioral characteristics in which words or utterances are not completed without completion, oral repetition characteristics in which plural completed words are repeated, word repetition characteristics in which whole words are repeated, syllable repetition characteristics in which syllables are repeated, single phoneme or diphthong Non-fluidity comprising at least one of a repeating feature in which only a portion of a vowel is repeated, an extended behavioral feature in which an element of a phoneme or vowel is long lasting, and a clogging behavioral feature in which the sound is interrupted due to a time lag that occurs when the phonological initiation or plosive is uttered. To perform tagging for features,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
According to claim 1,
Tokenizing for each word and the tagged non-fluency feature comprises:
deleting a content portion of the tagged portion and the content portion corresponding to the tagged non-fluent characteristic; and
performing tokenization for each word and the non-fluency feature from which the content portion has been deleted,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
4. The method of claim 3,
The step of performing word- and document-based embedding on the tokenized utterance data includes:
performing word embedding of vectorizing all words constituting the tokenized utterance data; and
performing document embedding of adding a vector representing characteristic information of the document to the word embedding result,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
5. The method of claim 4,
The step of performing word- and document-based embedding on the tokenized utterance data includes:
Among the tokens constituting the tokenized utterance data, if there is a new token that has not been learned in the pre-trained prediction model, the embedding is performed,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
6. The method of claim 5,
Further comprising the step of updating the prediction model based on the embedding result performed based on the new token,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
According to claim 1,
setting learning data for learning the predictive model to be input to an input terminal of the predictive model;
setting a preset prediction result value corresponding to a parameter affecting the cognitive impairment to be output to an output terminal of the prediction model; and
Further comprising the step of learning the predictive model in which the input terminal and the output terminal are set,
The training data is transcribed from predetermined utterance data, and after tagging is performed on a predetermined non-fluency characteristic among the transcribed utterance data, the tagged utterance data is tokenized and embedded,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
6. The method of claim 5,
The learning data is tokenized after the content part of the tag part and the content part corresponding to the tagged non-fluency characteristic is deleted,
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
7. The method of claim 6,
The training data includes word embedding for vectorizing all words constituting the tokenized training data, and document embedding for adding a vector representing characteristic information of a document constituting the training data to a result of the word embedding being performed. which is vectorized and
Cognitive impairment prediction method based on document classification model technique and fluency tagging.
In the cognitive impairment prediction system based on the document classification model technique and fluency tagging,
A communication module for receiving the subject's speech data for the document to be evaluated;
a memory in which a program is stored for outputting a prediction result value corresponding to a parameter affecting cognitive impairment through a prediction model learned in advance based on the speech data; and
Including a processor for executing the program stored in the memory,
As the processor executes the program stored in the memory, the utterance data of the subject is transcribed, the transcribed utterance data is tagged with a predetermined non-fluency characteristic, and the tagged non-fluency characteristic and each word are transcribed. After performing tokenization on the tokenized utterance data and performing word and document-based embedding on the tokenized utterance data, inputting the embedded utterance data into the prediction model to output a prediction result value,
Cognitive impairment prediction system based on document classification model technique and fluency tagging.

Images