KR20210037231A - Apparatus and method for assessing language level using artificial intelligence - Google Patents

Abstract

The present invention relates to an apparatus and a method for diagnosing and assessing the Korean language level of a child based on artificial intelligence to generate a professional personalized curriculum. To this end, the apparatus for diagnosing and assessing the Korean language level of a child based on artificial intelligence comprises: a diagnosis problem output module which selects a specific phoneme based on an appearance frequency in a general corpus, and outputs a diagnosis problem for the selected specific phoneme to a child learning application module of a learner client who is a client of a learner; an embedding module which receives diagnosis problem solving information which is an input of the learner for the diagnosis problem from the learner client, generates achievement information for the specific phoneme based on a distribution for the entire learners of the diagnosis problem solving information, and embeds the achievement information in a vector for the specific phoneme; and a level information generation module which includes an artificial neural network module for outputting prediction achievement information which is a prediction value of the achievement information for the entire phonemes except for the specific phoneme with the embedding information as an input vector, and merges the achievement information for the specific phoneme and the prediction achievement information for the rest of the entire phonemes to generate level information which is achievement information for the entire phonemes.

Description

Apparatus and method for assessing language level using artificial intelligence}

The present invention relates to an artificial intelligence-based diagnostic evaluation apparatus and method for each level of Hangul for infants.

As the education enthusiasm for various languages such as Korean, English, Japanese, and Chinese is overheating, recent parents have started to educate various languages such as Korean and English as early as 4 years old. In particular, in the case of Hangeul, which is the core of early education, according to statistics from the Korea Institute for Curriculum and Evaluation, the rate of Hangul education before enrollment reached 91.6%.

As for the existing Korean language education method, the method of teaching using materials such as Hangeul study papers during the caring time of parents or caregivers became the mainstream, and it was gradually shifting to a method of teaching online or offline through applications such as smartphones or tablets. . However, this Hangul education method was either a low-cost one-way Hangul education method according to a preset curriculum that is not related to the Hangul level of young children, or an expensive interactive Hangul education method that includes the guidance of a teacher at the place of study.

Because of this, when it is necessary to conduct Korean language education for infants at a low cost, even if the Korean language education is performed using an application, the difficulty is too high or too low to induce interest of infants. Or, there was a problem that the caregiver had to spend a lot of education time on Korean language education for young children. In addition, it was very difficult for parents and caregivers to choose which one would be most appropriate among the over 4,000 Korean language education workbooks and numerous Korean language education applications sold on the market.

In addition, even if the teacher of the workbook visits and takes the expensive interactive Korean language education method, most of the workbook teachers teach 15 minutes a week, and there is a problem that parents are responsible for teaching Korean language in the rest of the time.

(Patent Document 1) Korean Patent Registration 10-1921888, Online Evaluation-Based Language Rehabilitation Training Method, Korea (National Rehabilitation Center) (Patent Document 2) Korean Patent Laid-Open Publication No. 10-2017-0108622, English speaking prosody and fluency evaluation device and method thereof, Electronics and Telecommunications Research Institute (Patent Document 3) Korean Patent Publication 10-2017-0087367, Translingual Semantic Web Information Quality Evaluation Method, Korea Advanced Institute of Science and Technology (Patent Document 4) Registered Korean patent, 10-1943520, Automatic evaluation system of English speaking pronunciation of non-native speakers and evaluation method using it, Korea University of Foreign Studies Research Industry-Academic Cooperation Foundation

Accordingly, an object of the present invention is an artificial intelligence-based infant Hangul level diagnostic evaluation device for performing level-specific diagnostic evaluation for infants' Hangul proficiency using artificial intelligence, and creating a professional 1:1 personalized curriculum based on this And to provide a method.

Hereinafter, specific means for achieving the object of the present invention will be described.

An object of the present invention is to select a specific phoneme based on the frequency of appearance in the General Corpus of Hangul, and output a diagnosis problem for the selected specific phoneme to an early childhood learning application module of a learner client, which is a learner's client. module; The learner client receives diagnosis problem solving information, which is the learner's input for the diagnosis problem, and generates achievement level information for the specific phoneme based on the distribution of the diagnosis problem solving information to all learners, and the specific An embedding module for embedding the achievement level information in a vector for a phoneme; And an artificial neural network module for outputting predicted achievement information, which is a predicted value of achievement information for all phonemes other than the specific phoneme, as an output vector by using the embedding information as an input vector, and the achievement information for the specific phoneme and the A level information generation module for generating level information, which is achievement level information for all phonemes, by merging the predicted achievement information for the remaining whole phonemes; and wherein the artificial neural network module of the level information generation module comprises: Provides an artificial intelligence-based diagnostic evaluation device for each level of Hangul for infants, characterized in that it is a pre-learned artificial neural network using the achievement information for phonemes as an input vector and the achievement information for each of the remaining phonemes as an output vector. Can be achieved by

Another object of the present invention is a memory module for storing a diagnostic evaluation program code for each infant Hangul level, which is connected to a learner client, which is a client of a learner installed with an infant learning application module, to perform diagnostic evaluation for each infant Hangul level; And a processing module for processing the diagnostic evaluation program code for each level of Hangul for infants; wherein the diagnostic evaluation program code for each level of Hangul for children selects a specific phoneme based on the frequency of appearance in the General Corpus of Hangul, and the selected specific phoneme A diagnostic problem output step of outputting a diagnostic problem for a phoneme to the early childhood learning application module of the learner client; The learner client receives diagnosis problem solving information, which is the learner's input for the diagnosis problem, and generates achievement level information for the specific phoneme based on the distribution of the diagnosis problem solving information to all learners, and the specific An embedding step of embedding the achievement level information in a vector for a phoneme; And inputting the embedding information to the artificial neural network module to output predicted achievement information, which is a predicted value of achievement information for all other phonemes other than the specific phoneme, and the achievement information for the specific phoneme and the prediction for the remaining all phonemes. A level information generation step of generating level information, which is achievement level information for all phonemes by merging achievement level information, and is configured to be performed on a computer, and the artificial neural network module of the level information generation module includes the specific phoneme of the existing learners. Provides an artificial intelligence-based diagnosis and evaluation device for each level of Hangul for infants, characterized in that it is an artificial neural network that has been learned by using the achievement information for each as an input vector and the achievement information for each of the remaining phonemes as an output vector. Can be achieved.

Another object of the present invention is to provide a diagnostic problem output module to select a specific phoneme based on the frequency of appearance in the General Corpus of Hangul, and to determine the diagnosis problem for the selected specific phoneme. A diagnostic problem output step of outputting to the module; The embedding module receives diagnosis problem solving information, which is the learner's input for the diagnosis problem, from the learner client, and generates achievement level information for the specific phoneme based on the distribution of the diagnosis problem solving information to all learners. And embedding the achievement level information in a vector for the specific phoneme; And a level information generating module, by inputting the embedding information into an artificial neural network module included in the level information generating module, outputting predicted achievement level information, which is a predicted value of achievement level information for all phonemes other than the specific phoneme, and the specific phoneme Including, the artificial neural network module of the level information generating module, comprising: generating level information, which is achievement level information for all phonemes, by merging the achievement level information for and the predicted achievement level information for the remaining phonemes. Is a pre-learned artificial neural network using the achievement information for the specific phoneme of existing learners as an input vector and the achievement information for each of the remaining phonemes as an output vector. This can be achieved by providing diagnostic evaluation methods for each level of Hangul.

Another object of the present invention is a memory module configured in a learner client, which is a learner's client, and storing an infant Korean language application module installed in the learner client; And a processing module configured in the learner client and processing a program code of the infant Hangul application module; wherein the program code of the infant Hangul application module comprises: diagnostic evaluation for each infant Hangul level based on artificial intelligence according to claim 1 A diagnostic problem receiving step of receiving a diagnostic problem for a specific phoneme in the device; A diagnostic problem solving information generating step of generating diagnostic problem solving information input by the learner with respect to the diagnosis problem; And a diagnostic problem solving information transmitting step of transmitting the diagnosis problem solving information to the artificial intelligence-based diagnosis evaluation apparatus for each child's Hangul level; and configured to be performed on a computer, wherein the artificial intelligence-based diagnosis evaluation apparatus for each child's Hangul level includes It can be achieved by providing an artificial intelligence-based Hangul learning apparatus for young children, characterized in that receiving curriculum information output by level information generated based on the diagnosis problem solving information.

Another object of the present invention is to provide an infant Hangul application module configured in a learner client, which is a learner's client, and installed in the learner client, a diagnosis problem for a specific phoneme in the AI-based diagnostic evaluation apparatus for each level of Hangul for infants according to claim 1 A diagnostic problem receiving step of receiving; A diagnostic problem solving information generation step of generating, by the infant Hangul application module, information about solving a diagnosis problem input by the learner with respect to the diagnosis problem; And transmitting, by the infant Hangul application module, the diagnosis problem solving information to the artificial intelligence-based infant Hangul level-specific diagnostic evaluation device; including, wherein the AI-based infant Hangul level-specific diagnostic evaluation device It can be achieved by providing an artificial intelligence-based Hangul learning method for young children, characterized in that the curriculum information outputted by level information generated based on the diagnosis problem solving information is received.

As described above, the present invention has the following effects.

First, according to an embodiment of the present invention, there is an effect of enabling diagnostic evaluation for each level of Korean proficiency of infants.

Second, according to an embodiment of the present invention, an effect of generating a 1:1 personalized curriculum is generated according to the child's Korean language proficiency. As a result, there is an effect that professional Hangul personalized education can be made inexpensive.

Third, according to an embodiment of the present invention, an effect of selecting an optimal curriculum according to the child's Korean language proficiency is generated.

The following drawings attached to the present specification illustrate preferred embodiments of the present invention, and serve to further understand the technical idea of the present invention together with the detailed description of the present invention, so the present invention is limited only to the matters described in such drawings. And should not be interpreted.
1 is a schematic diagram showing a system for providing learning sessions for each level of Korean language for young children based on artificial intelligence according to an embodiment of the present invention;
2 is a schematic diagram showing an artificial intelligence-based diagnosis and evaluation apparatus for each level of Hangul for infants according to an embodiment of the present invention;
3 is a schematic diagram showing an artificial intelligence-based infant Hangul level diagnosis apparatus 1 according to a modified example of the present invention;
4 is a schematic diagram showing an embedding module 14 according to a modified example of the present invention,
5 is a schematic diagram showing an artificial neural network of the level information generating module 15 according to a modified example of the present invention;
6 is a schematic diagram showing an RNN module of the level information generating module 15 according to the first modified example of the present invention;
7 is a schematic diagram showing a generation module 150 and a classification module 151, which are artificial neural network modules of the level information generation module 15 according to the second modified example of the present invention;
8 is a schematic diagram showing an example of a ConvNet encoder that is a classification module 151 according to a second modified example of the present invention;
9 is a schematic diagram showing an artificial intelligence-based Hangul customized curriculum output device for infants according to an embodiment of the present invention;
10 is a schematic diagram showing an artificial neural network module of the curriculum selection module 20 according to an embodiment of the present invention;
11 is a schematic diagram showing reinforcement learning of the curriculum selection module 20 according to a modified example of the present invention;
12 is a schematic diagram showing a curriculum selection module 20 according to a modified example of the present invention,
13 is a flowchart showing an operation example of the curriculum selection module 20 according to an embodiment of the present invention;
14 is a schematic diagram showing an artificial intelligence-based infant Hangul curriculum detailed adjustment device according to an embodiment of the present invention;
15 is a schematic diagram showing an artificial neural network module of an artificial intelligence-based infant Hangul curriculum detailed adjustment device 3 according to a modified example of the present invention;
16 is a flowchart illustrating a method of providing a learning session for each level of Hangul for children based on artificial intelligence according to an embodiment of the present invention.

Hereinafter, exemplary embodiments in which the present invention can be easily implemented by those of ordinary skill in the art to which the present invention pertains will be described in detail with reference to the accompanying drawings. However, when it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention in describing the operating principle of the preferred embodiment of the present invention in detail, the detailed description thereof will be omitted.

In addition, the same reference numerals are used for portions having similar functions and functions throughout the drawings. Throughout the specification, when a specific part is said to be connected to another part, this includes not only a case in which it is directly connected, but also a case in which it is indirectly connected with another element interposed therebetween. In addition, the inclusion of a specific component does not exclude other components unless otherwise stated, but means that other components may be further included.

In the following description of the invention, a convolutional neural network, which is a neural network using a convolutional product, may be described as CNN, ConvNet, or the like.

In the following description of the invention, Hangul is composed of a consonant (touch sound) and a vowel (single sound) as a single sound (phoneme) character, and the syllable (sound node) consists of the first sound (initial sound), the middle sound (neutral), and the ending sound (last sound). It consists of three sets of single sounds (phonemes), with touch sounds (consonants) used for the first sound (initial sound) and the last sound (ending sound), and a hall sound (vowel sound) for the middle sound (neutral).

In modern Hangeul, there are 14 characters such as ㄱ,ㄴ,ㄷ, ㄹ,ㅁ etc. in the touch sound (consonant) that makes a single note, and there are 10 characters such as ㅏ, ㅑ, ㅓ, ㅕ, etc. in the single sound (vowel). There are 5 characters of ㄲ, ㄸ, ㅃ, ㅆ, ㅉ in the double consonant sound (double consonant) that makes the Gospel, and 11 characters such as ㅐ, ㅒ, ㅔ, ㅖ, etc. In addition, in modern Hangeul, when there is a final sound (jongseong), there is a single base or side base, all touch sounds (consonants) are used on the base, and 13 characters such as ㄲ, ㅆ, ㄳ, and ㄵ are used in the side base.

Modern Hangeul can write 11,172 words (19 first sound × 21 middle sound × (27 ending + 1 no ending sound)) by weaving single characters. Of the 11,172 characters, 399 are non-supporting characters and 10,773 are supporting characters.

According to the language regulations, there are fewer syllables actually used in the modern Korean standard language. The sound of Korean is composed of the first sound + the middle sound (+ the end sound). In the standard language, all 19 touch sounds are used for the first sound, but the ㅇ in the first sound does not sound. End sounds are gathered in 7 segments according to the 7-segment method, and all 8 prongs are added to the one with no ending sound, so the pronunciation of modern Korean is 19 first sound × 21 middle sound × 8 last sound = 3,192 sounds. According to the standard pronunciation method, double vowels ㅑ, ㅒ, ㅕ, ㅖ, ㅛ, ㅠ are short vowels ㅏ, ㅐ, ㅓ, ㅔ, ㅗ, ㅜ, so the first sound is 3 × middle sound. 8 = 144 The middle sound [ㅢ] with the first sound (which comes after the first sound instead of ㅇ) is omitted, except for b (in the case of b, it is recognized as another phoneme according to palatalization.) [ㅣ] Because it sounds as (Hangul Spelling Clause 9 and Standard Pronunciation Clause 5 Clue 3), the first sound 17 × the middle sound 1 × the end sound 8 = 136 sounds are dropped again. Therefore, currently 3192-144-136 = 2,912 types of phonetic measures actually used in the standard Korean language.

1 is a schematic diagram showing a system for providing learning sessions for each level of Korean language for children based on artificial intelligence according to an embodiment of the present invention. As shown in FIG. 1, the system 100 for providing a learning session for each level of Hangul for infants based on artificial intelligence according to an embodiment of the present invention includes an artificial intelligence-based diagnosis and evaluation device for each level of Hangul for infants (1), It may include an apparatus for generating a customized Korean language curriculum for infants (2), and a device for fine-tuning a Korean language curriculum for infants based on artificial intelligence (3). As shown in FIG. 1, the system 100 for providing learning sessions for each level of Korean language for young children based on artificial intelligence according to an embodiment of the present invention is an application module for learning Korean language for young children that is included in the memory module of the learner client and processed by the processing module. It is configured to transmit a diagnosis problem or problem to the device, receive diagnosis problem solving information or problem solving information input by the learner, and transmit curriculum information or detailed curriculum information generated based on diagnosis problem solving information/problem solving information. Can be.

Artificial Intelligence-based Diagnosis and Evaluation Device for Each Hangul Level in Young Children

2 is a schematic diagram showing an artificial intelligence-based diagnosis and evaluation apparatus for each level of Hangul for infants according to an embodiment of the present invention. As shown in Fig. 2, the AI-based diagnosis evaluation apparatus 1 for each level of Korean alphabet for infants according to an embodiment of the present invention includes a level information classification module 10, a diagnosis criterion setting module 11, and a diagnosis evaluation module. (12) may be included. According to an embodiment of the present invention, the artificial intelligence-based diagnosis and evaluation device 1 for each level of Korean language for infants is performed by a processing module of a computing device such as a specific web server, a virtual server such as a cloud server, and a smart phone, a tablet PC, and a desktop PC. It can be configured to be processed and stored in a memory module of each device.

The level information classification module 10 outputs a diagnosis problem, information on the number of incorrect answers of the diagnosis problem solving information input by the learner by solving the diagnosis problem, information on the input time for each phoneme of the learner, and a preset label (consonant problem) of the diagnosis problem. , Vowel problem, syllable problem, word problem) as input information, classifying the learner's level, and outputting level information. Level information, which is output information of the level information classification module 10, may be composed of a plurality of labels including a first level, which is a Hangul learning start level, a second level, which is a character recognition level, and a third level, which is a word recognition level.

The diagnostic criterion setting module 11 is a module that generates diagnostic criterion information by setting a diagnosis criterion of Korean language level based on the level information classified by the level information classification module 10. The diagnostic criteria set according to each level information are set differently by the diagnostic criteria setting module 11, and when a learner is classified as a Hangul learning start level, consonants, short vowels, and double vowels (ㅑ, ㅕ, ㅛ, ㅠ) It is set as the first diagnosis criterion, which is the diagnosis criterion for, and if the learner is classified by the level of character recognition, the first diagnosis criterion and diconsonants, double vowels (ㅘ, ㅟ, ㅝ, ㅚ, ㅖ, ㅙ, ㅢ, ㅒ, ㅞ) and the second diagnosis criterion, which is a diagnosis criterion for the support, may be configured to be set as a third diagnosis criterion, which is a criterion in which the ratio of syllables is increased from the second diagnosis criterion when the learner has a word recognition level.

The diagnosis evaluation module 12 is a module that generates a diagnosis item based on the diagnosis criteria information set in the diagnosis criteria setting module 11, and evaluates a learner by measuring the achievement level for each phoneme/syllable based on the diagnosis item. . Achievement for each phoneme/syllable according to an embodiment of the present invention is index information measuring whether it is possible to find the letter'A' after hearing the sound of'A', and how long it takes to find it. .

In addition, the diagnostic evaluation module 12 may be configured to adjust the difficulty or number of diagnostic items by performing diagnostic evaluation in real time while generating the diagnostic items. The diagnosis evaluation module 12 may be configured to adjust the difficulty of the diagnosis item by adjusting the time limit of the diagnosis item. In addition, the diagnostic evaluation module 12 may be configured to skip a predetermined number of diagnostic items or terminate the diagnosis, when evaluated above a specific level in the performed diagnostic evaluation.

With respect to the artificial intelligence-based infant Hangul level diagnosis apparatus 1 according to a modified example of the present invention, FIG. 3 is a schematic diagram showing the artificial intelligence-based infant Hangul level diagnosis apparatus 1 according to a modified example of the present invention. . As shown in Fig. 3, the artificial intelligence-based infant Hangul level diagnosis apparatus 1 according to a modified example of the present invention includes a diagnosis problem output module 13, an embedding module 14, and a level information generation module 15. It may include.

The diagnostic problem output module 13 outputs phonemes (elements that produce one phonological segmental sound such as consonants and vowels) as a diagnostic problem to the learner client based on the frequency of appearance in the General Corpus of Hangul, Node. It is a module that outputs a word or a unit element of sound that forms a part of a word and gives the feeling of one synthesized sound) to the learner's client as a diagnosis problem based on the frequency of appearance in the General Corpus of Hangul. For example, when a learner using the corresponding infant Hangul learning application module for the first time accesses the infant Hangul learning application module, phoneme information that is preset or selected based on the frequency of appearance in the General Corpus (for example, ㄱ,ㄴ, It may be configured to output the diagnosis problem to the early childhood Hangul learning application module of the learner client on the basis of ㅏ,ㅓ).

The embedding module 14 vectorizes all phonemes with a text information vectorization technique (Bag of Words, n-gram, TF-IDF, Word2Vec), and a vector corresponding to the diagnosis problem among all phonemes. It is a module that generates embedding information by merging achievement level information for. The achievement level information according to an embodiment of the present invention is based on the distribution of the phoneme input time/number of incorrect answers of all learners of the diagnosis problem solving information (phoneme input time, incorrect answer count, etc.) input by the learner client based on the corresponding diagnosis problem. It can mean information calculated by. For example, when learner A has a phoneme input time of 1.5 seconds (diagnosis problem solving information) for the phoneme'ㄱ', and this phoneme input time represents the distribution of the top 40% of all learners, learner A's achievement level information Can be generated as 0.6. Thereafter, the embedding module 14 may be configured to generate embedding information obtained by merging the achievement level information 0.6 into the vector of'a' of the learner A.

4 is a schematic diagram showing an embedding module 14 according to a modified example of the present invention. The embedding module 14 is a three-dimensional matrix (or initial-consonant, initial-double consonant, neutral-vowel, neutral- A six-dimensional matrix with the axis of the double vowel, final-consonant, final-single, and final-side support) is created, and achievement information for each phoneme of the learner for the phoneme corresponding to the diagnosis problem among all phonemes (or , Diagnosis problem solving information) may be merged to generate embedding information (achievement matrix). Table 1 below is an example showing a part of embedding information (achievement matrix). As shown in Table 1 below, the achievement matrix may be configured to be pair-wise for all phonemes.

Example achievement matrix 0 G N C 0 0 0.8 0.5 0.4 ㅏ 0.5 0.2 0.3 0.1 ㅑ 0.6 0.5 0.8 0.4 ㅓ 0.7 0.2 0 0

The level information generation module 15 predicts achievement level information for all phonemes other than the phoneme of the diagnosis problem based on the embedding information generated by the embedding module 14 to generate predicted achievement level information, and recalls the remaining all phonemes. It is a module that generates level information, which is information in which prediction achievement information is merged.

Regarding the prediction of the achievement level information for the remaining whole phonemes of the level information generation module 15, FIG. 5 is a schematic diagram showing an artificial neural network of the level information generation module 15 according to a modified example of the present invention. As shown in Fig. 5, the level information generation module 15 uses an input achievement matrix, which is achievement level information (or diagnosis problem solving information) for some phonemes of existing learners, as input information, and the achievement level for each of the remaining phonemes. A previously learned artificial neural network may be included by using information (or diagnosis problem solving information) as output information. In particular, the artificial neural network module of the level information generation module 15 is composed of a plurality of nodes in the output layer to correspond to each of the remaining phonemes, and outputs prediction achievement information (or diagnosis problem solving information) for each phoneme. It is configured so that the weight of each node can be learned by back propagation based on the difference (error) with the actual achievement level information. The predicted achievement information (or diagnostic problem solving information) for each of the remaining phonemes output from the artificial neural network of the level information generating module 15 is input to the embedding module 14 to generate an output achievement matrix, and the level information generating module (15) may be configured to merge the output achievement matrix and the input achievement matrix to generate level information including achievement information (or diagnostic problem solving information) for the entire phoneme.

Regarding the first modification of the prediction of the achievement level information for the remaining whole phonemes of the level information generation module 15, the level information generation module 15 includes achievement level information (or diagnosis problem solving information) for some phonemes of the corresponding learner. Recurrent Neural Network (RNN) or LSTM that sequentially generates predicted achievement information (or predicted value of diagnostic problem-solving information) for neighboring phonemes (or related phonemes) based on the input achievement matrix of) and outputs the output achievement matrix. (Long-Short Term Memory) may be included. 6 is a schematic diagram showing an RNN module of the level information generating module 15 according to the first modified example of the present invention. As shown in FIG. 6, the RNN module of the level information generation module 15 uses an input achievement matrix, which is achievement level information (or diagnosis problem solving information) for a specific phoneme of learners, as input information, and surrounding phonemes (or, It can be composed of a previously learned RNN using the output achievement matrix including achievement level information for related phonemes) as output information. The hidden vector of the current time t state, which is the hidden layer of the RNN module of the level information generating module 15 according to the first modified example of the present invention, is the state of the previous time t-1 as shown in Equation 1 below ( state) and can be configured to be updated as a function of the input vector.

In the above equation, h _t is the hidden vector of time t, W _hh is the weight of the hidden vector h _{t-1 of the previous time t-1, w xh} is the weight of the input vector x _t of time t, and b _h is It can mean a constant.

The surrounding phoneme or related phoneme according to the first modification of the present invention is a phoneme corresponding to a pixel adjacent to a pixel corresponding to a specific phoneme on a matrix having an initial sound (initial sound), a middle sound (neutral), and an end sound (ending sound) as axes. Or, when vectorized using a text information vectorization technique (Bag of Words, n-gram, TF-IDF, Word2Vec), it may mean a phoneme having a close mutual distance. According to the first modified example of the present invention, level information may be generated in the following steps.

(1) Create a 3D matrix (input achievement matrix) that includes achievement level information (or diagnostic problem solving information) for some phonemes of the learner and centers on the first sound (initial sound), middle sound (neutral), and ending sound (final sound). . In particular, the location of the achievement level information (or diagnosis problem solving information) for some phonemes is configured in the direction in which the convolution product of the 3D matrix proceeds.

(2) An input achievement matrix including achievement level information for some phonemes is input into the RNN module of the level information generation module 15. At this time, the RNN module constructs a masked kernel (masked convolution, masked filter) in which the future pixel to be predicted is masked in order to prevent the data of the future pixel to be predicted from being input as input information. That is, input the input vector x _{t to the RNN module}

(3) The RNN module predicts the achievement level information for the surrounding phoneme (or related phoneme), which is the pixel to be predicted, and predicts the achievement level information for the surrounding phoneme (or related phoneme) (or the predicted value of the diagnosis problem solving information). ) Output (y _t ) and update the input achievement matrix including the prediction achievement information (or a predicted value of the diagnosis problem solving information).

(4) The RNN module is updated based on the state of time t, h _{t, and the input vector x t} , which constitutes the kernel masked in the input achievement matrix, and the updated input achievement matrix is fed back to the updated RNN module.

(5) Repeat 1 to 4 to generate an output achievement matrix, which is predicted achievement information (or predicted value of diagnosis problem solving information) for the remaining phonemes, and level information that is a 3D matrix in which the input achievement matrix and the output achievement matrix are merged. Print

According to the first modification of the present invention, there is an effect that the learning speed based on the level information of all the phonemes of existing learners is relatively fast, and there is an effect of being able to track predicting the achievement of each phoneme. . In addition, there is an effect that learning can be performed stably.

Regarding the second variation of the prediction of the achievement level information for the remaining whole phonemes of the level information generation module 15, the level information generation module 15 includes achievement level information (or diagnosis problem solving information) for some phonemes of the corresponding learner. ), which is a module that generates level information including the output achievement matrix, which is the predicted achievement information (or the predicted value of the diagnosis problem solving information) for the remaining phonemes, based on the input achievement matrix, and generated by the corresponding generation module. It may include a classification module that is an encoder that learns the generation module by classifying whether the determined level information is similar to the level information on the entire phoneme of the existing learners. 7 is a schematic diagram showing a generation module 150 and a classification module 151, which are artificial neural network modules of the level information generation module 15 according to the second modified example of the present invention. As shown in FIG. 7, the level information generation module 15 may include a generation module 150 and a classification module 151.

The generation module 150 may be composed of an image generator composed of encoders and decoders such as VAE and GAN. In the inference step of the generation module 150, an input achievement matrix, which is achievement level information (or diagnosis problem solving information) for some phonemes of the learner, is received as input information, and predictions for all phonemes excluding the some phonemes. And generating an output achievement matrix, which is a vector including achievement information (or predicted value of diagnostic problem solving information), and merging the input achievement matrix and the output achievement matrix to generate level information, which is an achievement matrix for the entire phoneme.

In the learning step (training) of the generation module 150, the generation module 150 is learned by the classification module 151, and the generation module 150 provides information on the achievement level (or diagnosis problem solving information) on some phonemes of the existing learner. Receives an input achievement matrix, which is a vector containing ), as input information, generates an output achievement matrix, which is a vector containing predicted achievement information (or predicted value of diagnosis problem solving information) for all phonemes excluding the partial phonemes, and the input When the level information is generated by merging the achievement matrix and the output achievement matrix, the level information generated by the classification module 151, which is a ConvNet encoder, is classified (classified) as to whether the level information is similar to the level information of the existing learners. When the level information generated by the module 150 is classified as not a matrix similar to the level information of the existing actual learners in the classification module 151, the level information generated by the generation module 150 and the classification module ( In 151), a method of updating the generation module 150 by outputting an error based on the difference (cross entropy, etc.) of existing actual learners' level information (for example, back propagation) is used. The generation module 150 is learned by the output value of ). According to this, as the learning arrangement proceeds, the generation module 150 outputs a matrix similar to the level information of the existing actual learners.

8 is a schematic diagram showing an example of a ConvNet encoder that is a classification module 151 according to a second modified example of the present invention. As shown in FIG. 8, the ConvNet encoder, which is the classification module 151, may be constructed as [INPUT-CONV-RELU-POOL-FC]. In the case of the generated level information, which is input information, when the depth is composed of the first sound (initial sound), the middle sound (neutral), and the end sound (last sound), the number of branches of the initial voice is 19 characters, the number of branches of the neutrality is 21 characters, and the branches of the last sound. In order to contain 27 characters, the level information of the INPUT vector is 32 horizontally, vertically 32, and the initial-neutral-final channel, and the size of the input can be composed of a matrix of [32x32x3]. The CONV layer (Conv. Filter, 101) is connected to some areas of level information, and the dot product of this connected area and its own weight is calculated. The resulting volume will have the same size as [32x32x12]. The RELU layer is an activation function applied to each element, such as max(0,x). The RELU layer does not change the volume size ([32x32x12]). As a result, Activation map 1 (102) is generated. The POOL layer (pooling, 103) performs downsampling on the "horizontal, vertical" dimension to output a reduced volume (Activation map 2, 104) such as [16x16x12]. The fully-connected (FC) layer 106 connected to the n-th activation map n 105 calculates class scores and outputs an output layer 107 having a size of [1x1x1]. The output layer 107 outputs whether the level information generated and input by the generation module 150 is classified as similar to the level information of the existing learners or is classified as not similar to the level information of the existing learners. The FC layer is connected to all elements of the previous volume.

In this way, the ConvNet of the classification module 151 transforms the level information, which is the original matrix consisting of pixel values for the achievement level corresponding to a specific phoneme, into a class score (similarity with the level information of existing learners) through each layer. ). Some layers have parameters, while others do not. In particular, CONV/FC layers are activation functions that include not only input volume but also weight and bias. On the other hand, RELU/POOL layers are fixed functions. The parameters of the CONV/FC layer are learned with gradient descent so that the class score for each matrix is the same as the label of the matrix.

The parameters of the CONV layer consist of a series of learnable filters. Each filter is small in the horizontal/vertical dimension, but covers the entire depth in the depth dimension. In the forward pass, each filter is slid (convolved precisely) to the horizontal/vertical dimensions of the input volume, and an n-dimensional activation map is created. When sliding the filter over the input, a dot product is performed between the filter and the input volume. Through this process, ConvNet learns a filter that activates for a specific pattern at a specific location of the input information. The stacking of these activation maps in the depth dimension becomes the output volume. Therefore, each element of the output volume handles only a small area of the input, and neurons in the same activation map share the same parameters as the result of applying the same filter.

According to the second modified example of the present invention, it is possible to predict the achievement level of the corresponding learner relatively high, and the achievement level of the remaining phonemes can be predicted with high accuracy even with a diagnosis problem composed of several phonemes.

Artificial intelligence-based Hangul customized curriculum generation device for infants

9 is a schematic diagram showing an artificial intelligence-based Hangul customized curriculum output device for infants according to an embodiment of the present invention. As shown in FIG. 9, the apparatus 2 for generating a customized Korean curriculum for infants based on artificial intelligence according to an embodiment of the present invention includes a curriculum selection module 20, a repetition frequency adjustment module 21, and an area ratio adjustment module. (22), it may include a curriculum output module (23). According to an embodiment of the present invention, the apparatus 2 for generating a customized Korean curriculum for children based on artificial intelligence is provided by a specific web server, a virtual server such as a cloud server, and a processing module of a computing device such as a smartphone, a tablet PC, and a desktop PC. It can be configured to be processed and stored in a memory module of each device.

The curriculum selection module 20 is a module that generates curriculum information by selecting a curriculum suitable for the current level of the learner in the entire curriculum based on the achievement level for each phoneme/syllable generated by the diagnostic evaluation module 12. Curriculum information according to an embodiment of the present invention includes phoneme information, repetition count information, area ratio information (initial, neutral, final, consonant, double consonant, etc.) Area), information about the learning session (problem) including information on the properties such as content type information, speed information, time limit information, learning order information, etc., which are educational methods such as games, images, and videos. I can.

The repetition count adjustment module 21 adjusts the repetition count information by adjusting the repetition count of the curriculum based on the curriculum information selected by the curriculum selection module 20 and the achievement level information (or level information) for each phoneme/syllable. It is a module to create. The repetition count adjustment module 21 according to an embodiment of the present invention sets the number of repetitions to a predetermined number of times when the achievement level information for each phoneme/syllable for the curriculum selected by the curriculum selection module 20 is relatively high. A module that adjusts to a relatively low number of times, and adjusts the number of repetitions to a relatively higher number of times than a preset number when the achievement level information per phoneme/syllable for the curriculum selected by the curriculum selection module 20 is at a relatively low level to be.

The area ratio adjustment module 22 adjusts the area ratio of the corresponding curriculum based on the curriculum information selected by the curriculum selection module 20 and the achievement level information (or level information) for each phoneme/syllable. It is a module to create. The area ratio adjusted by the area ratio adjustment module 22 is the ratio between the areas to be output in the curriculum, such as vowels, consonants, final consonants, double vowels, double consonants, initials, neutrals, and endings. It can be used as a probability weight.

The curriculum output module 23 adjusts the curriculum information generated by the curriculum selection module 20 based on the repetition number information and area ratio information generated by the repetition number adjustment module 21 and the area ratio adjustment module 22. It is a module that outputs problem information according to the adjusted curriculum information.

With respect to the curriculum selection module 20 of the artificial intelligence-based Hangul customized curriculum generation device 2 for infants according to an embodiment of the present invention, FIG. 10 is a diagram of a curriculum selection module 20 according to an embodiment of the present invention. It is a schematic diagram showing an artificial neural network module. As shown in Fig. 10, the artificial neural network module of the curriculum selection module 20 according to an embodiment of the present invention includes level information received in a modified example of the diagnosis evaluation device 1 for each level of Korean language for infants based on artificial intelligence. Phoneme information, which is a vector for the phoneme or syllable in question to be output in the curriculum is input information, and the rate of change of the expected level information after the learner learns the curriculum based on the phoneme information (achievement for each phoneme) It can be composed of an artificial neural network using the sum of the rate of change of information) as output information. In the learning step of the curriculum selection module 20 according to a modified example of the present invention, the level information of the existing learner and phoneme information, which is a vector composed of at least one phoneme, are used as input information, and the rate of change of the level information expected as the output information and the actual The artificial neural network can be trained by back propagation of the difference (error) in the rate of change of level information.

According to the artificial neural network module of the curriculum selection module 20 according to an embodiment of the present invention, the rate of change of level information expected when specific phoneme information is used as a curriculum is output for level information of a specific learner, and the curriculum selection module ( 20) is configured to output the expected rate of change of level information for a plurality of phoneme information composed of various phoneme/syllable combinations, and to generate curriculum information based on at least one phoneme information outputting the rate of change of level information above a specific value. Can be.

The curriculum selection module 20 according to an embodiment of the present invention may specifically generate curriculum information in the following steps.

(1) Generate multiple input vectors by merging specific learner's level information and specific phoneme information

(2) Input a plurality of generated input vectors to the artificial neural network module of the curriculum selection module 20

(3) Level information change rate output for each of a plurality of phoneme information

(4) Select at least one phoneme information for which the rate of change of level information above a specific value is output

(5) Generate curriculum information based on selected phoneme information

(6) After the curriculum is over, update the learner's level information based on the learner's achievement level information (using moving average, etc.)

(7) Start over from step 1 based on the updated level information

According to the curriculum selection module 20 according to an embodiment of the present invention, an effect of generating a learning curriculum based on phonemes and syllables that can most increase the achievement of a learner is generated for each curriculum.

Regarding the curriculum selection module 20 of the artificial intelligence-based Hangul customized curriculum generating device 2 for infants according to a modified example of the present invention, FIG. 11 is a reinforcement learning of the curriculum selection module 20 according to a modified example of the present invention. It is a schematic diagram showing. As shown in FIG. 11, the curriculum selection module 20 according to a modified example of the present invention may be configured to update the selection of phoneme information of the curriculum by reinforcement learning. In terms of reinforcement learning, the objective of the curriculum selection module 20 according to a modified example of the present invention is to improve the amount of change in level information after learning of the curriculum, and the state and environment are the level information of the corresponding learner. It can mean (achievement information for each phoneme), action (Action) can mean phoneme information selection of the curriculum selection module 20, and reward (Reward) is curriculum information based on the selected phoneme information. It can mean the amount of change in the learner's level information after learning. According to a modified example of the present invention, when the phoneme information selection of the curriculum selection module 20 is learned by reinforcement learning, the learner's level information is improved more as the learners learn the curriculum selected by the curriculum selection module 20. There is an effect of being able to select a curriculum that can be made. That is, in order to improve the curriculum selection module 20, a lot of curriculum learning information is not necessarily required for one learner, and even when curriculum learning information is generated for several learners, the curriculum selection module 20 can be reinforced learning. There will be.

The curriculum selection module 20 according to a modified example of the present invention may be configured to update the phoneme information selection of the curriculum by more effective reinforcement learning by the following configuration. 12 is a schematic diagram showing a curriculum selection module 20 according to a modified example of the present invention. As shown in FIG. 12, the curriculum selection module 20 according to a modified example of the present invention includes a value network 211, which is an artificial neural network that learns a value function that outputs value in a specific state, and each attribute of curriculum information. A policy network 210 for learning a policy function that outputs a probability may be included, and the policy network 210 and the value network 211 according to a modified example of the present invention include a curriculum selection module 20 that predicts a rate of change of level information. ) Can be configured to be connected to the artificial neural network module. In addition, the policy network 210 and the value network 211 may be connected to an optimal curriculum search module to output selected curriculum attribute information 500.

The policy network 210 is an artificial neural network that determines the probability of specific attributes (phoneme information, repetition count information, area information, learning time information, content type information, etc.) selected in each state of the curriculum selection module 20. And, by learning the policy function, the selected curriculum attribute probability is output. The cost function of the policy network may be a function obtained by multiplying the policy function and the cost function of the value network to calculate the cross entropy and then taking the policy gradient. For example, it may be configured as in Equation 2 below. The policy network can be back propagated based on the product of the cross entropy and the time difference error, which is a cost function of the value network.

In Equation 1, π is a policy function, θ is a policy network parameter, π _θ (a _{i │s i} ) is the probability of performing a specific action (curriculum of a specific attribute) in the current episode, V is a value function, and w is a value network. Parameter, s _i is the status information of the current episode i, S _i+1 is the status information of the next episode i+1, r _i+1 is the reward expected to be acquired in the next episode, and V _w (s _i ) is The possibility of compensation in the current episode, V _w (s _i+1 ), may indicate the possibility of compensation in the next episode, and γ may indicate a depreciation rate. In this case, r _i+1 may be configured to be received by an artificial neural network module that predicts a rate of change of level information. As a result, the policy network 210 according to an embodiment of the present invention initially outputs a curriculum attribute that simulates a learner's curriculum history through a policy gradient.

The policy network 210 according to an embodiment of the present invention provides information on existing curriculum attributes (phoneme information, repetition number information, area information, learning time information, content type information, etc.) and performance according to the existing curriculum attribute information before reinforcement learning proceeds. Based on information (level information change rate), supervised learning is performed, and the weight of the policy network is updated so that the basics of the policy can be learned. That is, the weight of the policy network can be set by supervising learning based on existing curriculum attribute information and performance information. According to this, the policy network can be learned very quickly by recording the existing curriculum.

In addition, according to an embodiment of the present invention, when supervised learning of the policy network 210, it may be configured to perform supervised learning based on existing curriculum attribute information and result information including a random vector. The random vector may use, for example, a Gaussian distribution. Accordingly, there is an effect that the policy network can output challenging curriculum policies with random probability. When supervised learning of the policy network 210 is configured to be supervised learning based on the existing curriculum attribute information and corresponding performance information, the result of the selection of a curriculum being optimized within the existing policy appears. However, according to an embodiment of the present invention, if a random vector is included in the supervised learning of a policy network, the more effective reinforcement learning proceeds, the more effective the policy network can learn a curriculum than the existing policy.

The value network 211 is an artificial neural network that derives the possibility of achieving a reward in each state that the curriculum selection module 20 may have, and learns a value function. The value network 211 presents a direction in which direction the curriculum selection module 20, which is an agent, will be updated. To this end, the input variable of the value network 211 is set as state information, which is information on the state of the curriculum selection module 20, and the output variable of the value network 211 is the curriculum selection module 20 to achieve compensation. It can be set as the possibility of compensation possibility information (a rate of change of level information). Compensation possibility information according to an embodiment of the present invention may be calculated by a Q-function as shown in the following equation.

In Equation 3 above, Q _π may mean information about the total reward possibility expected in the future in the case of a state s and an action a in a specific policy π, R represents a reward for a specific period, and gamma represents a depreciation rate. S _t may represent a state of time t, A _t may represent an action of time t, and E may represent an expected value. Compensation possibility information (Q value) according to an embodiment of the present invention defines the update direction and size of the policy network 210.

At this time, the cost function of the value network may be a mean square error (MSE) function for the value function, and may be configured as in Equation 4 below. The value network 211 may be back propagated based on a time difference error, which is a cost function of the value network.

In Equation 4, V is a value function, w is a value network parameter, s _i is the state information of the current episode i, S _i+1 is the state information _{of the next episode i+1} , and r i+1 is the next episode. A reward expected to be acquired, V _w (s _i ) may mean a reward possibility in the current episode, V _w (s _i+1 ) represents a reward probability in the next episode, and γ represents a depreciation rate. In this case, r _i+1 may be configured to be received by an artificial neural network module that predicts a rate of change of level information.

Accordingly, when the state of the curriculum selection module 20 changes, the value network may update the cost function of Equation 4 in the direction of gradient descent.

According to an embodiment of the present invention, while learning the value network separately from the policy network, the Q value of the value network is supervised instead of starting at random, resulting in an effect of enabling rapid learning. According to this, the effect of being able to greatly reduce the burden of exploration in the action of selecting a combination of curriculum properties with very high complexity occurs.

According to the curriculum selection module 20 according to an embodiment of the present invention, when the policy network 210 that has completed supervised learning selects the curriculum attribute of the current episode i, the value network 211 proceeds with the selected curriculum attribute. It is learned to predict the reward (rate of change of level information). The policy network 210 and the value network 211 of the curriculum selection module 20 that have completed learning are combined with the simulation using the optimal curriculum search module, and finally curriculum properties (phoneme information, repetition count information, area information, learning time). Information, content type information, etc.).

In addition, according to the value network 211 according to an embodiment of the present invention, an effect of updating the policy network that outputs the probability of the selected curriculum attribute can be performed every episode. In the existing reinforcement learning, there is a problem in that the update of the reinforcement learning model proceeds after all episodes are finished, so it was difficult to apply it to the curriculum selection model.

The optimal curriculum search module is a configuration that searches for optimal curriculum properties by conducting multiple simulations for various states and various actions based on a plurality of agents calculated in the policy network and the value network. The optimal curriculum search module according to an embodiment of the present invention may utilize, for example, a Monte Carlo tree search, wherein each node of the tree is a state, and each edge is a specific action for a corresponding state. It represents the expected value accordingly, and it is a structure in which the leaf node expands every time it transitions to a new state by taking a new action with the current state as the root node. In the optimal curriculum search module according to an embodiment of the present invention, when the Monte Carlo tree search is used, the search for the optimal curriculum may be processed in four stages of selection, expansion, evaluation, and backup.

The selection step of the optimal curriculum search module is a step in which the action with the highest value among the selectable actions is selected and proceeded from the current state until the leaf node appears. At this time, the value of the value function stored in the edge and the visit frequency value to balance inquiry-use are used. The equation for action selection in the selection step is as follows.

In Equation 5 above, a _t is an action (curriculum execution) at time t, Q(s _t ,a) is the value of the value function stored in the tree, and u(s _t ,a) is the corresponding state-action pair. This value is inversely proportional to the number of visits, and is used to balance exploration and use.

The expansion step of the optimal curriculum search module is a step of adding a new node as a leaf node by taking action according to the probability of the policy network learned by supervised learning when the simulation proceeds to the leaf node.

In the evaluation step of the optimal curriculum search module, the value of the leaf node is evaluated through the value determined by using the value net from the newly added leaf node (reward possibility) and the reward obtained by proceeding until the end of the curriculum episode using the policy network from the leaf node. It is a step to do. The following equation is an example of evaluating the value of a new leaf node.

In Equation 6 above, V(s _L ) is the value of the leaf node, λ is the mixing parameter, v _θ (s _L ) is the value obtained through the value network, and z _L is the reward obtained by continuing the simulation.

The backup step of the optimal curriculum search module is a step of re-evaluating the value of the nodes visited during the simulation and updating the visit frequency by reflecting the value of the newly added leaf node. The following equation is an example of node value re-evaluation and visit frequency update.

In Equation 7 above, s _L ⁱ denotes the leaf node in the i-th simulation, 1(s,a,i) denotes whether the connection (s,a) was visited in the i-th simulation, and when the tree search is completed, the algorithm is It can be configured to select the most visited connection (s,a) from the root node. According to the optimal curriculum search module according to an embodiment of the present invention, an optimal curriculum attribute (phoneme information, repetition count information, area) is preceded by a plurality of simulations based on the value network for a plurality of curriculum attributes selected by the policy network. Information, learning time information, content type information, etc.) can be selected.

According to an embodiment of the present invention, the curriculum selection module 20 may be configured such that a plurality of agents are configured. When a plurality of agents are configured, the curriculum selected by the curriculum selection module 20 for each of a specific state and a specific curriculum attribute competes with each other, so that the most optimal curriculum can be selected.

13 is a flowchart showing an operation example of the curriculum selection module 20 according to an embodiment of the present invention. As shown in Fig. 13, when the state s(t) is input by the AI-based infant Hangul level diagnosis evaluation device 1 that tracks state information in the infant Hangul learning application module of a plurality of learner clients, the value network ( 211), various curriculum attributes are input to the optimal curriculum search module by a plurality of agents of the policy network 210, and the selected curriculum attribute probability a As the curriculum proceeds by (t), episode t ends and episode t+1 begins. In episode t+1, s(t+1), which is the change in state due to a(t), is inputted by the AI-based infant Hangul level diagnosis evaluation device (1), and the compensation according to a(t), r( t+1) is immediately input to update the value network 211 and the policy network 210.

Artificial Intelligence-based Hangul Curriculum for Preschool Children

14 is a schematic diagram showing an apparatus for fine-tuning a Hangul curriculum for infants based on artificial intelligence according to an embodiment of the present invention. As shown in Figure 14, the artificial intelligence-based infant Hangul curriculum detailed adjustment device 3 according to an embodiment of the present invention, a preference generation module 30, a detailed adjustment module 31, a curriculum output module 32 ) Can be included. The artificial intelligence-based Hangul customized curriculum output device 2 for infants according to an embodiment of the present invention is provided by a processing module of a computing device such as a specific web server, a virtual server such as a cloud server, a smartphone, a tablet PC, and a desktop PC. It can be configured to be processed and stored in a memory module of each device.

The preference generation module 30 is a module that generates preference information for each curriculum based on the number of times the learner has performed the curriculum, the achievement of each curriculum, and the learner's evaluation for each curriculum.

The detailed adjustment module 31 is a module that generates detailed adjustment curriculum information by fine-tuning each attribute of a corresponding curriculum based on the preference information generated by the preference generating module 30. Detailed adjustment according to an embodiment of the present invention may mean phoneme/syllable relearning and reconfirmation of each session, phoneme/syllable omission or skipping, phoneme/syllable order change, repetition frequency adjustment, area ratio adjustment, and the like.

The curriculum output module 32 is a module that outputs fine-tuned curriculum information to a learner client based on the fine-tuned curriculum information generated by the fine-tuning module 31.

Regarding the modification of the output of the detailed adjustment curriculum information of the artificial intelligence-based Hangul curriculum for infants (3), the curriculum information (phoneme information, repetition) generated by the artificial intelligence-based Hangul curriculum generation device (2) Based on the number information, area ratio information, content type information, etc.), after outputting the infant Hangul curriculum to the learner's client's infant Hangul learning application module, problem solving information received by the learner's input (phoneme input time, number of incorrect answers, etc.) ) Or, based on the current achievement level information on the phoneme information of the curriculum, the learner predicts churnability information, which is the possibility of a student's departure, and the curriculum information (phoneme information, repetition count information, area ratio information, content type) Information, etc.) can be configured to fine-tune.

Regarding the prediction of departure possibility information of the artificial intelligence-based infant Hangul curriculum detailed adjustment device 3 according to a modified example of the present invention, FIG. 15 is an artificial intelligence-based infant Hangul curriculum detailed adjustment device according to a modified example of the present invention ( 3) is a schematic diagram showing the artificial neural network module. As shown in Figure 15, the artificial neural network module of the artificial intelligence-based infant Hangul curriculum detailed adjustment device 3 according to a modified example of the present invention is curriculum information generated by the artificial intelligence-based infant Hangul curriculum generating device 2 (Phoneme information, repetition count information, area ratio information, content type information, etc.), problem solving information (phoneme input time, incorrect answer count, etc.), the learner's level information, the learner's level information and the problem-solving information. The current concentration information, which is the difference from the actual achievement level information, can be used as an input vector, and the departure possibility information can be configured as an output vector. In the learning step of the corresponding artificial neural network module, it may be configured to learn the artificial neural network module based on a difference between the output possibility of departure and the actual departure information of the learner.

According to a modified example of the present invention, it may be configured to predict departure probability information at the end of each curriculum, at a specific time, or after a specific number of curriculum learning, and curriculum information (level information) generated by the curriculum generating device 2 Based on the curriculum information that the amount of change is expected to be higher than a certain value), a plurality of candidate curriculum information in which each attribute value (phoneme information, repetition number information, area ratio information, content type information, etc.) It may be configured to include and input into an input vector of an artificial neural network module of the apparatus 3 for fine-tuning the Korean language curriculum for infants 3, and to output information on the possibility of departure corresponding to each candidate curriculum information. The artificial intelligence-based infant Hangul curriculum detailed adjustment device (3) is based on the departure possibility information corresponding to each candidate curriculum information, the departure possibility information is less than a certain value (for example, less than the departure possibility information from the previous curriculum) It may be configured to output at least one of the candidate curriculum information as detailed adjustment curriculum information.

According to a modified example of the present invention, among curriculums that are expected to have a high level of change of level information by the curriculum information generated by the artificial intelligence-based Hangul curriculum generating device 2 for infants, the deviation probability information falls below a specific value. The curriculum is calculated to be low, and the effect is automatically fine-tuned. In addition, according to the current concentration of learners, there is an effect that an optimal curriculum can be output.

Method of providing learning sessions by level of Korean language for young children based on artificial intelligence

Regarding the method of providing a learning session for each level of Korean language for young children based on artificial intelligence according to an embodiment of the present invention, FIG. 16 is a flowchart illustrating a method for providing a learning session for each level of Korean language for children according to an embodiment of the present invention. . As shown in FIG. 16, the method of providing a learning session for each level of Korean language for children based on artificial intelligence according to an embodiment of the present invention includes a diagnosis problem output step (S10), a diagnosis problem solution information reception step (S11), and a achievement level information generation step. (S12), level information generation step (S13), curriculum information generation step (S14), problem solving information receiving step (S15), departure possibility information generation step (S16), fine-tuning curriculum information generation step (S17). I can.

In the diagnosis problem output step (S10), the diagnosis and evaluation apparatus 1 for each level of Hangul for infants based on artificial intelligence outputs a diagnosis problem for a specific phoneme to a learner client.

In the step of receiving diagnosis problem solving information (S11), the AI-based diagnosis evaluation apparatus 1 for each level of Hangul for infants receives diagnosis problem solving information (phoneme input time information, incorrect answer count information, etc.) input for the diagnosis problem from the learner client. This is the receiving step.

The achievement level information generation step (S12) is a step in which the AI-based diagnostic evaluation apparatus 1 for each level of Hangul for infants generates achievement level information for a specific phoneme corresponding to the diagnosis problem.

In the level information generation step (S13), the AI-based diagnosis and evaluation device for each level of Hangul for infants (1) predicts the achievement level information for all the remaining phonemes based on the achievement level information for the specific phoneme generated in S12, and the predicted achievement level information. This is the step of generating level information, which is the achievement level information for the entire phoneme of the learner based on the predicted achievement level information.

In the curriculum information generation step (S14), the curriculum information (phoneme information, repetition number information) for which the level information change is expected to be more than a specific value based on the level information generated in S13 by the artificial intelligence-based Hangul customized curriculum generation device 2 , Area ratio information, content type information, etc.), and outputting the problem to the learner client based on the generated curriculum information. When a learner has previously learned Korean through an early childhood Hangul learning application module, and level information and curriculum information for the corresponding learner are previously stored, it may be configured to be performed from step S14 when the learner logs in.

In the step of receiving problem solving information (S15), the artificial intelligence-based infant Hangul curriculum detailed adjustment device 3 receives problem solving information (phoneme input time information, incorrect answer count information, etc.) input for the problem from a learner client. to be.

In the step of generating departure possibility information (S16), the artificial intelligence-based infant Hangul curriculum fine-tuning device 3 provides information on a plurality of candidate curriculums based on the problem solving information (phoneme input time information, incorrect answer count information, etc.) in S15. This is the step of generating information about the learner's possibility of departure.

In the step of generating detailed adjustment curriculum information (S17), the artificial intelligence-based infant Hangul curriculum detailed adjustment device 3 generates detailed adjustment curriculum information based on the information on the possibility of departure for a plurality of candidate curriculum information in S16. This is the step of outputting the problem to the learner's client based on the detailed curriculum information.

As described above, those skilled in the art to which the present invention pertains will appreciate that the present invention can be implemented in other specific forms without changing the technical spirit or essential features thereof. Therefore, the above-described embodiments are illustrative in all respects and should be understood as non-limiting. The scope of the present invention is indicated by the claims to be described later rather than the detailed description, and all changes or modified forms derived from the meaning and scope of the claims and equivalent concepts should be interpreted as being included in the scope of the present invention.

The features and advantages described herein are not all inclusive, and in particular many additional features and advantages will become apparent to those skilled in the art in view of the drawings, specification, and claims. Moreover, it should be noted that the language used herein has been selected primarily for readability and for teaching purposes, and may not be chosen to describe or limit the subject matter of the invention.

The above description of the embodiments of the present invention has been presented for purposes of illustration. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Those skilled in the art will understand that many modifications and variations are possible in light of the above disclosure.

Therefore, the scope of the invention is not limited by the detailed description, but by any claims in the application on which it is based. Accordingly, the disclosure of the embodiments of the present invention is illustrative and does not limit the scope of the present invention described in the following claims.

1: Artificial intelligence-based diagnosis and evaluation device for each level of Hangul for infants
2: Artificial intelligence-based Hangul customized curriculum generation device for infants
3: Artificial Intelligence-based Hangul Curriculum for Preschool Children Fine-tuning Device
10: level information classification module
11: Diagnostic Criteria Setting Module
12: diagnostic evaluation module
13: diagnostic problem output module
14: embedding module
15: level information generation module
20: Curriculum Selection Module
21: repetition count adjustment module
22: area ratio adjustment module
30: Affinity Generation Module
31: fine tuning module
32: curriculum output module
100: AI-based learning sessions for infants by Korean level
101: Conv. Filter
102: Activation map 1
103: Pooling layer
104: Activation map 2
105: Activation map n
106: Fully connected layer
107: output layer
150: generation module
151: division module
210: policy network
211: value chain
500: Curriculum attribute information

Claims (5)

A diagnosis problem output module for selecting a specific phoneme based on the frequency of appearance in the General Corpus of Hangul, and outputting a diagnosis problem for the selected specific phoneme to an early childhood learning application module of a learner client, which is a learner's client;
The learner client receives diagnosis problem solving information, which is the learner's input for the diagnosis problem, and generates achievement level information for the specific phoneme based on the distribution of the diagnosis problem solving information to all learners, and the specific An embedding module for embedding the achievement level information in a vector for a phoneme; And
And an artificial neural network module for outputting predicted achievement information, which is a predicted value of achievement information for all phonemes other than the specific phoneme, as an output vector by using the embedding information as an input vector, and the achievement information for the specific phoneme and the rest A level information generation module for generating level information, which is achievement level information for all phonemes, by merging the predicted achievement level information for all phonemes;
Including,
The artificial neural network module of the level information generation module is an artificial neural network previously learned by using the achievement level information for the specific phoneme of the existing learners as an input vector and the achievement level information for each of the remaining phonemes as an output vector. Characterized in that,
Artificial intelligence-based diagnosis and evaluation device for each level of Hangul for infants.
A memory module that stores a diagnostic evaluation program code for each level of Korean language for infants, which is connected to a learner client, which is a client of a learner on which the early childhood learning application module is installed, and performs diagnosis evaluation for each level of Korean language in the infants; And
A processing module that processes the diagnostic evaluation program code for each level of the infant Hangul level;
Including,
The diagnostic evaluation program code for each level of Korean language for infants,
A diagnostic problem output step of selecting a specific phoneme based on the frequency of appearance in the General Corpus of Hangul, and outputting a diagnosis problem for the selected specific phoneme to the early childhood learning application module of the learner client;
The learner client receives diagnosis problem solving information, which is the learner's input for the diagnosis problem, and generates achievement level information for the specific phoneme based on the distribution of the diagnosis problem solving information to all learners, and the specific An embedding step of embedding the achievement level information in a vector for a phoneme; And
By inputting the embedding information to the artificial neural network module, the predicted achievement information, which is a predicted value of achievement information for all other phonemes other than the specific phoneme, is output, and the achievement information for the specific phoneme and the predicted achievement for the remaining all phonemes A level information generation step of merging the information to generate level information, which is achievement level information for all phonemes;
It is configured to run on a computer, including,
The artificial neural network module of the level information generation module is an artificial neural network previously learned by using the achievement level information for the specific phoneme of the existing learners as an input vector and the achievement level information for each of the remaining phonemes as an output vector. Characterized in that,
Artificial intelligence-based diagnosis and evaluation device for each level of Hangul for infants.
The diagnostic problem output module selects a specific phoneme based on the frequency of appearance in the General Corpus of Hangul, and outputs the diagnosis problem for the selected specific phoneme to the early childhood learning application module of the learner's client, the learner's client. step;
The embedding module receives diagnosis problem solving information, which is the learner's input for the diagnosis problem, from the learner client, and generates achievement level information for the specific phoneme based on the distribution of the diagnosis problem solving information to all learners. And embedding the achievement level information in a vector for the specific phoneme; And
The level information generation module inputs the embedding information into the artificial neural network module included in the level information generation module to output predicted achievement level information, which is a predicted value of achievement level information for all other phonemes other than the specific phoneme, and outputs prediction achievement level information to the specific phoneme. A level information generation step of generating level information, which is achievement level information for all phonemes, by merging the achievement level information for and the predicted achievement level information for the remaining whole phonemes;
Including,
The artificial neural network module of the level information generation module is an artificial neural network previously learned by using the achievement level information for the specific phoneme of the existing learners as an input vector and the achievement level information for each of the remaining phonemes as an output vector. Characterized in that,
Artificial intelligence-based diagnosis and evaluation method for each level of Hangul for infants.
A memory module configured in a learner client that is a learner's client and storing an infant Hangul application module installed in the learner client; And
A processing module configured in the learner client and processing a program code of the early childhood Hangul application module;
Including,
The program code of the infant Korean language application module,
A diagnostic problem receiving step of receiving a diagnostic problem for a specific phoneme in the artificial intelligence-based diagnostic evaluation apparatus for each Hangul level for infants according to claim 1;
A diagnostic problem solving information generating step of generating diagnostic problem solving information input by the learner with respect to the diagnosis problem; And
A diagnostic problem solving information transmitting step of transmitting the diagnosis problem solving information to the artificial intelligence-based diagnostic evaluation apparatus for each level of Korean characters for infants;
It is configured to run on a computer, including,
Receiving curriculum information output by level information generated based on the diagnosis problem solving information in the artificial intelligence-based diagnosis evaluation apparatus for each level of Korean language for infants,
An artificial intelligence-based Hangul learning device for young children.
A diagnostic problem receiving step in which an infant Hangul application module configured in a learner client, which is a learner's client, and installed in the learner client, receives a diagnosis problem for a specific phoneme in the AI-based diagnostic evaluation device for infant Hangul levels according to claim 1 ;
A diagnostic problem solving information generation step of generating, by the infant Hangul application module, information about solving a diagnosis problem input by the learner with respect to the diagnosis problem; And
Transmitting, by the infant Hangul application module, the diagnosis problem solving information to the AI-based infant Hangul level diagnostic evaluation apparatus;
Including,
Receiving curriculum information output by level information generated based on the diagnosis problem solving information in the artificial intelligence-based diagnosis evaluation apparatus for each level of Korean language for infants,
A method of learning Korean for young children based on artificial intelligence.

Images