Method and device for providing cognitive training

Abstract

A device for providing cognitive training according to one aspect comprises: memory in which at least one program is stored; and at least one processor for executing the at least one program, wherein the at least one processor determines a cognitive training program to be provided to a patient on the basis of prior data of the patient, and provides feedback on the basis of the amount of progress made in the cognitive training program by the patient.

Description

Method and device for providing cognitive training

[0001] The present disclosure relates to a method and apparatus for providing cognitive training.

[0002] In general, human cognitive function consists of various sub-components such as attention, memory, judgment, and language ability, and these functions can decline due to causes such as aging, disease, and trauma. Accordingly, various training programs have been proposed to prevent or improve cognitive decline.

[0003] However, conventional cognitive training methods generally provide uniform content to users, which limits their ability to offer appropriate feedback or training paths tailored to individual cognitive abilities or responses. Furthermore, even when some methods collect user response data, they lack the functionality to analyze or incorporate it in real-time, which reduces training efficiency.

[0004] Therefore, there is a growing need for a more sophisticated and personalized method of providing cognitive training that can accurately evaluate the user's current cognitive state and dynamically adjust training content based on the evaluation results.

[0005] The invention provides a method and apparatus for providing cognitive training. Additionally, the invention provides a computer-readable recording medium that records a program for executing the above method on a computer. The technical problems to be solved are not limited to those described above, and other technical problems may exist.

[0006] According to one aspect of the present disclosure, a method for diagnosing a disease using an artificial intelligence model can be provided, comprising: determining a cognitive training program to be provided to a patient based on the patient's prior data; and providing feedback based on the degree of performance of the patient's cognitive training program.

[0007] An apparatus according to another aspect of the present disclosure comprises: a memory in which at least one program is stored; and at least one processor for executing said at least one program, wherein the at least one processor determines a cognitive training program to be provided to said patient based on said patient’s prior data and can provide feedback based on the degree of performance of said cognitive training program by said patient.

[0008] A computer-readable recording medium according to another aspect of the present disclosure includes a recording medium that records a program for executing the above-described method on a computer.

[0009] According to one embodiment of the present invention, by using an artificial intelligence model to communicate with the user in real-time and bidirectionally, it is possible to help improve the user's cognitive function.

[0010] In addition, cognitive reserve can be effectively activated by selectively providing a cognitive training method suitable for the user based on the results of the user's cognitive training performance.

[0011] In addition, it can provide cognitive training to users while simultaneously achieving promotional effects for commercial items.

[0012] In addition, by providing cognitive training to the user based on minimal rules, it may be easier to guide the patient's associative flow toward the planning direction of the cognitive training compared to cases where only artificial intelligence models are used.

[0013] The objectives of the present invention are not limited to those mentioned above, and other unmentioned objectives will be clearly understood by those skilled in the art from the description below.

[0014] FIG. 1 is a diagram illustrating an example of a method for providing cognitive training according to one embodiment.

[0015] FIG. 2 is a block diagram illustrating an example of a device for providing cognitive training according to one embodiment.

[0016] FIG. 3 is a flowchart illustrating an example of a method for providing cognitive training according to one embodiment.

[0017] FIG. 4 is a flowchart illustrating an example of a method for determining a cognitive training program according to one embodiment.

[0018] FIGS. 5 and 6 are flowcharts illustrating an example of a method for determining training words related to commercialization item names according to one embodiment.

[0019] FIG. 7 is a diagram illustrating an example of a method for providing unit training to a patient using training words related to a commercialized item according to one embodiment.

[0020] FIG. 8 is a flowchart illustrating an example of a method for providing real-time feedback to a patient according to one embodiment.

[0021] FIG. 9 is a flowchart illustrating an example of a method for determining whether a patient's input is correct according to one embodiment.

[0022] FIG. 10 is a diagram illustrating an example of a method for determining a patient's input as a correct or incorrect answer in association training according to one embodiment.

[0023] FIGS. 11a and FIGS. 11b are drawings illustrating an example of a method for providing feedback to a patient based on determining the patient's input as the correct answer according to one embodiment.

[0024] FIG. 12 is a diagram illustrating an example of a method for providing feedback to a patient based on determining the patient's input as an incorrect answer according to one embodiment.

[0025] FIG. 13 is a flowchart illustrating an example of a method for providing final feedback to a patient according to one embodiment.

[0026] FIG. 14 is a diagram illustrating an example of a method for deriving recall evaluation data according to one embodiment.

[0027] The present invention relates to a method and apparatus for providing cognitive training. An apparatus for providing cognitive training according to one embodiment of the present invention comprises: a memory storing at least one program; and at least one processor for executing said at least one program, wherein the at least one processor determines a cognitive training program to be provided to said patient based on said patient's prior data and provides feedback based on the degree of performance of said cognitive training program by said patient.

[0028] The terms used in the embodiments have been selected to be as widely used and general as possible; however, these may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been selected at the applicant's discretion, and in such cases, their meanings will be described in detail in the relevant description section. Therefore, terms used in the specification must be defined not merely by their names, but based on their meanings and the overall content of the specification.

[0029] When a part of the specification is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "~ unit" or "~ module" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software.

[0030] Additionally, terms including ordinal numbers, such as "first" or "second," used in the specification may be used to describe various components, but said components should not be limited by said terms. Such terms may be used for the purpose of distinguishing one component from another.

[0031] The present disclosure will be described in detail below with reference to the attached drawings. Specifically, a method for providing cognitive training according to one embodiment will be described in more detail with reference to FIGS. 1 to 14. However, embodiments may be implemented in various different forms and are not limited to the examples described herein.

[0032]

[0033] FIG. 1 is a diagram illustrating an example of a method for providing cognitive training according to one embodiment.

[0034] Hereinafter, with reference to FIG. 1, an example of a method for providing cognitive training will be described.

[0035] Referring to FIG. 1, the patient (1) can perform a cognitive training program using a device (10) that provides cognitive training.

[0036] A device (10) providing cognitive training can provide a cognitive training program to a patient (1) and provide feedback on the input of the patient (1) performing the provided cognitive training program. Here, the input of the patient (1) may include all forms of input, such as voice, text, touch, motion, and biosignals (e.g., brainwaves, gaze, heart rate, etc.).

[0037] In the present disclosure, a cognitive training program can be provided to a patient (1) using an artificial intelligence model of a device (10) that provides cognitive training.

[0038] For example, a device (10) providing cognitive training can provide a cognitive training program to a patient (1) and obtain input from the patient (1) performing the provided cognitive training program. Additionally, the device (10) providing cognitive training can input the input of the patient (1) obtained as input data for an artificial intelligence model into an artificial intelligence model, analyze the input of the patient (1) to derive feedback as output data for the artificial intelligence model, and provide it to the patient (1).

[0039] Here, an artificial intelligence model refers to a set of machine learning algorithms that use a layered algorithmic structure based on deep neural networks in machine learning technology and cognitive science.

[0040] For example, an artificial intelligence model may be composed of an input layer that receives input signals or data from the outside, an output layer that outputs output signals or data corresponding to the input data, and at least one hidden layer located between the input layer and the output layer, which receives a signal from the input layer, extracts a feature, and transmits it to the output layer. The output layer receives a signal or data from the hidden layer and outputs it to the outside.

[0041] More specifically, the artificial intelligence model in this disclosure may refer to a Large Language Model (LLM), and the Large Language Model is a language model having a vast amount of parameters, meaning an artificial intelligence model capable of understanding and generating human language patterns by learning text data.

[0042] That is, the artificial intelligence model in the present disclosure is an artificial intelligence model capable of understanding and generating human language patterns, and can determine whether emotional words included in the input of the patient (1) are used appropriately, whether the context of the text is natural, whether a predetermined word is used in an appropriate position, and whether various types of words (e.g., nouns, pronouns, numerals, particles, adjectives, determiners, adverbs, verbs, interjection words, etc.) correspond to each other.

[0043] Additionally, the artificial intelligence model described below in the present disclosure may include, in one artificial intelligence model, a unit training providing model that provides unit training to a patient (1), a unit training determining model that determines unit training to be provided to the patient (1), an embedding model that generates embedding vectors, a training word determining model that determines training words to be provided to the patient (1), an image generating model that generates images using training words, a recall evaluation model that evaluates the patient (1)'s ability to recall training words, a voice evaluation model that evaluates the patient (1)'s voice, a word evaluation model that evaluates words used by the patient (1), a sentence structure evaluation model that evaluates the sentence structure included in the patient (1)'s input, a description evaluation model that evaluates the patient (1)'s ability to describe, an emotion evaluation model that evaluates the emotion included in the patient (1)'s input, a correct answer determination model that determines whether the patient (1)'s input is correct or incorrect, a feedback providing model that provides feedback to the patient (1), a general sentence generating model that provides answers according to the patient (1)'s request, and a basic language model that provides assistance in the patient (1)'s cognitive training performance, and the model that provides feedback to the patient (1) may include a hint generating model that provides hints to the patient (1). It may include an association sentence generation model that provides the following association cognitive training to the patient (1).

[0044] Meanwhile, the aforementioned unit training provision model, unit training decision model, embedding model, training word decision model, image generation model, recall evaluation model, speech evaluation model, word evaluation model, sentence structure evaluation model, description evaluation model, sentiment evaluation model, correct answer determination model, feedback provision model (e.g., hint generation model and associative sentence generation model), general sentence generation model, and basic language model may be included in a single artificial intelligence model, but are not limited thereto, and at least one artificial intelligence model may be an individual artificial intelligence model.

[0045] The method of providing cognitive training described in the present disclosure may provide cognitive training to a patient (1) using at least one artificial intelligence model. Furthermore, the method of providing cognitive training described in the present disclosure may not merely use at least one artificial intelligence model, but may also utilize a minimum set of rules for providing cognitive training to the patient (1), thereby maximizing the cognitive training effect on the patient (1). In other words, the present invention may provide cognitive training to a patient by combining at least one artificial intelligence model and a set of rules, that is, at least one artificial intelligence model and a rule-based algorithm.

[0046] Here, the pre-set rules may include, but are not limited to, a pre-set determination rule for determining whether the input of the patient (1) is correct, and a pre-set feedback rule for providing feedback to the patient (1), and details will be described later.

[0047] Accordingly, the patient (1) performing cognitive training can communicate bidirectionally with the device (10) providing the cognitive training to receive real-time feedback, and can also receive final feedback after the cognitive training is finished.

[0048]

[0049] FIG. 2 is a block diagram illustrating an example of a device for providing cognitive training according to one embodiment.

[0050] Referring to FIG. 2, a device (hereinafter referred to as the 'device') (200) that provides cognitive training may include a communication unit (210), a processor, and a memory (230). Only the components related to the embodiment are shown in the device (200) of FIG. 2. Therefore, it is obvious to a person skilled in the art that other general-purpose components may be included in addition to the components shown in FIG. 2.

[0051] The communication unit (210) may include one or more components that enable wired / wireless communication with an external server or external device. For example, the communication unit (210) may include a short-range communication unit (not shown) and a mobile communication unit (not shown) for communication with an external server or external device.

[0052] The memory (230) is hardware that stores various data processed within the device (200) and can store programs for processing and controlling the processor.

[0053] For example, the memory (230) may store various data such as the patient's medical data, data evaluating the patient's cognitive function, data on the patient's previous cognitive training program performance, data on the patient's cognitive training program performance, cognitive training program, commercial item name, keywords related to the commercial item, words related to keywords, cognitive training, unit training, training words, response words, hint data, various types of words such as nouns, pronouns, numerals, particles, adjectives, determiners, adverbs, verbs, interjections, matching data between various types of words, real-time feedback data, final feedback data, and data generated according to the operation of the processor. Additionally, the memory (230) may store an operating system (OS) and at least one program (e.g., a program required for the processor to operate).

[0054] The memory (230) may include RAM (random access memory), such as DRAM (dynamic random access memory) and SRAM (static random access memory), ROM (read-only memory), EEPROM (electrically erasable programmable read-only memory), CD-ROM, Blu-ray or other optical disc storage, HDD (hard disk drive), SSD (solid state drive), or flash memory.

[0055] The processor controls the overall operation of the device (200). For example, the processor can control the input unit (not shown), display (not shown), communication unit (210), memory (230), etc., by executing programs stored in memory (230).

[0056] The processor may be implemented using at least one of ASICs (application specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field programmable gate arrays), controllers, micro-controller units, microprocessors, and other electrical units for performing functions.

[0057] The processor can control the operation of the device (200) by executing programs stored in memory (230). For example, the processor can perform at least some of the methods for providing cognitive training described with reference to FIGS. 3 to 14.

[0058]

[0059] FIG. 3 is a flowchart illustrating an example of a method for providing cognitive training according to one embodiment.

[0060] Hereinafter, with reference to FIG. 3, an example of a method for providing cognitive training will be described.

[0061] Referring to FIG. 3, a method for providing cognitive training may include steps 310 to 320. However, it is not limited thereto, and other general operations in addition to those shown in FIG. 3 may be further included in the method for providing cognitive training. Additionally, as described above with reference to FIG. 1 and FIG. 2, at least one of the operations of the flowchart shown in FIG. 3 may be processed by a processor.

[0062] First, in step 310, the processor may determine a cognitive training program to be provided to the patient based on the patient's prior data. Here, the prior data may include data evaluating the patient's cognitive function, the patient's medical data, and the patient's previous cognitive training program performance data.

[0063] For example, the processor can determine unit training in areas where the patient's cognitive ability is impaired based on prior data, and determine the training words to be used in the unit training based on the patient's word usage.

[0064] In addition, at step 320, the processor can provide feedback based on the patient's level of performance in the cognitive training program.

[0065] For example, feedback may include real-time feedback provided during the performance of the cognitive training program and final feedback provided after the performance of the cognitive training program.

[0066] As an example, the processor can provide real-time feedback to the patient during the execution of a cognitive training program.

[0067] For example, the processor can determine whether the input of a patient performing a cognitive training program is correct. More specifically, the processor can extract input words included in the patient's input, calculate the phonemic distance and semantic distance of the input words, and determine whether the patient's input is correct based on the phonemic distance and semantic distance.

[0068] For example, the processor can provide real-time feedback based on the result of determining whether the patient's input is correct. More specifically, the processor can provide real-time hint feedback to the patient based on the result of determining that the patient's input is not correct.

[0069] As another example, the processor can provide final feedback to the patient after performing a cognitive training program.

[0070] For example, the processor can derive evaluation data for each of multiple parameters related to cognitive function based on the result data of the patient performing a cognitive training program, and provide the evaluation data to the patient as final feedback.

[0071] Hereinafter, with reference to FIGS. 4 to 14, a method by which a processor provides cognitive training will be described in more detail.

[0072] First, the processor can determine a cognitive training program to be provided to the patient based on the patient's prior data. Here, the prior data may include data evaluating the patient's cognitive function, the patient's medical data, data on the patient's previous performance in cognitive training programs, and demographic information such as the patient's gender, age group, and occupation.

[0073] For example, data evaluating a patient's cognitive function may refer to the results of standardized cognitive tests that take into account the patient's school age and age. In other words, data evaluating a patient's cognitive function may refer to the results of cognitive tests conducted to determine in which domain the patient's cognitive function has declined and to what extent.

[0074] In addition, the patient's medical data may include the diagnosis received from a medical professional regarding cognitive function, cognitive function status, vision, hearing, motor function, occupation, etc.

[0075] Meanwhile, the processor can control the communication unit to obtain data evaluating the patient's cognitive function and the patient's medical data from the hospital via a designated server or cloud, or to obtain them through the patient's input.

[0076] In addition, the patient's previous cognitive training program performance data may refer to the result data of the patient performing the previous cognitive training program, and may refer to the final feedback data in this disclosure. Here, the final feedback data is evaluation data for each of a plurality of parameters related to cognitive function, and details will be described later.

[0077] In addition, the patient's demographic information may refer to information related to the patient's demographic and social background, such as gender, age group, occupation, education level, marital status, income, place of residence, and retirement status, and is not limited to the examples above.

[0078] Therefore, the processor can determine a cognitive training program suitable for the patient using the patient's prior data and provide the determined cognitive training program to the patient.

[0079] That is, the processor can determine unit training included in the cognitive training program and determine training words to be used in the determined unit training. Here, the training words may include general training words widely used in daily life without being focused on a specific category, and training words focused on a specific category and related to commercial item names. Accordingly, the training words in the present disclosure may be understood to include general training words and / or training words related to commercial item names.

[0080] As one example, the processor can determine the training words (e.g., general training words) to be used in unit training based on the patient's word usage. As another example, the processor can determine the training words related to the names of commercialized items provided in unit training based on prior data.

[0081]

[0082] FIG. 4 is a flowchart illustrating an example of a method for determining a cognitive training program according to one embodiment. FIG. 5 and FIG. 6 are flowcharts illustrating an example of a method for determining a training word related to a commercialized item name according to one embodiment.

[0083] Hereinafter, with reference to FIGS. 4, 5, and 6, an example of a method for determining a cognitive training program to be provided to a patient by a processor is described. More specifically, with reference to FIG. 4, an example of a method for determining unit training and general training words to be provided to a patient by a processor is described, and with reference to FIGS. 5 and 6, an example of a method for determining training words related to commercialized item names by a processor is described.

[0084] First, referring to FIG. 4, in step 410, the processor can determine unit training for the area where the patient's cognitive ability has declined based on prior data.

[0085] Based on prior data, the processor can determine the areas where the patient's cognitive ability has declined and determine unit training that can improve cognitive ability in those areas.

[0086] For example, the processor can determine the area where a patient's cognitive ability has declined by using demographic information such as the patient's occupation and retirement status, speech information such as pronunciation accuracy, word information such as the difficulty and category of words used, and cognitive function test information such as SNSB, CERAD, MMSE, and digital cognitive function tests.

[0087] For example, based on the patient's prior data, the processor can determine the area where the patient's cognitive ability is impaired among the attention, language, visuospatial, memory, and frontal lobe areas related to cognitive ability.

[0088] Therefore, the processor can determine unit training that can strengthen the areas where the patient's cognitive ability has declined.

[0089] In addition, the processor can determine unit training by taking into account the patient's physical difficulties. Specifically, the processor can determine unit training that primarily utilizes visual elements if the patient's hearing is impaired, and unit training that primarily utilizes auditory elements if the patient's vision is impaired.

[0090] Additionally, the processor can determine the unit training by considering the patient's accuracy rate for each unit training included in the previous cognitive training program performance data. Specifically, the processor can set higher weights for unit training with lower accuracy rates and lower weights for unit training with higher accuracy rates, as included in the patient's previous cognitive training program performance data. Therefore, the processor can determine the unit training with the higher weight as the unit training to be provided to the current patient.

[0091] Unit training may include intensive training, association training, and associative training, depending on the specific details of the training. More specifically, unit training may include intensive training for the repetitive learning of training words, association training for learning words associated with training words, and associative training for learning the meanings between training words.

[0092] As an example, intensive training can be a type of training that strengthens the patient's frontal lobe. For instance, intensive training could utilize a word chain game, where the patient and the unit training provider model can perform the game by exchanging words in real time.

[0093] As another example, association training can be a type of training that strengthens a patient's temporal lobe. For instance, association training may involve having the patient input another word or phrase that is sensorially associated with a specific word or phrase.

[0094] As another example, association training can be a type of training that strengthens the patient's parietal lobe. For instance, association training can be a type of training where the patient is provided with at least two words and asked to tell a story that is semantically connected using the provided words.

[0095] Meanwhile, as described above, the processor can determine unit training using a unit training decision model. Specifically, the processor inputs the patient's prior data as input data for the unit training decision model, and as output data for the unit training decision model, it can strengthen areas where the patient's cognitive function is impaired and obtain information regarding unit training that considers physical difficulties and the accuracy rate for each unit training. Therefore, the processor can determine the unit training to be provided to the patient based on the information regarding unit training obtained as output data for the unit training decision model.

[0096] In step 420, the processor can determine the training words to be used in unit training based on the patient's word usage.

[0097] The processor can determine the training words to be used in unit training based on the patient's prior data.

[0098] For example, the processor can determine training words based on the patient's word usage data included in the patient's previous cognitive training program performance data.

[0099] For example, the processor can derive specific words that the patient does not use or words related to specific categories from the patient's previous cognitive training program performance data. Specifically, the processor can determine that the patient frequently used words related to animals but not words related to art when performing the previous cognitive training program. Therefore, the processor can decide to prioritize providing words related to art over words related to animals as training words in that unit of training.

[0100] In addition, the processor can determine training words by taking into account the patient's cognitive ability.

[0101] For example, the processor can determine the difficulty and educational level of each word using the National Institute of Korean Language's word API (Application Programming Interface). Additionally, the processor can determine words with a difficulty level corresponding to the patient's cognitive ability as training words.

[0102] Therefore, the processor can determine training words by considering both words related to specific categories that the patient does not use and the patient's cognitive abilities. Specifically, even if the patient does not use words related to art, the processor can selectively determine high-difficulty words (e.g., Impressionism, Pop Art, Renaissance, etc.) or low-difficulty words (e.g., artist, exhibition, masterpiece, etc.) based on the patient's cognitive abilities.

[0103] Meanwhile, the processor can input the patient's previous cognitive training program performance data as input data for the training word decision model. Additionally, the processor can acquire words related to specific categories not used by the patient and words of difficulty considering the patient's cognitive ability as output data for the training word decision model. Therefore, the processor can determine the acquired words as training words.

[0104] Hereinafter, with reference to FIGS. 5 and 6, an example of a method in which a processor determines training words related to commercialization item names is described.

[0105] Referring to FIG. 5, in step 510, the processor can determine a commercialization item corresponding to the patient's demographic information based on prior data.

[0106] The processor can determine commercialization items based on the patient's gender, age group, and occupation. Here, commercialization items may refer to all items subject to commerce, regardless of the form of provision, such as goods or services.

[0107] For example, if the patient is male, the processor may determine commercial items that men typically have a primary interest in, such as automobiles, sports equipment, razors, alcoholic beverages, and men's products, as the commercial items that serve as the basis for determining training words. Additionally, if the patient is female, the processor may determine commercial items that women typically have a primary interest in, such as beauty products, yoga, Pilates, and women's products, as the commercial items that serve as the basis for determining training words.

[0108] For example, the processor may determine commercial items such as games, school supplies, clothing, and shoes as the basis for determining training words when the patient is in their teens or twenties, commercial items such as children's education, automobiles, and financial management when the patient is in their thirties or forties, and commercial items such as health, premium food, and medical care when the patient is in their fifties or sixties.

[0109] For example, the processor may determine commercial items such as sunscreen products, heating products, and cooling products as the basis for determining training words if the patient's occupation group involves a lot of outdoor activities, and commercial items such as chairs, office supplies, and drawers if the patient's occupation group involves a lot of indoor activities.

[0110] However, the method for determining commercialization items is not limited to the examples described above, and the processor may determine commercialization items based on whether the time period during which the patient performs unit training is a weekday or weekend, whether it is morning or afternoon, the number of people living with the patient, the patient's residential area, etc.

[0111] Meanwhile, in step 520, the processor can determine training words using keywords related to commercialization items.

[0112] The processor can acquire keywords related to commercialization items determined based on prior data.

[0113] For example, the processor may obtain keywords related to a commercialization item from a user. Here, the user may refer to, but is not limited to, a company of the commercialization item that wishes to promote the commercialization item.

[0114] Therefore, the processor can obtain at least one keyword from the user based on the concept, features, advantages, and target customer base of the commercialization item.

[0115] For example, if the commercialization item is ramen, the processor can obtain keywords from the user such as "chewy noodles," "spicy broth," "eating alone," "late-night snack," "easy meal," etc. In this case, specific keywords desired by the user can be obtained.

[0116] Meanwhile, the processor may obtain keywords through web crawling rather than from the user, and may obtain keywords by various methods not described in this disclosure. Here, web crawling may refer to the operation of software that visits and collects web pages to extract data such as text and images.

[0117] Referring to FIG. 6, in step 610, the processor can preprocess text related to the keyword to determine the similarity between the obtained word and the keyword. Here, text related to the keyword may refer to a corpus, and the dictionary meaning of a corpus is a collection of text, voice, etc., that is organized by collecting data for a specific purpose.

[0118] The processor can obtain at least one piece of text related to a keyword for each keyword through web crawling. For example, the processor can obtain data such as words, sentences, and images related to keywords by crawling domains such as social network services (SNS), blogs, and news articles.

[0119] In addition, the processor can generate embedding vectors by performing a predetermined preprocessing on text related to keywords.

[0120] For example, the processor can remove noise by filtering text related to the acquired keywords. In other words, the processor can remove noise such as words unrelated to the keywords or typos.

[0121] For example, a processor can generate word-unit tokens by tokenizing text related to keywords. Here, tokenization refers to the process of converting text data, such as corpus data, into tokens at the word or character level. Therefore, a processor can generate word-unit tokens using text related to keywords.

[0122] For example, the processor can normalize word-unit tokens. Specifically, the processor can normalize word-unit tokens based on standardization methods such as surface form standardization, which organizes the Unicode, width, and spacing of word-unit tokens and unifies uppercase and lowercase letters; morphological standardization, which restores the base form of verbs and makes them headwords; and synonym standardization, which unifies synonyms and abbreviations into their original forms.

[0123] Additionally, the processor can generate a fixed-length embedding vector using normalized word-unit tokens. In other words, the processor inputs normalized word-unit tokens into the embedding model as input data and obtains a fixed-length embedding vector as output data from the embedding model. Here, the embedding model may be an artificial intelligence model that has been pre-trained to generate an embedding vector as output data using text, tokens, etc., as input data. Furthermore, an embedding vector may refer to data such as text or images represented as a real number vector.

[0124] Meanwhile, the processor can preprocess text related to the keyword to determine the similarity between the obtained word and the keyword.

[0125] For example, a processor can determine the embedding similarity between the keyword and the embedding vector of a word-unit token generated using text related to the keyword. Here, embedding similarity refers to a numerical value representing the distance between vectors, where embeddings representing data such as text or images as real-valued vectors become closer as they share the same meaning.

[0126] Therefore, the processor can determine embedding similarity by calculating the distance between the embedding vector of a word-unit token and the embedding vector of a keyword. In other words, the processor can calculate the similarity between a normalized word and a keyword using word-unit tokens generated by preprocessing text related to the keyword.

[0127] Meanwhile, in step 620, the processor can determine candidate training words based on similarity.

[0128] For example, the processor may determine a word-unit token, i.e., a normalized word, whose similarity satisfies a pre-set condition, and determine the determined normalized word as a candidate training word. Here, the pre-set condition may mean a condition in which the similarity between the word-unit token, i.e., the normalized word, and the keyword is included within the top x% (where x is a number between 0 and 100), and x may be 20, but is not limited thereto. That is, the top x% of words with high similarity between at least one normalized word and the keyword may be determined as candidate training words.

[0129] In addition, the processor can determine the difficulty of candidate training words based on candidate training word information.

[0130] For example, the processor can determine the difficulty of a candidate training word based on similarity information, word level information, and usage frequency information. Specifically, the processor can determine a candidate training word as having high difficulty (i.e., being difficult) if the similarity between the candidate training word and the keyword is low, the level of the candidate training word is high (i.e., being difficult), or the usage frequency of the candidate training word is low. Conversely, the processor can determine a candidate training word as having low difficulty (i.e., being easy) if the similarity between the candidate training word and the keyword is high, the level of the candidate training word is low (i.e., being easy), or the usage frequency of the candidate training word is high.

[0131] In addition, at step 630, the processor can determine a training word among candidate training words based on the patient's unit training performance data.

[0132] For example, the processor can determine the training words to be used in unit training based on the patient's unit training performance data.

[0133] For example, the processor can determine training words based on the patient's word usage data included in the patient's previous unit training performance data.

[0134] For example, the processor can derive specific words that the patient does not use or words related to specific categories from the patient's previous unit training performance data. Specifically, the processor can determine that the patient frequently used words related to animals but not words related to art when performing previous unit training. Therefore, the processor can decide to prioritize providing words related to art over words related to animals as training words in that unit training.

[0135] In addition, the processor can determine training words by taking into account the patient's cognitive ability.

[0136] For example, the processor can determine the difficulty and educational level of each word using the National Institute of Korean Language's word API (Application Programming Interface). Additionally, the processor can determine words with a difficulty level corresponding to the patient's cognitive ability as training words.

[0137] Therefore, the processor can determine training words by considering both words related to specific categories that the patient does not use and the patient's cognitive abilities. Specifically, even if the patient does not use words related to art, the processor can selectively determine high-difficulty words (e.g., Impressionism, Pop Art, Renaissance, etc.) or low-difficulty words (e.g., artist, exhibition, masterpiece, etc.) as training words based on the patient's cognitive abilities.

[0138] Meanwhile, the processor can input the patient's previous unit training performance data as input data for the training word decision model. Additionally, the processor can acquire words related to specific categories not used by the patient and words of difficulty considering the patient's cognitive ability as output data for the training word decision model. Furthermore, the processor can determine the acquired words as training words.

[0139] Meanwhile, with reference to FIG. 4, a method for determining general training words is explained, and with reference to FIG. 5 and FIG. 6, a method for determining training words related to commercialization items is explained, and the methods for determining general training words and training words related to commercialization items are explained as being distinct from each other. However, this is only for convenience of explanation, and the method of FIG. 4 can be used to determine training words related to commercialization items, and the methods of FIG. 5 and FIG. 6 can be used to determine general training words.

[0140] The processor can provide a cognitive training program to a patient using unit training and training words (at least one of general training words or training words related to commercial items) determined by the method described above. That is, the processor can provide unit training to the patient based on the training words.

[0141] As described above, unit training may include intensive training, association training, and association training depending on the details of the training. More specifically, unit training may include intensive training for repeatedly learning training words, association training for learning words associated with training words, and association training for learning the meanings between training words.

[0142] Below, we will specifically describe how a processor provides intensive training, association training, and associative training to a patient using general training words or training words related to commercial items.

[0143] First, specifically explain how the processor provides intensive training, association training, and associative training to the patient using general training words.

[0144] As an example, the processor can provide intensive training to the patient using general training words.

[0145] Specifically, the processor can acquire a specific word (hereinafter referred to as Word A) as input from a patient and input the acquired Word A (e.g., barbershop) into a unit training provision model. Additionally, the processor can acquire Word B (e.g., pine tree) starting with the last letter of Word A (e.g., barbershop) from the unit training provision model and provide it to the patient, and can again acquire Word C (e.g., rainbow) starting with the last letter of Word B (e.g., pine tree) from the patient.

[0146] In addition, the processor can adjust the difficulty of the intensive training by controlling the number of words used in the word chain game.

[0147] As another example, the processor can provide association training to the patient using general training words.

[0148] Specifically, the processor can provide the patient with the question, "What is a monkey's butt like?" and obtain the input "A monkey's butt is red" from the patient in response to that question. In other words, the processor can induce the patient to recall a visually associated phrase, "A monkey's butt is red." Additionally, the processor can input the patient's answer, "A monkey's butt is red," into a unit training model, obtain the phrase "If it's red?" as output data from the model, and provide this to the patient, thereby inducing the patient to provide a response regarding a word or phrase visually associated with the color red. Furthermore, the processor can obtain the answer "If it's red, it's an apple" from the patient, input this response into a unit training model, and obtain the phrase "What about an apple?" as output data. Additionally, the processor can provide the phrase "What about an apple?" to the patient again, thereby inducing the patient to provide a response regarding a word or phrase gustatorily associated with an apple.

[0149] Here, the unit training provider model may be a model trained to determine whether a patient's input corresponds to a sensory word. That is, the unit training provider model can determine that the patient's input (apple) and the sensory word (red) correspond when the patient's input to "If it is red?" is "apple," and can determine that the patient's input (banana) and the sensory word (red) do not correspond when the patient's input is "banana."

[0150] In addition, the processor can adjust the difficulty of intensive training by controlling the number of words used for association training.

[0151] As another example, the processor can provide associated training to the patient using common training words.

[0152] Specifically, the processor can provide the patient with words and emotions (e.g., fun, fear, sadness, etc.) that are not semantically close to each other, thereby inducing the patient to create a story that matches the provided words and emotions. That is, the processor can provide the patient with "cherry blossoms" and "ramen" as semantically unrelated words and "sadness" as an emotion, and obtain input from the patient such as, "I went to the Han River to see the cherry blossoms and ate ramen, but I spilled the ramen and felt sad." Additionally, the processor inputs the obtained input into a unit training provider model and obtains a result from the unit training provider model determining whether the story entered by the patient is constructed to be semantically related to all the provided words and emotions.

[0153] In addition, the processor can adjust the difficulty of intensive training by controlling the number of words undergoing combined training.

[0154]

[0155] Below, we will specifically describe how a processor provides intensive training, association training, and association training to a patient using training words related to commercialized items.

[0156] FIG. 7 is a diagram illustrating an example of a method for providing unit training to a patient using training words related to a commercialized item according to one embodiment.

[0157] Hereinafter, with reference to FIG. 7, an example of a method in which a processor provides unit training to a patient using training words related to commercialized items is described.

[0158] Referring to FIG. 7, the processor can provide unit training to a patient using a commercial item name (710) and a training word (720) related to the commercial item.

[0159] As an example, the processor can provide intensive training to the patient using training words (720) related to commercial items.

[0160] For example, intensive training may be training that involves repeatedly providing the patient with the commercialization item name (710) and training words (720) related to the commercialization item to help them remember the training words (720) related to the commercialization item.

[0161] Specifically, the processor may repeatedly provide the patient with a commercial item name (710) and training words (720) related to the commercial item. For example, if the commercial item name (710) is "Patent Ramen," the processor may repeatedly provide the patient with training words (720) related to ramen, such as "Kimchi," "Pickled Radish," and "Late-night snack." Additionally, the patient may memorize the provided commercial item name (710) and training words (720) related to the commercial item in the form of a sentence. That is, the patient may memorize the commercial item name (710) and training words (720) related to the commercial item provided by the processor in the form of a sentence such as, "Last night, I ate Patent Ramen with Kimchi and Pickled Radish as a late-night snack."

[0162] Accordingly, the processor can provide the patient with a commercialization item name (710) and obtain the patient's input regarding training words (720) related to the commercialization item. Additionally, the processor can provide feedback to the patient regarding the patient's input, and details regarding the method of providing feedback will be described later.

[0163] As another example, the processor can provide association training to the patient using training words related to commercialized items.

[0164] For example, association training may be training that involves having a patient input another word or phrase that is sensually associated with a specific word, phrase, or image.

[0165] Specifically, the processor may provide the patient with a commercial item name (710) and obtain input from the patient such as words, phrases, etc. associated with the commercial item name (710). For example, the processor may provide the patient with the word "Patent Ramen" and obtain input from the patient such as visual, gustatory, and tactile associations such as "chewy," "bouncy," "red," and "spicy," which are training words (720) associated with "Patent Ramen." That is, the processor may make the patient recall words that are sensorially associated with the commercial item name (710) "Patent Ramen," and the words that are sensorially associated may be training words (720).

[0166] Additionally, the processor can generate an image (730) containing a commercial item based on the commercial item name (710) and training words (720) associated with the commercial item.

[0167] For example, the processor may input a commercial item name (710) and a training word (720) related to the commercial item as input data to an image generation model, and obtain an image (730) containing the commercial item as output data. Here, the image generation model may be an artificial intelligence model trained to generate an image containing the commercial item and evoking the training word (720) using the commercial item name (710) and the training word (720) related to the commercial item.

[0168] Additionally, the processor can provide an image (730) containing a commercial item to the patient to obtain the patient's input containing words related to the commercial item, and provide feedback based on the patient's input.

[0169] For example, the processor can provide the patient with an image (730) containing a commercial item called "patented ramen" and obtain visual, gustatory, and tactile inputs from the patient, such as "chewy," "bouncy," "red," and "spicy," which are training words (720) associated with "patented ramen." That is, the processor can make the patient recall words that are sensuously associated with the image (730) containing the commercial item called "patented ramen," and the words that are sensuously associated may be training words (720).

[0170] In addition, the processor can provide feedback to the patient based on the patient's input, and details regarding the method of providing feedback will be described later.

[0171] Meanwhile, for convenience of explanation, only an image (730) containing a commercial item called "patented ramen" is shown in FIG. 7, but commercial items are not limited to this, and images containing all commercial items including goods and services can be provided to the patient. That is, if a commercial item called "branded clothing care machine" is included in the image, the processor can obtain input from the patient such as "fine dust," "antibacterial control," and "custom clothing care," and provide feedback thereon.

[0172] Accordingly, the processor can provide the patient with the commercial item name (710) of "Patent Ramen" or an image (730) containing the commercial item of "Patent Ramen", and obtain the patient's input regarding a training word (720) related to the commercial item of "Patent Ramen". Additionally, the processor can provide feedback to the patient regarding the patient's input, and details regarding the method of providing feedback will be described later.

[0173] As another example, the processor can provide associated training to the patient using training words related to commercialized items.

[0174] For example, the association training may be a training in which a patient is provided with a commercial item name (710) and a training word (720) related to the commercial item, so that the patient tells a story that is semantically connected using the provided commercial item name (710) and the training word (720). Additionally, the association training may be a training in which an image generated using the commercial item name (710) and the training word (720) related to the commercial item is provided to the patient, so that the patient can recall the commercial item name (710) or the training word (720) related to the commercial item from the provided image.

[0175] Specifically, the processor can provide the patient with a commercial item name (710) and training words (720) related to the commercial item, and induce the patient to come up with a story that includes the provided commercial item name (710) and training words (720) related to the commercial item. That is, the processor can provide the patient with "patented ramen" as the commercial item name (710) and "chewy," "bouncy," "red," and "spicy" as training words (720) related to the commercial item, and can obtain input from the patient such as, "I ate patented ramen last night, the noodles were chewy and bouncy, and the red broth was spicy but delicious."

[0176] Additionally, as described above, the processor can generate an image (730) containing a commercial item based on the commercial item name (710) and training words (720).

[0177] For example, the processor may provide the patient with an image (730) containing a generated commercial item and induce the patient to make up a story by looking at the image (730) containing the commercial item. That is, the processor may provide the patient with an image containing "patented ramen" as a commercial item and obtain input from the patient such as "the noodles of the patented ramen are bouncy and the red broth is spicy."

[0178] For example, the processor may provide a question regarding the name of the commercial item (710), “What is the name of the ramen that the person is eating in the image?” along with an image (730) containing the generated commercial item, and obtain an input from the patient, “It is patented ramen.” Additionally, the processor may provide a question regarding the training word (720), “What is the taste of the ramen that the person is eating in the image?” along with an image (730) containing the generated commercial item, and obtain an input from the patient, “Patented ramen is spicy.”

[0179] Accordingly, the processor can provide the patient with the commercial item name (710) "Patent Ramen" and a training word (720) or an image (730) containing the commercial item "Patent Ramen", and obtain input from the patient regarding the commercial item name (710) "Patent Ramen" or the training word (720) related to the commercial item name (710). Additionally, the processor can provide feedback to the patient regarding the patient's input, and details regarding the method of providing feedback will be described later.

[0180] Meanwhile, the processor can provide feedback based on the patient's level of performance in the cognitive training program. Here, the feedback may include real-time feedback provided during the performance of the cognitive training program and final feedback provided after the performance of the cognitive training program.

[0181] As one example, the processor can provide real-time feedback based on patient input obtained during the execution of a cognitive training program. As another example, the processor can provide final feedback based on patient input obtained after the execution of the cognitive training program.

[0182]

[0183] First, with reference to FIGS. 8, FIGS. 9, and FIGS. 10, an example of a method by which a processor provides real-time feedback will be described.

[0184] FIG. 8 is a flowchart illustrating an example of a method for providing real-time feedback to a patient according to one embodiment. FIG. 9 is a flowchart illustrating an example of a method for determining whether a patient's input is correct according to one embodiment. FIG. 10 is a diagram illustrating an example of a method for determining a patient's input as correct or incorrect in association training according to one embodiment.

[0185] Referring to FIG. 8, in step 810, the processor can determine whether the input of a patient performing a cognitive training program is correct.

[0186] For example, the processor can determine whether the input from a patient performing unit training is the correct answer that aligns with the intent of that unit training.

[0187] First, we will explain how the processor determines whether the input from a patient performing unit training based on general training words is correct.

[0188] As an example, the processor can determine whether the input from a patient performing intensive training is correct.

[0189] For example, the processor may acquire a specific word (hereinafter referred to as Word A) as input from a patient and input the acquired Word A (e.g., barbershop) into a correct answer determination model. Additionally, the processor may acquire Word B (e.g., pine tree) starting with the last letter of Word A (e.g., barbershop) from the correct answer determination model and provide it to the patient. Furthermore, the processor may input the patient's corresponding answer into the correct answer determination model and obtain a result determining whether the patient's answer is correct.

[0190] For example, the correct answer determination model can determine that the patient's input is the correct answer if it obtains word C (e.g., rainbow) that starts with the last letter of word B (e.g., pine tree), and can determine that the patient's input is not the correct answer (i.e., incorrect) if it obtains word D (e.g., forsythia) that does not start with the last letter of word B (e.g., pine tree).

[0191] As another example, the processor can determine whether the input from a patient performing association training is correct.

[0192] For example, the processor can provide a specific word or phrase to a patient, obtain a word or phrase from the patient that is sensorially associated with the specific word, and input it into a correct answer determination model. Additionally, the processor can obtain a result from the correct answer determination model determining whether the word or phrase obtained from the patient corresponds sensorially.

[0193] For example, the correct answer discrimination model can determine that if the patient's input to "If it is red?" is "apple," the patient's input (apple) corresponds to the sensory word (red) (i.e., the correct answer), and if the patient's input is "banana," the patient's input (banana) does not correspond to the sensory word (red) (i.e., the incorrect answer).

[0194] As another example, the processor can determine whether the input from a patient performing federated training is correct.

[0195] For example, the processor can provide the patient with words and emotions that are not semantically close to each other and obtain a story created by the patient that matches those words and emotions. Additionally, the processor can input the story created by the patient into a ground truth model and obtain a result from the model determining whether the words connect naturally within the flow of the story and whether the story was constructed to match the corresponding emotion.

[0196] For example, if the provided words are "cherry blossoms" and "ramen" and the emotion is "sadness," the correct answer discrimination model can determine that the patient's input, "I went to the Han River to see the cherry blossoms and ate ramen, but I spilled the ramen and felt sad," is semantically naturally connected and is a story that fits the provided emotion (i.e., correct answer), and can determine that the patient's input, "I made ramen with cherry blossoms and it was fun," is not semantically naturally connected and is a story that does not fit the provided emotion (i.e., incorrect answer).

[0197] In other words, the processor can train a correct answer discrimination model by semantically and sensorially analyzing the words included in the patient's input to determine whether the patient's input is suitable for the purpose of the unit training—that is, whether it is the correct answer. Furthermore, the processor can use the trained correct answer discrimination model to determine whether the patient's input is correct or incorrect.

[0198] The following describes how a processor determines whether the input of a patient performing unit training based on training words related to commercialized items is correct.

[0199] As an example, the processor can determine whether the input from a patient performing intensive training is correct.

[0200] For example, the processor can perform intensive training by repeatedly providing the patient with the commercial item name "Patented Ramen" and training words "Kimchi," "Pickled Radish," and "Late-night Snack" so that the patient remembers the provided words.

[0201] For example, the processor inputs the patient's input into a correct answer determination model and can obtain a result from the correct answer determination model determining whether the patient's input is correct.

[0202] Specifically, the processor inputs the patient's inputs of "patent ramen," "kimchi," "pickled radish," and "dinner" into a correct answer determination model, and can obtain a result in which the patient's input is determined to be correct (or incorrect) because it is identical or similar (or different) to the commercial item name or training word. Here, the processor can obtain as output data of the correct answer determination model a result in which the patient's input "dinner" is determined to be correct because, although "dinner" among the patient's inputs is not identical to the training word "late-night snack," it is semantically similar, and the method by which the correct answer determination model determines "dinner" and "late-night snack" as correct because they are semantically similar is described later.

[0203] As another example, the processor can determine whether the input from a patient performing association training is correct.

[0204] For example, the processor may provide a patient with the name of a commercial item or an image containing a commercial item, and obtain a word or phrase that is sensorially associated with the name of the commercial item or the image containing the commercial item from the patient, and input it into a correct answer discrimination model. Additionally, the processor may obtain a result from the correct answer discrimination model determining whether the word or phrase obtained from the patient sensorially corresponds (i.e., is identical or similar) to a training word associated with the name of the commercial item.

[0205] Specifically, the processor can provide the patient with the commercial item name "Patent Ramen" or an image containing "Patent Ramen," and obtain the words "red" and "spicy" from the patient and input them into a correct answer determination model. Additionally, the processor can obtain a result as output data of the correct answer determination model in which the patient's input is judged to be correct (or incorrect) because it is identical or similar (or different) to the training word. Here, the processor can obtain a result as output data of the correct answer determination model in which the patient's input "spicy" is judged to be correct because, although it is not identical to the training word "spicy," it is semantically similar; and the method by which the correct answer determination model judges "dinner" and "late-night snack" as correct because they are semantically similar will be described later.

[0206] As another example, the processor can determine whether the input from a patient performing federated training is correct.

[0207] For example, the processor can provide the patient with commercial item names and training words, and obtain a story created by the patient using the commercial item names and training words. Additionally, the processor can provide the patient with images generated using the commercial item names and training words, and the patient can obtain commercial item names or training words that come to mind from the provided images, or obtain answers to questions related to the provided images.

[0208] For example, the processor inputs a story created by a patient into a correct answer determination model and obtains a result from the model determining whether the words are naturally connected in accordance with the flow of the story and whether the story was constructed to fit the atmosphere. Additionally, the processor inputs words that come to mind from images obtained from the patient into the correct answer determination model and obtains a result from the model determining whether the words are identified as correct (or incorrect) because they are identical or similar (or differ) to commercial item names or training words.

[0209] Specifically, the processor provides the patient with "patented ramen" as the name of a commercialized item and "chewy," "bouncy," "red," and "spicy" as training words, and obtains the patient's input, "I ate patented ramen last night; the noodles were chewy and bouncy, and the red broth was spicy but delicious," and inputs it into a correct answer discrimination model. Additionally, the processor can obtain the result of judging the patient's input as the correct answer as output data of the correct answer discrimination model, because all the words included in the patient's input are naturally connected in a flow to form a story.

[0210] Additionally, the processor can provide the patient with an image containing the commercial item and a question regarding the name of the commercial item, "What is the name of the ramen that the person is eating in the image?", and obtain the patient's input, "It is patented ramen," and input it into a correct answer determination model. Additionally, the processor can obtain the result of determining the patient's input as the correct answer as the output data of the correct answer determination model, in which the patient's input is judged to be the correct answer because it is identical or similar to the name of the commercial item.

[0211] The following describes how a processor uses a correct answer discrimination model to determine whether the meaning between a word entered by a patient and the correct answer word is identical or similar.

[0212] The processor can extract the patient's input in single-word units and calculate the phonemic distance and semantic distance of the extracted words.

[0213] As one example, the processor can calculate the phoneme distance between a word included in the patient's input and a general training word. As another example, the processor can calculate the phoneme distance between a word included in the patient's input and a commercial item name or a training word related to the commercial item. For the sake of convenience of explanation, words included in the patient's input are referred to as response words, and words that can serve as correct answers, such as general training words, commercial item names, and training words related to commercial items, are referred to as correct answer words.

[0214] Referring to FIG. 9, in step 910, the processor can extract input words included in the patient's input and calculate the phoneme distance and semantic distance of the input words.

[0215] The processor can extract the patient's input in single-word units and calculate the phonemic distance and semantic distance of the extracted words.

[0216] For example, the processor can calculate the phoneme distance between a word included in the patient's input (hereinafter, the response word) and the correct word.

[0217] Specifically, the processor can convert the pronunciation (or phoneme) of the patient's response word. Here, the method by which the processor converts the pronunciation of the response word may include, but is not limited to, a dictionary-based conversion method, a method using a G2P (grapheme-to-phoneme) model, or a method based on the IPA (international phonetic alphabet). Accordingly, the processor can convert the pronunciation of the response word and calculate the phoneme distance based on the converted pronunciation.

[0218] Additionally, the processor can convert the response word into a spectrogram and input the converted response word as input data to an embedding model. Additionally, the processor can embed the input response word to generate a predetermined embedding vector and obtain the generated embedding vector as output data to the embedding model. Thus, the processor can calculate the phoneme distance using the cosine similarity and DTW (dynamic time warping) of the embedding vector.

[0219] For example, the processor can calculate the semantic distance between the patient's response word and the correct word.

[0220] Specifically, the processor can input the patient's response word into an embedding model and generate a predetermined embedding vector by embedding the input response word. Additionally, the processor can calculate the semantic distance between the patient's response word and the correct answer word by calculating the cosine similarity or Euclidean distance of the generated embedding vector.

[0221] Referring again to FIG. 9, in step 920, the processor can determine whether the patient's input is correct based on phoneme distance and semantic distance.

[0222] For example, the processor can determine whether a patient's input is correct by calculating the phonemic distance and semantic distance of the patient's response word during unit training and comparing the patient's response word with the correct word of the corresponding unit training.

[0223] For example, the processor may determine that the response word is the correct answer if the phoneme distance of the response word is less than or equal to a preset value. Additionally, the processor may determine that the response word is the correct answer if the cosine similarity of the response word is greater than or equal to a preset value, that is, if the semantic distance is less than or equal to a preset value.

[0224] That is, as described above, the processor can determine whether the patient's input is correct by calculating the phoneme distance and semantic distance between the patient's response word and the correct word input for each unit of training.

[0225] For example, the processor can determine whether the patient's input is correct for each unit of training using general training words.

[0226] As one example, the processor can determine whether a patient's input is correct based on whether the input is a word suitable for the word chain game to proceed. As another example, the processor can determine whether a patient's input is correct based on whether the input corresponds to a sensory word during association training. As yet another example, the processor can determine whether a patient's input is correct based on whether the input is a story or sentence suitable for the provided word and emotion during association training.

[0227] For example, the processor can determine whether the patient's input is correct for each unit of training using training words related to commercialized items.

[0228] As an example, if the input of a patient undergoing intensive training—that is, the response word—is "dinner," and the training word related to the commercialization item—that is, the correct word—is "late-night snack," the processor can determine whether the patient's input is the correct answer by calculating the phonemic and semantic distances between "dinner" and "late-night snack." In other words, the processor can determine that the response word "dinner" is the correct answer by judging that the similarity between "dinner" and "late-night snack" is greater than or equal to a preset value.

[0229] As another example, if a patient performing association training gives the response word "kalkalhada" and the correct answer word is "maepda," the processor can determine whether the patient's input is the correct answer by calculating the phonemic and semantic distances between "kalkalhada" and "maepda." In other words, the processor can determine that the response word "kalkalhada" is the correct answer by judging that the similarity between "kalkalhada" and "maepda" is greater than or equal to a preset value.

[0230] As another example, although not described above, when a patient performing federated training has a response word of "patent surface" and a correct answer word of "patent ramen," the processor can determine whether the patient's input is correct by calculating the phoneme distance and semantic distance between "patent surface" and "patent ramen." That is, the processor can determine that the response word "patent surface" is correct by determining that the similarity between "patent surface" and "patent ramen" is greater than or equal to a preset value.

[0231] Meanwhile, if the unit training provided to the patient is associative training, the processor can determine whether the patient's input is correct in a way different from the method described above.

[0232] That is, as described above, the processor can provide the patient with cognitive training sentences that cause the patient to continuously associate words associated with the training words, such as "A monkey's butt is red," "If it's red, it's an apple," and "Apples are delicious."

[0233] Accordingly, the processor can determine whether a patient's input is correct or incorrect based on a pre-established discrimination rule, using a correct answer discrimination model, and based on Type 1 input words and Type 2 input words included in the patient's input to the first cognitive training sentence. Here, Type 1 words may refer to noun words and Type 2 words may refer to adjective words, but this is for the convenience of explanation and is not limited thereto. That is, the processor can determine whether a patient's input is correct or incorrect based not only on noun words and adjective words, but also on pronouns, numerals, particles, determiners, adverbs, verbs, and interjections included in the patient's input.

[0234] Additionally, the input word refers to a word included in the patient's input, the storage word described below refers to a word stored in memory or other storage device, the generated word refers to a word generated by the processor, and the representative word may refer to a word that is most highly associated with a predetermined word and can represent the predetermined word.

[0235] Accordingly, the first type input word may refer to a specific type of word entered by the patient. Additionally, the first type stored word may refer to a specific type of word already stored in memory or other storage devices. Additionally, the first type generated word may refer to a specific type of word generated by the processor. Furthermore, the first type representative word may refer to a specific type of word capable of representing a predetermined word.

[0236] Here, specific types of words may include nouns, adjectives, pronouns, numerals, particles, determiners, adverbs, verbs, and interjections, as described above; and descriptions regarding Type 2 input words, Type 2 stored words, Type 2 generated words, and Type 2 representative words are omitted as they overlap with the descriptions of Type 1 words.

[0237] For example, the processor may extract a first type input word and a second type input word from at least one word included in the patient's input, generate at least one second type representative word corresponding to the first type input word based on the first type input word using a correct answer determination model, and determine the patient's input as correct or incorrect according to a pre-set determination rule based on at least one second type representative word and the second type input word.

[0238] In addition, the processor can determine whether the patient's input is correct or incorrect based on a pre-set discrimination rule using a correct answer determination model, based on whether the second type input word corresponds to at least one second type representative word.

[0239] Referring to FIG. 10, the processor may provide a patient with a first cognitive training sentence (1000) containing a predetermined word (e.g., a training word) as an association training.

[0240] For example, the processor can generate a first cognitive training sentence (1000) using at least one first type word and at least one second type word. Additionally, the processor can provide the generated first cognitive training sentence (1000) to a patient.

[0241] More specifically, referring to the first cognitive training sentence (1000) illustrated in FIG. 10, the processor can generate a first cognitive training sentence (1000) including first type words such as "monkey" and "buttocks" and second type words such as "red" and "if red," which are conjugated and connected words of the second type word "red," and can provide the generated first cognitive training sentence (1000) to a patient.

[0242] Additionally, the processor can generate a first cognitive training sentence (1000) based on at least one first-type word and at least one second-type word using a basic language model. That is, the processor can generate a first cognitive training sentence (1000) as output data by using at least one first-type word and at least one second-type word as input data to the basic language model.

[0243] The processor can obtain the patient's input (1010) for the first cognitive training sentence (1000) from the patient.

[0244] For example, the processor may obtain a patient's input (1010) of "if it is red, it is an apple" as a response to a first cognitive training sentence (1000) from the patient. Additionally, the patient's input (1010) may include at least one first type word and at least one second type word.

[0245] More specifically, referring to the patient's input (1010) illustrated in FIG. 10, the patient's input (1010) may include a first type of input word (1020) such as "apple" and a second type of input word (1030) such as "red".

[0246] Based on the acquired patient input (1010), the processor can determine whether the patient's input (1010) is a correct response corresponding to the first cognitive training sentence (1000), that is, whether it is a correct answer or an incorrect answer.

[0247] For example, the processor can extract a first type input word (1020) and a second type input word (1030) among at least one word included in the patient's input (1010).

[0248] For example, if the processor contains multiple first-type words in the patient's input (1010), it can extract the first-type word most closely associated with the second-type input word (1030) as the first-type input word (1020). Additionally, if the first-type word is a compound word consisting of two first-type words separated by a space, such as "monkey butt," the processor can extract the compound word "monkey butt" as a single first-type input word (1020).

[0249] For example, if a second type input word (1030) included in the patient's input (1010) is not a base form, the processor can extract the second type input word and normalize it to a base form. More specifically, if a second type input word (1030) included in the patient's input (1010) is not a base form, such as "if it is red," the processor can extract "if it is red" and normalize it to a base form, such as "red" or "red."

[0250] For example, the processor can use the patient's input (1010) as input data for the correct answer determination model to obtain, as output data, the first type input word (1020) that is most closely associated with the second type input word (1030) among a plurality of first type words. That is, the correct answer determination model may refer to a model that has been trained to calculate the semantic distance between a plurality of words and to derive first type words and second type words that match each other according to the calculated semantic distance.

[0251] The processor can generate at least one second type representative word corresponding to the first type input word (1020) based on the first type input word (1020).

[0252] For example, the processor can use a first type input word (1020) as input data for a correct answer determination model to obtain at least one second type representative word corresponding to the first type input word (1020) as output data. That is, the correct answer determination model may refer to a model that has been trained to generate at least one second type representative word that modifies the first type input word (1020) in everyday life.

[0253] More specifically, the processor can generate words such as "round," "pretty," "crispy," "delicious," "sour," "fragrant," "smooth," "cool," "red," "fresh," etc. as second-type representative words modifying "apple," which is a first-type input word (1020) included in the patient's input (1010). Additionally, if the first-type input word included in the patient's input is "tiger," the processor can generate words such as "fierce," "frightening," "agile," "strong," "spirited," "courageous," "imposing," "majestic," "rough," "fast," etc. as second-type representative words modifying "tiger."

[0254] Meanwhile, the processor may also generate at least one second-type representative word corresponding to the first-type input word (1020) based on the first-type input word (1020) according to a prompt, that is, a pre-set generation rule.

[0255] For example, the processor can generate at least one second-type representative word corresponding to the first-type input word (1020) according to a pre-set generation rule to generate at least one second-type input word that people generally think of for the first-type input word (1020).

[0256] More specifically, the processor can generate words such as "round," "pretty," "crispy," "delicious," "sour," "fragrant," "smooth," "cool," "red," "fresh," etc. as representative words of the second type that modify "apple," which is a first type input word (1020) included in the patient's input (1010), according to a pre-set generation rule. Additionally, if the first type input word included in the patient's input is "tiger," the processor can generate words such as "fierce," "scary," "agile," "strong," "spirited," "courageous," "imposing," "majestic," "rough," "fast," etc. as representative words of the second type that modify "tiger."

[0257] The processor can determine whether the patient's input (1010) is correct or incorrect according to a pre-set discrimination rule based on at least one second-type representative word and a second-type input word (1030).

[0258] For example, the processor can determine whether the patient's input (1010) is correct or incorrect based on a pre-set determination rule based on whether the second type input word (1030) corresponds to at least one second type representative word using a correct answer determination model.

[0259] For example, a pre-set determination rule may mean a rule that determines the patient's input (1010) as correct or incorrect depending on whether the second type input word (1030) corresponds to at least one second type representative word. In other words, the pre-set determination rule may mean a rule that determines the patient's input (1010) as correct if the second type input word (1030) corresponds to at least one second type representative word that people typically use when describing the first type input word (1020), and determines the patient's input (1010) as incorrect if there is no correspondence.

[0260] More specifically, the processor can determine whether the second type input word (1030), "if red," corresponds to at least one second type representative word when the second type input word (1030) included in the patient's input (1010) is "if red," and at least one generated second type representative word is "round," "pretty," "crispy," "delicious," "sour," "fragrant," "smooth," "cool," "red," or "fresh." That is, the processor can determine whether the second type input word (1030) is identical or similar to at least one second type representative word. Therefore, the processor can determine the patient's input (1010) as the correct answer, as the second type input word (1030), "if red," corresponds to the same or similar meaning as a second type representative word such as "red."

[0261] Additionally, the processor can determine the patient's input (1010) as the correct answer, even if the second type input word is "thin" and the second type representative word is "flat," since "thin" and "flat" can correspond to second type words that are used interchangeably.

[0262] Conversely, when a patient input such as “if red, it’s a tiger” is obtained as an answer to a first cognitive training sentence (1000) such as “A monkey’s butt is red. If it’s red?”, the processor can generate at least one second type representative word that modifies the first type input word “tiger,” and can determine whether the patient’s input is correct or incorrect using the generated at least one second type representative word and the second type input word “if red.”

[0263] More specifically, the processor can determine whether the second-type input word "red mask" corresponds to at least one second-type representative word when the second-type input word included in the patient's input is "red mask" and at least one generated second-type representative word is "fierce," "scary," "agile," "strong," "spirited," "courageous," "imposing," "majestic," "rough," or "fast." Therefore, since the second-type input word "red mask" is not identical or similar to all second-type representative words, the processor can determine the patient's input as an incorrect answer.

[0264] By utilizing a correct answer discrimination model together with pre-set discrimination rules, the processor can effectively guide the patient's associative flow in a direction consistent with the training intent, compared to the case where only an artificial intelligence model is used. In other words, the processor can determine the patient's input as correct or incorrect based on pre-set discrimination rules; however, by using the correct answer discrimination model to identify the patient's intent regarding the input, the processor can determine the patient's input as correct even if it is not the correct answer according to the pre-set discrimination rules, provided that the input aligns with the training intent. Here, the pre-set discrimination rules can be configured differently depending on the concept of cognitive training; for example, different discrimination rules can be set according to a concept that evokes a specific mascot character or a concept that evokes a specific product.

[0265] More specifically, if a patient answers "If it's red, it's a car" as a response to the first cognitive training sentence (1000), according to a pre-set discrimination rule, people generally do not think of words like "red" as association words for "car," so the patient's input can be determined as an incorrect answer. However, if the concept of the cognitive training is one that evokes a red sports car, the processor can use a correct answer determination model to identify the intent of the patient's input "If it's red, it's a car" and determine whether the patient's input aligns with the concept of the cognitive training, that is, whether it aligns with the training intent. In other words, the processor can use the patient's input as input data for the correct answer determination model and derive a correct or incorrect answer judgment result that considers the patient's input and the concept of the cognitive training as output data. Therefore, in this case, the processor can identify the training intent of the concept that evokes a red sports car and ultimately determine the patient's input "If it's red, it's a car" as the correct answer.

[0266] Conversely, if the patient answers "If it's red, it's a car" as a response to the first cognitive training sentence (1000), if only the artificial intelligence model is used, the processor can determine the patient's input as the correct answer because "red car" also exists. However, if the artificial intelligence model and a pre-set discrimination rule are used together, although "red car" may actually exist, the processor can determine the patient's input as the incorrect answer because people generally do not think of words like "red" as association words for "car."

[0267] In this way, the correct answer discrimination model and pre-established discrimination rules are used complementarily depending on the patient's input, the concept of cognitive training, etc., to determine the patient's input as correct or incorrect, and to more effectively guide the patient's associative flow in a direction consistent with the training intention.

[0268] Meanwhile, the processor may use a correct answer determination model to perform at least one of the operations included in the method for determining whether the patient's input (1010) described above is correct or incorrect. For example, the correct answer determination model may refer to a model that has been trained to use the patient's input (1010) as input data to derive a result determining whether the patient's input (1010) is correct or incorrect as output data. That is, the processor may use the correct answer determination model to perform at least one of the operations included in the method for determining whether the patient's input (1010) described above is correct or incorrect, even for operations that are not explicitly described as using the correct answer determination model.

[0269]

[0270] Referring again to FIG. 8, in step 820, the processor may provide real-time feedback based on the result of determining whether the patient's input is correct. Here, the real-time feedback may include praise feedback provided when the patient's input is correct, feedback providing a new cognitive training sentence as the next problem, and hint feedback provided when the patient's input is incorrect. Additionally, the hint feedback may include a hint sentence generated based on a training word or a word included in the patient's input.

[0271] For example, the processor can provide praise feedback to the patient based on the result of determining that the patient's input is correct, and provide the next problem as feedback.

[0272] For example, the processor can provide real-time hint feedback to the patient based on the result that the patient's input is determined to be incorrect. Additionally, the processor can provide real-time hint feedback even if it fails to acquire the patient's input for a preset period of time, that is, if the patient's non-response persists.

[0273] As an example, if the input from a patient performing intensive training is incorrect or there is no input from the patient for a preset period of time, the processor can provide the patient with words before and after the incorrect input as real-time hint feedback.

[0274] For example, when a patient performs intensive training using general training words, if the patient cannot think of "pine tree" during a word chain game involving "barber shop," "pine tree," and "rainbow," the processor can provide real-time hint feedback to the patient asking, "What word goes between 'barber shop' and 'rainbow'?"

[0275] In addition, the processor can provide hint feedback to the patient, such as "What word goes between 'barbershop' and 'rainbow'?", even if the patient cannot recall "pine tree" during the review process after performing intensive training.

[0276] Meanwhile, when a patient performs intensive training using training words related to a commercialized item, if the patient does not remember the name of the commercialized item, "Patented Ramen," the processor can provide one of the training words related to ramen, such as "Kimchi," "Pickled Radish," or "Late-night Snack," to the patient as real-time hint feedback. Conversely, if the patient does not remember the training words related to ramen, such as "Kimchi," "Pickled Radish," or "Late-night Snack," the processor can provide the name of the commercialized item, "Patented Ramen," to the patient as real-time hint feedback.

[0277] In addition, the processor can provide the training word or commercial item name as hint feedback to the patient even if the patient cannot recall the commercial item name or training word during the review process after performing intensive training.

[0278] As another example, if the input from a patient performing association training is incorrect or there is no input from the patient for a preset period of time, the processor can provide the patient with sensory images related to the incorrect answer entered by the patient as real-time hint feedback.

[0279] For example, when a patient performs association training using general training words, if the patient fails to recall "apple" during the training sequence "A monkey's butt is red," "If it's red, it's an apple," and "Apples are delicious," the processor can provide the patient with real-time hint feedback in the form of visual and gustatory imagery such as "What is red and delicious?"

[0280] In addition, the processor can provide hint feedback to the patient, such as "What is red and delicious?", even if the patient cannot recall "apple" during the review process after performing association training.

[0281] Meanwhile, when a patient performs association training using training words related to a commercial item, if the patient is unable to recall a word that is sensorially associated from the commercial item name "Patent Ramen" or an image containing the commercial item "Patent Ramen," the processor can provide real-time hint feedback to the patient that can recall visual, gustatory, and tactile images, such as "What are the noodles of the ramen like?" or "What are the broth of the ramen like?"

[0282] In addition, similarly during the review process after performing association training, if the patient is unable to recall a word that is sensorially associated with the commercial item name "Patent Ramen" or an image containing the commercial item "Patent Ramen," the processor can provide the patient with hint feedback that can recall visual, gustatory, and tactile images, such as "What are the noodles of the ramen like?" or "What are the broth of the ramen like?"

[0283] As another example, if the input from a patient performing combined training is incorrect or there is no input from the patient for a preset period, the processor can provide the patient with the content of a story related to the incorrect answer entered by the patient as real-time hint feedback.

[0284] For example, when a patient performs associative training using general training words, and during associative training where the patient creates a story using the words "cherry blossom" and "ramen" and the emotion "sadness," if the patient responds, "I went to the Han River, saw the cherry blossoms, and ate ramen, and I felt sad," the processor can provide the patient with real-time hint feedback to naturally continue the story, such as "Why did I feel sad after eating ramen?"

[0285] In addition, the processor can provide hint feedback to the patient during the review process after performing associated training, such as, "What did you eat while looking at the cherry blossoms at the Han River and spill all of it?" even if the patient cannot remember what they ate at the Han River.

[0286] Meanwhile, when a patient performs associated training using training words related to commercial items, if the patient does not remember the correct answer to the question regarding the name of the commercial item, such as "What is the name of the ramen that a person is eating in the image?" provided along with an image containing the commercial item, the processor can provide real-time hint feedback to the patient using training words that help remember the name of the commercial item, such as "What is the name of the ramen with chewy noodles and spicy red broth?"

[0287] In addition, even if the patient is unable to answer a question during the review process after performing associated training, the processor can provide the patient with hint feedback using training words that help remember the name of a commercial item, such as, "What is the name of the ramen with chewy noodles and spicy red broth?"

[0288]

[0289] Meanwhile, if the unit training provided to the patient is associative training, the processor can provide real-time feedback to the patient in a manner different from the method described above.

[0290] For example, the processor can use a feedback provision model to provide feedback to the patient according to a pre-set feedback rule based on a first-type input word or a second-type input word included in the patient's input determined to be correct or incorrect.

[0291] For example, a pre-set feedback rule may mean a rule that, when the patient's input is determined to be correct, provides the patient with a second cognitive training sentence using a second-type stored word or a second-type generated word corresponding to a first-type input word, and when the patient's input is determined to be incorrect, provides the patient with a hint sentence using a first-type generated word corresponding to a second-type input word.

[0292] For example, if the processor determines the patient's input to be correct, it may generate a second cognitive training sentence for performing the next cognitive training based on the first type of input words included in the patient's input and provide the second cognitive training sentence to the patient as feedback. For another example, if the processor determines the patient's input to be incorrect, it may generate a hint sentence for the patient to input the correct answer to the first cognitive training sentence based on the second type of input words included in the patient's input and provide the hint sentence to the patient as feedback.

[0293] First, explain how to provide feedback to the patient when the processor determines the patient's input to be the correct answer.

[0294] For example, the processor can provide the patient with a second cognitive training sentence as the next cognitive training based on determining the patient's input as the correct answer. That is, based on the result of judging the patient's input as correct, the processor can provide praise feedback to the patient and generate a second cognitive training sentence to provide as the next problem.

[0295] As one example, the processor may derive a Type 2 stored word corresponding to a Type 1 input word based on determining the patient's input as the correct answer, and provide a Type 2 cognitive training sentence to the patient based on the Type 1 input word and the Type 2 stored word. As another example, the processor may obtain a Type 2 generated word based on the Type 1 input word from an associative sentence generation model based on determining the patient's input as the correct answer, and provide a Type 2 cognitive training sentence to the patient based on the Type 1 input word and the Type 2 generated word.

[0296] That is, the processor can generate a second cognitive training sentence based on the patient's input being determined as the correct answer, and provide the second cognitive training sentence to the patient as feedback, that is, as the next problem in association training.

[0297] FIGS. 11a and FIGS. 11b are drawings illustrating an example of a method for providing feedback to a patient based on determining the patient's input as the correct answer according to one embodiment.

[0298] Hereinafter, with reference to FIGS. 11a and 11b, an example of a method for providing feedback to a patient based on the processor determining the patient's input as the correct answer will be described.

[0299] First, referring to FIG. 11a, the processor can derive a second type storage word (1130) corresponding to a first type input word (1120) included in the patient's input (1110).

[0300] The processor can derive a second type stored word (1130) corresponding to a first type input word (1120) based on determining the patient's input (1110) as the correct answer.

[0301] For example, the processor may determine whether the first type input word (1120) is stored in the database based on determining the patient's input (1110) as the correct answer. Here, the database is a space capable of storing certain data and may include a certain storage device, a server, or the memory described in FIG. 2.

[0302] Additionally, the processor can derive a second type stored word (1130) corresponding to the first type input word (1120) from the database based on the determination that the first type input word (1120) is stored in the database.

[0303] For example, the processor may derive a second type stored word (1130) corresponding to a first type input word (1120) from a database based on a pre-set feedback rule. More specifically, the processor may derive a second type stored word (1130) from a database that does not overlap with the second type input word, and may derive the second type stored word (1130) based on a pre-set feedback rule to derive a second type word (1130) that corresponds to a characteristic of the first type input word (1120) that the patient can easily perceive, such as the shape, color, movement, taste, sound, or texture, among the second type words that modify the first type input word (1120). Additionally, the pre-set feedback rule may be set so that even if a second-type word modifies a first-type input word (1120), a second-type word that symbolically or metaphorically modifies the first-type input word (1120) is not derived as a second-type stored word (1130).

[0304] Meanwhile, referring to FIG. 11b, the processor can derive a second type generated word (1131) corresponding to the first type input word (1121) using an associative sentence generation model (1140) based on the determination that the first type input word (1121) is not stored in the database. That is, the processor can derive a second type generated word (1131) corresponding to the first type input word (1121) according to a pre-set feedback rule to derive the second type generated word (1131) using the associative sentence generation model (1140) when the first type input word (1121) is not stored in the database.

[0305] For example, the processor may obtain a second type generated word (1131) based on a first type input word (1121) from an associative sentence generation model (1140) based on the patient's input (1110) being determined as the correct answer. That is, the associative sentence generation model (1140) may refer to a model that has been trained to generate a second type generated word (1131) corresponding to the first type input word (1121) as output data, using the first type input word (1121) as input data.

[0306] Accordingly, if the first type input word (1121) is not stored in the database, the processor can obtain a second type generated word (1131) corresponding to the first type input word (1121) using an associative sentence generation model (1140). Additionally, the processor can store the obtained second type generated word (1131) in the database and, in subsequent cognitive training, use the second type generated word (1131) stored in the database as a second type stored word (1130) to provide feedback to the patient.

[0307] Meanwhile, the second type stored word (1130) and the second type generated word (1131) may include at least one second type word corresponding to the first type input word (1121). More specifically, the second type stored word (1130) and the second type generated word (1131) may mean at least one second type word that modifies the first type input word (1121), and with reference to FIG. 11a and FIG. 11b, the second type stored word (1130) and the second type generated word (1131) may include second type words such as "red," "sour," "sweet," and "round" that modify "apple," which is the first type input word (1121).

[0308] Accordingly, the processor can generate a second cognitive training sentence for performing the next cognitive training using a second type stored word (1130) stored in a database or a second type generated word (1131) derived using an associative sentence generation model (1140).

[0309] The processor can determine, according to a pre-set feedback rule, which of the second type stored word (1130) or the second type generated word (1131) does not overlap with the second type input word included in the patient's input (1110).

[0310] More specifically, the processor may determine, based on a pre-set feedback rule, whether there exists a Type 2 word among the Type 2 stored words (1130) or Type 2 generated words (1131) that corresponds to the Type 2 input word "red" included in the patient's input (1110). Additionally, if there exists a Type 2 word "red" among the Type 2 stored words (1130) or Type 2 generated words (1131) that corresponds to "red," the processor may randomly select any one of the remaining Type 2 words excluding that Type 2 word, i.e., "red." However, not limited thereto, the processor may select a Type 2 word suitable for the patient among the Type 2 stored words (1130) or Type 2 generated words (1131) based on the patient's age, gender, cognitive function, previous cognitive training results, difficulty of current cognitive training, etc.

[0311] Accordingly, the processor can provide a second cognitive training sentence to the patient based on the selected second type word and the first type input word (1120, 1121).

[0312] As one example, the processor may provide a second cognitive training sentence to the patient based on a first type input word (1120) and a second type stored word (1130). As another example, the processor may provide a second cognitive training sentence to the patient based on a first type input word (1121) and a second type generated word (1131).

[0313] More specifically, if the processor selects "round" from among the second type stored words (1130) or second type generated words (1131) that does not overlap with the second type input word "red," it can generate a second cognitive training sentence using the first type input word (1120) "apple" and the selected second type word "round," and provide the generated second cognitive training sentence to the patient.

[0314] For example, the processor can generate a second cognitive training sentence such as “Apples are round. Now, please tell me what comes to mind when you say round” using the first type input word (1120) “apple” and the selected second type word “round”, and can provide the generated second cognitive training sentence as feedback to the patient.

[0315] Meanwhile, the processor may perform at least one of the operations included in the method of providing feedback based on determining the above-described patient input (1110) as the correct answer by using an associative sentence generation model. For example, the associative sentence generation model may refer to a model that has been trained to generate a second cognitive training sentence as output data by using a first type input word (1120, 1121) included in the patient input (1110) as input data. That is, the processor may perform at least one of the operations included in the method of providing feedback based on determining the above-described patient input (1110) as the correct answer by using an associative sentence generation model, even for operations that are not explicitly described as using an associative sentence generation model.

[0316] First, explain how to provide feedback to the patient when the processor determines the patient's input to be incorrect.

[0317] Based on determining the patient's input as an incorrect answer, the processor obtains at least one first-type generated word corresponding to a second-type input word from a hint generation model, and can provide a hint sentence to the patient based on association information of at least one first-type generated word and the second-type input word.

[0318] FIG. 12 is a diagram illustrating an example of a method for providing feedback to a patient based on determining the patient's input as an incorrect answer according to one embodiment.

[0319] Hereinafter, with reference to FIG. 12, an example of a method for providing feedback to a patient based on the determination that the processor has determined the patient's input to be incorrect will be described.

[0320] Referring to FIG. 12, the processor can derive at least one first type generated word (1240) corresponding to a second type input word (1230) included in the patient's input (1210).

[0321] Based on determining that the patient's input (1210) is incorrect, the processor can obtain at least one first-type generated word (1240) corresponding to a second-type input word (1230) according to a pre-set feedback rule from a hint generation model.

[0322] For example, if the patient's input (1210) is incorrect, the processor may generate a first-type generated word (1240) that can be modified by a second-type input word (1230) included in the patient's input (1210), derive association information of a word among the first-type generated words (1240) that does not overlap with the first-type input word, generate a hint sentence, and provide the hint sentence to the patient according to a pre-set feedback rule to provide the generated hint sentence to the patient. Additionally, the pre-set feedback rule may be configured not to derive association information that is too abstract for the patient to associate, or too specific for the patient to associate too easily. Here, association information may refer to various types of information related to the word, such as the word's category, definition, example sentence, initial consonant, etc., and this is merely an example and is not limited thereto.

[0323] For example, the processor can use a second type input word (1230) as input data for a hint generation model to derive at least one first type generation word (1240) as output data. That is, the hint generation model may refer to a model that has been trained to generate a first type word corresponding to the second type input word (1230), that is, a word that the second type input word (1230) can modify.

[0324] Accordingly, the processor can obtain at least one first-type generated word (1240), such as “apple,” “rose,” “lips,” “fire truck,” etc., which “red mask” can modify, based on a second-type input word (1230), such as “red mask,” included in the patient’s input (1210).

[0325] Additionally, the processor can determine any one of the generated first type words (1240) that does not overlap with the first type input word “tiger” included in the patient’s input (1210).

[0326] The processor can provide a hint sentence to the patient based on association information of at least one first type generated word (1240) and a second type input word (1230).

[0327] For example, the processor may determine at least one first-type word among the first-type generated words (1240) that does not overlap with the first-type input word, and derive association information of the determined first-type word. Additionally, the processor may generate a hint sentence using the association information of the derived first-type word and the second-type input word (1230), and provide the generated hint sentence as feedback to the patient.

[0328] More specifically, if the processor selects "apple" among at least one first-type generated word (1240) that does not overlap with the first-type input word "tiger," it can derive "fruit" as the associated information (more specifically, the category among the associated information) of "apple." Additionally, the processor can generate a hint sentence such as "What are some red fruits?" using "fruit," which is the associated information of "apple," and the second-type input word "red." Thus, the processor can provide feedback to the patient by providing the generated hint sentence. Meanwhile, for the sake of convenience of explanation, only the method of generating a hint sentence using the category among the associated information of the first-type generated word (1240) has been described; however, as described above, the method of generating a hint sentence using the definition, example sentence, initial consonant, etc. of the first-type generated word (1240) included in the associated information can be applied in the same way.

[0329] The processor may obtain input for a first cognitive training sentence again from a patient who has been provided with a hint sentence. Additionally, if the input obtained again from the patient is correct, the processor may generate a second cognitive training sentence and provide it to the patient in the manner described above with reference to FIGS. 11a and 11b, thereby enabling the patient to perform cognitive training. However, the processor may stop the cognitive training if the patient fails to input the correct answer for the first cognitive training sentence or the second cognitive training sentence despite being repeatedly provided with hint sentences.

[0330] In addition, the processor can generate a hint sentence based on the patient's hint request, not only when the patient's input (1210) is incorrect, and provide the generated hint sentence to the patient. That is, the processor can continuously generate a hint sentence based on the patient's input and provide the hint sentence to the patient.

[0331] More specifically, the processor determines whether the patient's input includes expressions requesting a hint, such as "what," "I don't know," "hint," or "well," and if such expressions are included, it can provide a hint sentence to the patient in the manner described above.

[0332] Meanwhile, the processor may use a hint generation model to perform at least one of the operations included in the method of providing feedback based on determining the above-described patient input (1210) as an incorrect answer. For example, the hint generation model may refer to a model that has been trained to generate a hint sentence as output data using a second type input word (1230) included in the patient input (1210) as input data. That is, the processor may use the hint generation model to perform at least one of the operations included in the method of providing feedback based on determining the above-described patient input (1210) as an incorrect answer, even for operations that are not explicitly described as using the hint generation model.

[0333] Meanwhile, the processor can provide final feedback after performing the cognitive training program by utilizing the patient's input obtained while performing the cognitive training program.

[0334]

[0335] FIG. 13 is a flowchart illustrating an example of a method for providing final feedback to a patient according to one embodiment. FIG. 14 is a diagram illustrating an example of a method for deriving recall evaluation data according to one embodiment.

[0336] Hereinafter, with reference to FIG. 13, an example of a method in which a processor provides final feedback to a patient is described, and with reference to FIG. 14, an example of a method in which a processor derives recall evaluation data to provide final feedback to a patient is described.

[0337] First, referring to FIG. 13, in step 1310, the processor can derive evaluation data for each of a plurality of parameters related to cognitive function based on the result data of the patient performing a cognitive training program. Here, the plurality of parameters related to cognitive function may include recall, voice, words, sentence structure, description, and emotion parameters.

[0338] The processor can derive evaluation data for each of multiple parameters based on the patient's response to evaluation questions or result data from performing the cognitive training program after the patient performs the cognitive training program, and provide the derived evaluation data to the patient as final feedback.

[0339] The processor can derive evaluation data for recall parameters.

[0340] Referring to FIG. 14, the processor can provide a cognitive training evaluation to the patient based on the training words (1410) provided in the cognitive training.

[0341] For example, the processor can provide a cognitive training evaluation to a patient who has received feedback. In other words, the processor can provide a cognitive training evaluation to a patient who has performed cognitive training while receiving real-time feedback.

[0342] For example, the processor can provide a recall evaluation problem to the patient and derive evaluation data for recall parameters based on the patient's input to the provided recall evaluation problem. Here, the recall evaluation problem may include words trained in a cognitive training program performed by the patient.

[0343] For example, the processor can derive evaluation data for recall parameters by providing the patient with a recall evaluation problem regarding whether they can recall words trained in the cognitive training program even after a predetermined period has elapsed since the time the patient performed the cognitive training program.

[0344] For example, the processor may use a basic language model to provide a first evaluation of recalling the training words (1410) provided in the cognitive training in the order in which the cognitive training was performed, and a second evaluation of recalling the training words (1410) provided in the cognitive training based on clues regarding the training words (1410). More specifically, the processor may provide a first evaluation of recalling the training words (1410) in the order in which the training words (1410) were provided in the association training, and a second evaluation of recalling the training words (1410) based on clues regarding the training words (1410) provided in the association training.

[0345] The processor may provide a recall evaluation to the patient based on all training words (1410) provided to the patient performing cognitive training. Here, the recall evaluation may mean an evaluation to determine how well the patient performing cognitive training can recall the training words (1410).

[0346] For example, the processor may use a basic language model to provide a first evaluation of recalling the training words (1410) provided in the cognitive training in the order in which the cognitive training was performed, and a second evaluation of recalling the training words (1410) provided in the cognitive training based on clues about the training words.

[0347] First, the processor can provide the patient with a first evaluation of recalling the training words (1410) provided in the cognitive training in the order in which the cognitive training was performed.

[0348] For example, the processor may provide a first evaluation problem to the patient to recall the training words (1410) in the order in which the training words (1410) were provided in cognitive training. Additionally, the processor may obtain the patient's input regarding the first evaluation problem and determine whether the patient's input is correct or incorrect.

[0349] More specifically, the processor can generate a first evaluation problem as output data by using training words (1410) used in cognitive training as input data for the basic language model. For example, the processor can provide the patient with a first evaluation problem such as, "Please tell me the training words (1410) that you exchanged with me in order during cognitive training." Additionally, the processor can obtain the patient's input regarding the first evaluation problem and determine whether the patient's input is correct or incorrect. For example, if the processor obtains the patient's input such as, "A monkey's butt is red, if it is red it is an apple, and an apple is round," the processor can determine the patient's input as correct. For another example, if the processor obtains the patient's input such as, "A monkey's butt is red, if it is red it is an apple, and so on," the processor can determine the patient's input as incorrect. In addition, the processor can determine the patient's input as incorrect not only when the patient's input does not contain all training words (1410), but also when there is no input from the patient within a set time, or when the order is incorrect.

[0350] The processor may provide the patient with a second evaluation of recalling the training word (1410) provided in cognitive training based on a clue for the training word (1410).

[0351] For example, the processor may generate a second evaluation problem containing clues for the training word (1410) and provide the generated second evaluation problem to the patient.

[0352] More specifically, the processor can generate a second evaluation problem as output data by using the training words (1410) used in cognitive training as input data for the basic language model. For example, the processor can provide the patient with a second evaluation problem such as, "Among the training words (1410) that you exchanged with me during cognitive training, what is red and round?" Additionally, the processor can obtain the patient's input regarding the second evaluation problem and determine whether the patient's input is correct or incorrect. As an example, if the processor obtains the patient's input "apple," it can determine the patient's input as correct. As another example, if the processor obtains the patient's input "grapefruit," it can determine the patient's input as incorrect.

[0353] Therefore, the processor can evaluate whether the patient has faithfully performed the cognitive training by providing the patient with the first assessment problem and / or the second assessment problem. Additionally, the processor may provide the second assessment problem after providing the patient with the first assessment problem, but is not limited thereto; conversely, the processor may provide the first assessment problem after providing the patient with the second assessment problem.

[0354] In addition, the processor can provide recall evaluations to the patient at different times.

[0355] As one example, the processor may provide the patient with a recall assessment problem immediately after the patient performs a cognitive training program and derive recall assessment data regarding the patient's short-term memory using the patient's input. As another example, the processor may provide the patient with a recall assessment problem after a predetermined amount of time has elapsed (e.g., one day later) after the patient performs the cognitive training program and derive recall assessment data regarding the patient's long-term memory using the patient's input.

[0356] For example, the processor can use the patient's input on a recall evaluation problem to extract the retention rate and recall error types of words learned by the patient in a cognitive training program, and adjust the training difficulty based on the extracted retention rate and recall error types.

[0357] For example, the processor can determine the difficulty of cognitive training based on the results of the cognitive training evaluation. In other words, the processor can determine the difficulty of cognitive training based on the results of the recall evaluation.

[0358] In other words, the processor can adjust the difficulty of cognitive training by increasing or decreasing the number of times cognitive training is performed based on the results of the recall evaluation.

[0359] For example, the difficulty of cognitive training may increase as the number of times the training is performed increases and decrease as it decreases, and the processor can adjust the difficulty of the cognitive training by adjusting the number of times the training is performed based on the patient's recall evaluation results. In other words, the difficulty of the cognitive training can be adjusted according to the number of times the processor provides cognitive training sentences to the patient.

[0360] In addition, the processor can adjust the difficulty of cognitive training by determining the difficulty of the training words. For example, the processor determines the difficulty of the training words to be provided in cognitive training based on the abstractness of the training words themselves and the frequency of use of the training words included in the cognitive training evaluation results, and can adjust the difficulty of the cognitive training using the determined training words. That is, the higher the abstractness of the training words themselves and the lower their frequency of use, the higher the difficulty of the corresponding training words can be determined—i.e., they can be considered difficult training words; and the more difficult (or easier) the training words are, the more difficult (or easier) the cognitive training can become.

[0361] In addition, the processor can adjust the difficulty of cognitive training based on factors such as the frequency of providing hints—specifically, hint sentences—to the patient and the specificity of the hints. For example, the processor can adjust the difficulty of cognitive training by changing the frequency of providing hints to the patient. That is, the difficulty of cognitive training may become harder (or easier) as the frequency of providing hints to the patient increases (or decreases). Furthermore, the processor can adjust the difficulty of cognitive training by controlling the degree of specificity of the hints provided to the patient. That is, the difficulty of cognitive training may become harder (or easier) as the specificity of the hints decreases (or increases). Here, specificity can be determined by how specific the description of the training word is; for instance, if the training word is "tiger" and the hint is "beast," the specificity is determined to be low, while if the hint is "a beast with stripes and yellow and black fur," the specificity is determined to be high.

[0362] More specifically, if the patient's inputs for both the first and second assessments provided to the patient during a pre-set period are determined to be correct, the processor may increase the number of times cognitive training sentences are provided. Additionally, if the patient's inputs for both the first and second assessments provided to the patient during a pre-set period are determined to be incorrect, the processor may decrease the number of times cognitive training sentences are provided. Meanwhile, if at least one of the patient's inputs for both the first and second assessments provided to the patient during a pre-set period is determined to be incorrect, the processor may maintain the number of times cognitive training sentences are provided at the current level.

[0363] Meanwhile, the processor may use a basic language model to perform at least one of the operations included in the method of providing a cognitive training assessment to the patient described above. That is, the basic language model may refer to a model that has been trained to generate a cognitive training assessment problem, i.e., a recall assessment problem, as output data using training words provided in cognitive training as input data. Additionally, the basic language model may refer to a model that has been trained to derive a result determining whether the patient's input is correct or incorrect as output data using the patient's input regarding the recall assessment problem provided as input data.

[0364] For example, the processor can extract a retention rate regarding how well a patient remembers words learned in a cognitive training program. Additionally, the processor can determine whether the patient cannot remember the words learned in the cognitive training program at all (i.e., storage failure) or remembers them incorrectly (i.e., retrieval failure), and extract the type of recall error of the patient based on the determination.

[0365] In addition, as described above, the processor inputs the patient's input regarding the recall evaluation problem as input data for the recall evaluation model, and can obtain recall evaluation data such as the retention rate and recall error type as output data for the recall evaluation model.

[0366] Therefore, the processor can derive recall evaluation data including the patient's word retention rate, recall error types, etc.

[0367] The processor can derive evaluation data for voice parameters.

[0368] For example, the processor can derive evaluation data for voice parameters based on result data of a cognitive training program performed by the patient or the patient's input regarding a voice evaluation problem. Here, the voice evaluation problem may include words trained in the cognitive training program performed by the patient.

[0369] For example, the processor can determine whether there is confusion between "ㄱ" (Giyeok) and "ㅋ" (Kieuk) by producing a result of comparing phonemes between the patient's voice input included in the result data of a cognitive training program performed by the patient, or between the patient's voice input regarding voice evaluation problems and a native speaker's voice input. In addition, the processor can detect speech distortion (dysarthria) using time-frequency analysis (STFT, Mel-spectrogram).

[0370] In addition, the processor can use the patient's voice input to measure the time between the input of a specific word and the input of another specific word, and use the measured time to derive voice evaluation data regarding the patient's speech fluency and rhythm. That is, if the time between the input of a specific word and the input of another specific word exceeds a preset time, the processor can determine that the patient's cognitive load has increased, and if the patient fails to input an appropriate word, the processor can determine that the patient has agnosia.

[0371] In addition, as described above, the processor inputs the patient's input regarding cognitive training program result data or voice evaluation problems as input data for a voice evaluation model, and can obtain voice evaluation data such as speech fluency and rhythm as output data for the voice evaluation model.

[0372] Therefore, the processor can derive voice evaluation data including the patient's speech fluency, rhythm, etc.

[0373] The processor can derive evaluation data for word parameters.

[0374] For example, the processor can derive evaluation data for word parameters based on the result data of a patient's cognitive training program.

[0375] For example, the processor can derive voice evaluation data by determining whether a patient performing a cognitive training program used appropriate words during the training process. Specifically, the processor can derive evaluation data indicating low vocabulary richness if the patient fails to recall the correct word and remains unresponsive, or if the patient responds using pronouns instead of the correct word. In other words, the processor can derive word evaluation data indicating low vocabulary richness if the correct word is "dog" but the patient provides no input, or if the patient inputs using a pronoun, such as "it ate rice," instead of the correct word, "the dog ate rice."

[0376] In addition, as described above, the processor inputs the patient's input regarding the patient's cognitive training program result data or word evaluation problem as input data for the word evaluation model, and can obtain recall evaluation data such as vocabulary richness as output data for the word evaluation model.

[0377] Therefore, the processor can derive word evaluation data including the patient's vocabulary richness, etc.

[0378] The processor can derive evaluation data for sentence structure.

[0379] For example, the processor can derive evaluation data for sentence structure parameters based on result data of a cognitive training program performed by the patient or the patient's input regarding a sentence structure evaluation problem. Here, the sentence structure evaluation problem may include words trained in the cognitive training program performed by the patient.

[0380] For example, the processor can determine whether the patient's input regarding the result data of a cognitive training program performed by the patient or a sentence structure evaluation problem contains a specific sentence structure. Here, the specific sentence structure is a sentence structure that occurs primarily in patients with cognitive impairments such as dementia, and may include sentence structures that use repetitive expressions, contain grammatical errors, or have a broken narrative flow.

[0381] For example, the processor can derive sentence structure evaluation data related to the degree of cognitive decline in a patient if the patient's input consists of sentence structures where only words with short distances are repeated, or sentence structures of similar forms are repeated. Additionally, the processor can input the patient's input as input data for a sentence structure evaluation model, calculate the distances between words included in the input, or identify the relationships between subjects, predicates, objects, complements, attributive modifiers, and adverbs included in the input to calculate the similarity of repeated sentences, thereby deriving sentence structure evaluation data related to the patient's cognitive decline as output data.

[0382] For example, the processor can derive sentence structure evaluation data by determining whether the relationships between subject, predicate, object, complement, attributive, and adverbial phrases in the patient's input are grammatically correct. Additionally, the processor can input the patient's input as input data for a sentence structure evaluation model and derive the patient's sentence structure evaluation data as output data by identifying whether the patient's input corresponds to tense and speaker, and whether there are errors in the topic role within the context.

[0383] For example, the processor can derive sentence structure evaluation data by determining whether the patient's input progresses in chronological order. In other words, the processor can derive the patient's sentence structure evaluation data by determining whether the patient's input flows chronologically according to the narrative structure and whether the flow of the story between the preceding and succeeding sentences is correctly connected.

[0384] In addition, the processor inputs the patient's input as input data for a sentence structure evaluation model, calculates the distance between words included in the patient's input, or identifies the relationships between subjects, predicates, objects, complements, attributive modifiers, adverbs, etc. included in the patient's input to calculate the similarity of repeated sentences, thereby deriving sentence structure evaluation data related to the patient's cognitive decline as output data.

[0385] In addition, as described above, the processor inputs the patient's cognitive training program result data as input data for a sentence structure evaluation model, and determines whether there are repetitive expressions in the patient's input, whether there are grammatical errors, and whether the flow of the story is natural, thereby obtaining sentence structure evaluation data as output data.

[0386] Therefore, the processor can derive sentence structure evaluation data.

[0387] The processor can derive evaluation data for the description.

[0388] For example, the processor can derive evaluation data for description parameters based on result data of a cognitive training program performed by the patient or the patient's input regarding a description evaluation problem. Here, the description evaluation problem may include words trained in the cognitive training program performed by the patient.

[0389] For example, the processor can derive description evaluation data by determining whether the characteristics of words trained in a cognitive training program or words included in a description evaluation problem are correctly described.

[0390] For example, the processor can provide the patient with a problem to describe a specific word as a description evaluation problem, and derive description evaluation data by obtaining the patient's input regarding the color, shape, smell, taste, surrounding environment, etc., of the specific word. Specifically, the processor can derive description evaluation data by providing the patient with a problem to describe a butterfly, obtaining the patient's input "a yellow butterfly flutters around the flower," and determining whether there is an error in the patient's input.

[0391] In addition, the processor inputs the patient's input as input data for a description evaluation model and determines whether the patient's input correctly describes the provided word, thereby obtaining description evaluation data as output data. That is, the processor can determine whether the patient correctly described the provided word by considering factors such as whether the patient's input correctly described the color of the provided word, whether the basic form of the word was correctly described, whether vividness was enhanced by adding supporting elements or expressions that match the word, whether hints were used, the time taken to the first response, and the total time spent on training.

[0392] In addition, the processor can derive descriptive evaluation data by acquiring not only the patient's text or voice input regarding the provided words, but also images drawn directly by the patient.

[0393] The processor can derive evaluation data for sentiment parameters.

[0394] For example, the processor can derive evaluation data for sentiment parameters based on the result data of a cognitive training program performed by the patient or the patient's input regarding a voice evaluation problem. Here, the sentiment evaluation problem may include words trained in the cognitive training program performed by the patient.

[0395] For example, the processor can derive emotional evaluation data by determining whether the patient's input regarding the results of a cognitive training program performed by the patient or a voice evaluation problem accurately reflects the presented emotion, whether the emotion is consistently maintained, and whether the characteristics of the voice match the content of the story.

[0396] Specifically, the processor can determine whether the patient's input adequately reflects the presented emotion based on whether adjectives, adverbs, and predicates expressing emotion included in the patient's input are sufficiently used, and whether the presented words and the words expressing the patient's emotion emotionally match. Additionally, the processor can determine whether the emotion in the patient's input is consistently maintained based on whether the emotion in the patient's input is maintained consistently or changes without logical context. Furthermore, regarding the patient's voice input, the processor can determine whether the patient's voice characteristics match the content of the story based on whether the pitch and speech rate of the patient's voice match the emotion of the story—that is, whether a sad story is told in a bright and cheerful tone, or whether emotionally important words are spoken in the same tone as other unimportant words.

[0397] In addition, as described above, the processor inputs the patient's input regarding the result data of the cognitive training program or the emotion evaluation problem as input data for the emotion evaluation model, and can obtain the result of determining whether the patient's input accurately reflects the presented emotion, whether the emotion is consistently maintained, and whether the characteristics of the voice match the content of the story as output data for the emotion evaluation model.

[0398] Therefore, the processor can derive sentiment evaluation data.

[0399] Referring again to FIG. 13, in step 1320, the processor can provide evaluation data to the patient as final feedback.

[0400] For example, the processor can synthesize acquired recall evaluation data, voice evaluation data, word evaluation data, sentence structure evaluation data, descriptive evaluation data, and sentiment evaluation data to generate final feedback data, and provide the generated final feedback data to the patient.

[0401] In addition, the processor can provide the patient with final feedback data, including the patient's cognitive ability assessment, word usage, and training performance evaluation results, along with the respective evaluation data.

[0402] In addition, as described above, the processor can determine the cognitive training program to be provided to the patient using the final feedback data. That is, the processor can determine the unit training and training words to be provided to the patient using the final feedback data, and can also determine the difficulty level of the unit training and training words.

[0403] For example, the processor can determine the patient's unit training and training words by utilizing the training performance evaluation results, the patient's cognitive ability evaluation results, and the patient's word usage results included in the final feedback data. Additionally, the processor can determine the words to be provided to the patient as training words among various words related to the commercialization item by utilizing the training performance evaluation results, the patient's cognitive ability evaluation results, and the patient's word usage results included in the final feedback data. For example, the processor can determine words suitable for the patient among words related to the commercialization item as training words based on the difficulty level of the unit used by the patient, the patient's cognitive ability evaluation results, etc.

[0404] Meanwhile, the processor may use a general sentence generation model to perform at least one of the operations included in the method of providing cognitive training to the patient described above. That is, the general sentence generation model may refer to a model that has been trained to provide a response to the patient's request as output data, using various inputs from the patient, such as sentence re-request inputs and rule queries, as input data. In other words, the general sentence generation model may refer to a model that has been trained to provide a response to the patient's request overall while the patient performs cognitive training.

[0405] In the method described above in the present disclosure, a device that provides a cognitive training program to a patient can communicate in real time and provide the cognitive training program to the patient.

[0406] In addition, the patient can receive real-time feedback from the device providing the cognitive training program while performing the cognitive training program, and can also receive final feedback after performing the cognitive training program.

[0407] In addition, the patient's cognitive function can be effectively improved by providing a cognitive training program tailored to the patient's cognitive abilities.

[0408] In addition, patients can perform cognitive training anytime and anywhere through a device that provides cognitive training regardless of location.

[0409] In addition, by using an artificial intelligence model to analyze emotional words and sentence contexts included in the patient's input, accurate feedback on the patient's input can be provided.

[0410] In addition, by providing cognitive training to patients based on minimal rules, it may be easier to guide the patient's associative flow toward the planning direction of the cognitive training compared to cases where only artificial intelligence models are used.

[0411] In addition, a device that provides cognitive training using the patient and the commercial item name can communicate in real time and provide unit training to the patient.

[0412] In addition, by providing cognitive training using the names of commercial items—familiar words encountered in real life—patient interest can be easily stimulated, enabling patients to actively participate in the cognitive training.

[0413] In addition, by providing cognitive training to patients using the name of a commercial item, the company of the commercial item, etc., can obtain a natural promotional effect.

[0414] In addition, the patient can receive real-time feedback while performing unit training from a device that provides cognitive training using commercial item names, and can also receive final feedback after performing unit training.

[0415] Meanwhile, the above-described method can be written as a program executable on a computer and can be implemented on a general-purpose digital computer that operates the program using a computer-readable recording medium. In addition, the structure of the data used in the above-described method can be recorded on a computer-readable recording medium through various means. The computer-readable recording medium includes storage media such as magnetic storage media (e.g., ROM, RAM, USB, floppy disk, hard disk, etc.) and optical reading media (e.g., CD-ROM, DVD, etc.).

[0416] A person skilled in the art related to the present embodiment will understand that it may be implemented in modified forms without departing from the essential characteristics of the description above. Therefore, the disclosed methods should be considered in an illustrative rather than a restrictive sense, and the scope of rights is defined in the claims rather than the description above, and should be interpreted to include all differences within the scope of equivalence.

Claims

1. A step of determining a cognitive training program to be provided to the patient based on the patient's prior data; and The method includes the step of providing feedback based on the degree of performance of the cognitive training program by the patient. The above feedback includes real-time feedback provided during the execution of the cognitive training program and final feedback provided after the execution of the cognitive training program. Method of providing cognitive training.

2. In Paragraph 1, The above dictionary data is, A method comprising data evaluating the cognitive function of the patient, medical data of the patient, and data on the patient's previous cognitive training program performance.

3. In Paragraph 1, The above-mentioned determining step is, A step of determining unit training in the area of ​​reduced cognitive ability of the patient based on the above prior data; and A method comprising the step of determining a training word to be used in the unit training based on the word usage of the patient.

4. In Paragraph 1, The steps provided above are, A step of determining whether the input of the patient performing the above cognitive training program is the correct answer; and A method comprising the step of providing real-time feedback based on the result of determining whether the patient's input is correct.

5. In Paragraph 4, The above-mentioned judgment step is, A step of extracting input words included in the input of the patient and calculating the phoneme distance and semantic distance of the input words; and A method comprising the step of determining whether the patient's input is correct based on the phoneme distance and the semantic distance.

6. In Paragraph 4, The steps provided above are, A method comprising the step of providing real-time hint feedback to the patient based on the result of determining that the patient's input is not the correct answer.

7. In Paragraph 1, The steps provided above are, A step of deriving evaluation data for each of a plurality of parameters related to cognitive function based on the result data of the patient performing the cognitive training program; and A method comprising the step of providing the evaluation data to the patient as the final feedback.

8. A computer-readable recording medium storing a program for executing the method of claim 1 on a computer.

9. Memory in which at least one program is stored; and It includes at least one processor that executes the above at least one program, and The above-mentioned at least one processor is, A cognitive training program to be provided to the patient is determined based on the patient's prior data, and feedback is provided based on the degree of the patient's performance of the cognitive training program. The above feedback includes real-time feedback provided during the execution of the cognitive training program and final feedback provided after the execution of the cognitive training program. A device that provides cognitive training.

Images