KR102636582B1 - Artificial intelligence tutoring system that support diagnosis of learning proficiency and collaborative learning - Google Patents

Abstract

The present invention is an artificial intelligence tutoring system that supports diagnosis of learning comprehension and collaborative learning by remotely connecting the learner's first terminal and the teacher's second terminal, and establishes a network with the first and second terminals and the web and external servers. While communicating, the learner's learning log ( A control unit that matches management data to generate customization information to guide the learner's future learning and the teacher's future teaching, and outputs a preset notification signal according to the customization information; and a notification unit provided in the first and second terminals and visually displayed on an arbitrary screen according to a notification signal output from the control unit, wherein the diagnostic information for each learner includes understanding by learning area and understanding by unit. It may be data about at least one learning competency, and the control unit regenerates the correct answer to the test solved by the learner in each test based on the learning log ( It can be configured to learn to extract a random understanding level that can predict whether the answer is correct, and the customized information is included in at least one learning/instructor customized guide among understanding visualization, recommended content, recommended evaluation questions, and team career. It may include data about

Description

An artificial intelligence tutoring system that supports diagnosis of learning comprehension and collaborative learning {ARTIFICIAL INTELLIGENCE TUTORING SYSTEM THAT SUPPORT DIAGNOSIS OF LEARNING PROFICIENCY AND COLLABORATIVE LEARNING}

The present invention utilizes an existing deep learning-based knowledge tracking AI model based on a transformer structure among deep learning algorithms, so that the output value of the AI model is a value directly inferred from the understanding of the concept, rather than the probability of correcting the next evaluation question. Through this, it is possible to support a more accurate diagnosis of the learner's level of understanding of learning, thereby more efficiently achieving a management environment to guide future customized learning and teaching to learners and teachers, and at the same time, interacting between learners. It is about an artificial intelligence tutoring system that supports diagnosis of learning understanding and collaborative learning to achieve a more interesting learning environment through a configuration that enables collaborative learning through action.

In general, the main characteristics of artificial intelligence (AI) assistants are the convenience of a voice interface that understands and executes the intent of commands uttered by the user and controls peripheral devices, while the artificial intelligence tutoring system requires the intervention of teachers and instructors for learners. The main feature is a context-centered active response that identifies necessary situations and provides diagnosis and prescription appropriate to the situation.

Currently, the domestic edutech (Education Technology) market, which includes such artificial intelligence tutoring technology, is expanding, but its growth rate is lower than the global market and the proportion of small businesses is high.

For example, while sales are concentrated in a few large businesses, sales of small businesses, which account for half of the total number of businesses, are only 1.6%, creating polarization within the industry. To solve this, large edutech companies are starting existing edutech companies. By acquiring businesses, we are launching services using artificial intelligence technology and robot technology, and we are seeking new opportunities by directly discovering and supporting promising edutech startups.

Accordingly, edtech startups are growing in various fields such as customized learning services, game-based learning, foreign language education, and coding education based on technological capabilities. They analyze big data such as learner achievement and learning history to diagnose learning ability and provide personalized learning. The use of AI (artificial intelligence) to provide is most noticeable.

Meanwhile, custom learning technologies that have been commercialized in the past either rely on connections between learning contents or are limited to diagnosing weak concepts by simply counting the number of incorrect problems, so there is a problem in that the diagnosis of the learner's learning ability is not sophisticated.

In addition, with the current technology, in the future management of the learner's learning results, an enhanced artificial intelligence learning process based on deep learning technology is implemented by reflecting the learner's level of understanding (understanding) and tendency for each learning area/unit. And there is no learning/teaching guide system that has gone through an accurate analysis process.

In other words, even when trying to provide personalized learning outcome management, there were limitations in securing the efficiency and reliability of the customized learning system by being based on inaccurate learning competency diagnosis, and in addition, the personalized evaluation and diagnosis process It had the disadvantage of being somewhat low in terms of learning interest.

Accordingly, there is an increasing need for a personalized learning outcome management service that solves the problems of the prior art and provides more accurate learning ability diagnosis technology and the effectiveness of improving learning interest.

Republic of Korea Patent Publication No. 10-2056822 (Title of invention: Method and device for providing learning services)

Therefore, the present invention was created to solve the above problems, and utilizes an existing deep learning-based knowledge tracking AI model based on a transformer structure among deep learning algorithms, so that the output value of the AI model is not the probability of answering the next evaluation question. A more accurate diagnosis of the learner's learning understanding can be supported through a configuration in which the understanding of the concept is directly inferred, and accordingly, a management environment to guide learners and teachers in future customized learning and teaching. An artificial intelligence tutoring system that supports diagnosis of learning understanding and collaborative learning that can achieve learning more efficiently and at the same time achieve a more interesting learning environment through a configuration that enables collaborative learning through interaction between learners. The purpose is to provide.

However, the technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly apparent to those skilled in the art from the description below. It will be understandable.

The present invention, which is a technical means for achieving the above objective, is to remotely connect the learner's first terminal and the teacher's second terminal to diagnose the learner's learning understanding and support collaborative learning in a non-face-to-face environment. In the artificial intelligence tutoring system, a network is established and communicated with the first and second terminals and the web and external servers, and the learner's learning log ( Analyzing through a module, diagnostic information for each learner is extracted, and the diagnostic information for each learner is matched with pre-designated learning management data to generate arbitrary customized information to guide the learner's future learning and the teacher's future teaching. , a control unit that outputs a preset notification signal according to the customizing information; and a notification unit provided in the first and second terminals and visually displayed on an arbitrary screen according to a notification signal output from the control unit, wherein the diagnostic information for each learner includes the level of understanding for each learning area and each unit. It may be data about at least one learning competency among the levels of understanding, and the control unit regenerates whether the learner answered the test correctly in each test based on the learning log (X) and evaluates the full amount in a later test. It can be configured to learn to extract random understanding that can predict whether the questions are correct, and the customizing information is customized for learning/teaching at least one of understanding visualization, recommended content, recommended evaluation questions, and team career. May contain data about the guide.

In addition, the comprehension visualization data among the customizing information may be displayed on the screen of the notification unit in the form of cells connected to interrelated learning areas of the unit, and each learning area forming the cell may be displayed for each learner. The diagnostic information may be displayed on the screen of the notification unit in a preset color.

In addition, among the customizing information, the team career data is a ranking (ranking) of learners who are matched as a team with learning types within a similar range preset in response to the diagnostic information for each learner according to the number of individual recommended questions and their performance. ) first data displayed on the screen of the notification unit in a formatted and listed form; second data displayed on the screen of the notification unit in the form of an emoticon capable of expressing mutual praise and encouragement between learners in the first data; Third data that is displayed on the screen of the notification unit in the form of a unified team mission by combining the number of individual recommended questions and their performance levels of the learners in the first data; and fourth data displayed on the screen of the notification unit in the form of a gauge in which the perimeter of each learning area forming the cell changes according to the comprehensive and unified performance level in the third data.

In addition, the external server may include a product mall server for providing arbitrary learner products, and the control unit may respectively correspond to third data of the team career data through partnership with the product mall server. A product provision department that specifies the product and compensation period in advance and ensures that when the overall and unified performance reaches 100% within the compensation period and the team mission is completed, a designated product corresponding to the third data is provided. It can be included.

In addition, the control unit includes a data generation unit that acquires the learner's learning log (X) from the first terminal and converts it into data; a database unit that records the data of the learning log generated by the data generation unit, and stores and manages learning management data and arbitrary deep learning algorithm data predetermined by the second terminal; The diagnostic sub-module is generated using the deep learning algorithm data stored in the database unit, and the diagnostic information for each learner is generated by analyzing the data of the learning log recorded in the database unit through the generated diagnostic sub-module. A comprehension extraction unit that extracts; And generating the customized information by matching the diagnostic information for each learner extracted from the understanding extraction unit with the learning management data stored in the database unit, and controlling the output of a notification signal corresponding to the generated customized information. It may include a notification signal output unit.

In addition, the diagnostic submodule may apply a DKT (Deep Knowledge Tracing) AI model based on a transformer structure for natural language processing among the deep learning algorithms.

In addition, the learning log (X) is a response result for T tests, It can be defined as, and the response result for each test ( ) can be composed of responses to E evaluation questions corresponding to the test, and the response result for each evaluation question is a set element (tuple) of questions and incorrect answer results, It can be defined as:

In addition, the control unit can extract the learner's level of understanding θ when the following learning log (X) exists, and the level of understanding θ is a real value between [0,1] for all learning concepts. When the numerical value for a specific learning concept is high, it expresses high understanding, and when the numerical value is low, it expresses low understanding. It is proportional to the probability of correct answers for the evaluation questions of the learning concept, and individual evaluation questions are calculated based on this. It may be possible to predict the probability of a correct answer to

In addition, the diagnostic sub-module includes an initial understanding diagnostic module that generates preliminary diagnostic information while providing preset initial understanding values for each grade and semester based on the learning log (X); a comprehension tracking module that updates preliminary diagnosis information generated through the initial comprehension diagnosis module; The learner's guessing probability, probability of making a mistake (Slip), difficulty of the evaluation items, and discriminative power of the evaluation items are reflected in the pre-diagnosis information updated through the understanding tracking module, while providing direct understanding and individual evaluation items. A comprehension and item mapping module that extracts interpretable understanding θ by connecting the correct answer prediction probabilities; And by analyzing the learning log ( It may include a question-solving error prediction module that allows the probability of one problem-solving mistake at each point in time to be reflected as the probability of making a mistake (Slip) when deriving the final correct-correct prediction probability.

In addition, the initial comprehension diagnosis module, assuming that there is a learning log ( , uses only the M initial diagnostic tests, utilizes the transformer structure to extract hidden representations for any one test, and evaluates within the test as a dummy item indicating the start of the evaluation item. The refined final correlation of the [S0Q] node, which plays the role of summarizing the response results to the questions, is used as the correlation of the test unit expressing the corresponding test results, and positional encoding indicates the order of the evaluation questions. This may not apply.

In addition, the initial comprehension diagnostic module utilizes the correlation between the extracted M tests to assign an initial value of memory corresponding to Concept x memory (number of concepts, C=391, × memory dimension, D=48). And in this process, the correlation of the M tests is concatenated and through linear transition. You can find the value of the matrix.

In addition, the initial understanding diagnostic module is In the matrix, the linear transformation and the sigmoid function are used to obtain the degree of understanding θ for each learning concept, and the linear transformation weight, which changes D dimension to 1 dimension, is used equally for all learning concepts, With the above conversion, the the matrix It can be mapped to a comprehension area, and the mapping can be used in the same way in the comprehension tracking module.

Additionally, the comprehension tracking module will remind you which learning concepts will be affected by the current test and how. Decide whether to update the matrix, and then determine the level of understanding at each point in time. can be derived, using the same structure and approach as the initial comprehension diagnosis module and parameters different from the initial comprehension diagnosis module, for each of the N(TM) comprehension tracking tests. Extract correlations between tests and each individual test from can be extracted.

In addition, the comprehension tracking module applies a transformer encoder together with the position encoding through time series modeling (Sequence Modeling) to integrate information from time series tests, when the transformer encoder is performed, at a future point in time. The attention value is masked to prevent future correct answers from being reflected in the present, and in accordance with the characteristics of the transformer encoder, the correlation of each test is harmonized with the correlation of other tests in the time series. go It can be changed to .

In addition, the comprehension tracking module compares the summarized correlation between the evaluation questions within the test in the initial comprehension diagnosis module and the response results and encodes the information of the test excluding the response results, at this time, the initial comprehension diagnosis module. The same architecture as the comprehension diagnosis module is applied, and unlike the initial comprehension diagnosis module that uses incorrect answer results as input, the test is conducted while utilizing the parameters of the comprehension and item mapping module. Extract correlations that summarize only the evaluation item information, and each individual test above from can be extracted, and the extraction results are used to find out whether the response results of the test can change the understanding of any learning concept, and this means that regardless of the response results, the information on the evaluation questions determines what kind of test the test is. It can be configured to find out whether it is related to the learning concept.

In addition, the comprehension tracking module extracts correlations using only response results without the evaluation item information, and extracts correlations using only response results without the evaluation item information, and from is extracted, which can be used to track the probability of the learner's solution mistake, and the extracted class By utilizing the existing Update the matrix and use the same functions and weights as the comprehension and item mapping module to create the updated From the matrix, the degree of understanding θ for each learning concept can be obtained, and the existing In updating the matrix, first of all, the above What learning concepts and connections does the current test have? After obtaining, for updates using By finding The current P can be updated in proportion to It is possible to select learning concepts that are highly relevant to the current test by utilizing the linear transformation and softmax function.

In addition, the architecture of the question-solving error prediction module is It is composed of LSTM (Long Short-Term Memory) utilizing , and may be composed of the linear transformation and the sigmoid function to calculate the probability of making a mistake (Slip) at each point in time.

In addition, the diagnostic sub-module has two learner response result prediction losses and four regularization losses for learning stabilization in order to learn the diagnostic module, tracking module, mapping module, and prediction module that constitute it. It can be configured to be learned by a combination of a total of 6 losses, and the learner response result prediction loss includes Auto-encoding loss for reproducing the current test result and Next exam prediction loss for predicting the test result at the next time. The regularization loss is Curriculum regularization loss for regularizing understanding at each learning stage, Knowledge map regularization loss for regularizing understanding of knowledge map connection nodes, Proficiency smoothing loss for limiting dramatic changes in understanding, and solving mistakes. Noise smoothing loss can be included to limit dramatic changes in probability.

The artificial intelligence tutoring system that supports diagnosis and collaborative learning for learning understanding according to the present invention utilizes an existing deep learning-based knowledge tracking AI model based on a transformer structure among deep learning algorithms, so that the output value of the AI model is the following evaluation questions. Through a configuration that consists of a directly inferred value of the understanding of the concept rather than the probability of correctness, it is possible to support a more accurate diagnosis of the learner's learning understanding, thereby guiding learners and teachers in future customized learning and teaching. It is possible to achieve a management environment more efficiently, and at the same time, it is possible to achieve a more interesting learning environment through a structure that allows collaborative learning through interaction between learners.

However, the effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects not mentioned above will be clearly understood by those skilled in the art from the description below. You will be able to.

Figure 1 is a diagram showing the external configuration and use example of an artificial intelligence tutoring system that supports diagnosis of learning comprehension and collaborative learning according to an embodiment of the present invention.
Figure 2 is a block diagram schematically showing the electronic configuration of the artificial intelligence tutoring system.
Figure 3 is a block diagram schematically showing the sub-configuration of the control unit of the artificial intelligence tutoring system according to Figure 2.
Figure 4 is an image showing the sub-configuration of the diagnostic sub-module.
FIG. 5 is an image illustrating the process of extracting hidden relationships using an initial understanding diagnostic module among the sub-configurations of the diagnostic sub-module according to FIG. 4.
Figure 6 is an image illustrating the process of updating prior diagnosis information using a comprehension tracking module among the sub-configurations of the diagnosis sub-module according to Figure 4.
FIG. 7 is an image illustrating the Proficiency Pallet function for extracting interpretable comprehension θ using the comprehension and question mapping modules among the sub-configurations of the diagnostic sub-module according to FIG. 4.
Figure 8 is a usage example showing an example in which comprehension visualization data among customized information by the control unit of the artificial intelligence tutoring system is displayed in the notification unit in relation to the present invention.
Figure 9 is a usage example showing another example in which comprehension visualization data among customized information by the control unit of the artificial intelligence tutoring system is displayed in the notification unit in relation to the present invention.
Figure 10 is a usage example showing an example in which recommended evaluation item data among customizing information by the control unit of the artificial intelligence tutoring system is displayed in the notification unit in relation to the present invention.
Figure 11 is a usage example showing an example in which team career data is displayed in the notification unit among customizing information by the control unit of the artificial intelligence tutoring system in relation to the present invention.
12 is an image illustrating a screen of the notification unit related to customizing information according to another embodiment of the present invention.

Hereinafter, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the present invention. However, since the description of the present invention is only an example for structural or functional explanation, the scope of the present invention should not be construed as limited by the examples described in the text. In other words, since the embodiments can be modified in various ways and can have various forms, the scope of rights of the present invention should be understood to include equivalents that can realize the technical idea. In addition, the purpose or effect presented in the present invention does not mean that a specific embodiment must include all or only such effects, so the scope of the present invention should not be understood as limited thereby.

The meaning of terms described in the present invention should be understood as follows.

Terms such as “first” and “second” are used to distinguish one component from another component, and the scope of rights should not be limited by these terms. For example, a first component may be named a second component, and similarly, the second component may also be named a first component. When a component is referred to as being “connected” to another component, it should be understood that it may be directly connected to the other component, but that other components may also exist in between. On the other hand, when a component is referred to as being “directly connected” to another component, it should be understood that there are no other components in between. Meanwhile, other expressions that describe the relationship between components, such as "between" and "immediately between" or "neighboring" and "directly neighboring" should be interpreted similarly.

Singular expressions should be understood to include plural expressions, unless the context clearly indicates otherwise, and terms such as “comprise” or “have” refer to the specified features, numbers, steps, operations, components, parts, or them. It is intended to specify the existence of a combination, and should be understood as not excluding in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

All terms used herein, unless otherwise defined, have the same meaning as commonly understood by a person of ordinary skill in the field to which the present invention pertains. Terms defined in commonly used dictionaries should be interpreted as consistent with the meaning they have in the context of the related technology, and cannot be interpreted as having an ideal or excessively formal meaning unless clearly defined in the present invention.

Figure 1 is a diagram showing the external configuration and use example of an artificial intelligence tutoring system that supports diagnosis of learning comprehension and collaborative learning according to an embodiment of the present invention, and Figure 2 schematically shows the electronic configuration of the artificial intelligence tutoring system. Figure 3 is a block diagram schematically showing the sub-configuration of the control unit of the artificial intelligence tutoring system according to Figure 2, Figure 4 is an image showing the sub-configuration of the diagnostic sub-module, and Figure 5 is an image illustrating the process of extracting hidden relationships using the initial understanding diagnostic module among the sub-configurations of the diagnostic sub-module according to FIG. 4, and FIG. 6 is one of the sub-configurations of the diagnostic sub-module according to FIG. This is an image illustrating the process of updating pre-diagnosis information using a comprehension tracking module, and FIG. 7 shows a comprehension level that can be interpreted using the comprehension and question mapping module among the sub-configurations of the diagnostic sub-module according to FIG. 4. It is an image illustrating the Proficiency Pallet function for extracting θ, and Figure 8 shows an example in which comprehension visualization data among the customized information by the control unit of the artificial intelligence tutoring system is displayed in the notification unit in relation to the present invention. It is an exemplary diagram, and Figure 9 is a usage example showing another example in which comprehension visualization data is displayed in the notification unit among the customizing information by the control unit of the artificial intelligence tutoring system in relation to the present invention, and Figure 10 is a usage diagram showing the present invention and In relation to this, it is an example usage diagram showing an example in which recommended evaluation item data among the customizing information by the control unit of the artificial intelligence tutoring system is displayed in the notification unit, and Figure 11 is a diagram illustrating an example of displaying the recommended evaluation item data in the notification unit. This is an example usage diagram showing an example of team career data being displayed in the notification unit among customizing information, and Figure 12 is an image showing an example of a screen of the notification unit related to customizing information according to another embodiment of the present invention.

As shown in Figures 1 and 2, the artificial intelligence tutoring system 100 that supports diagnosis of learning comprehension and collaborative learning according to the present invention includes a first terminal 110, a second terminal 120, and a web ( 130), an external server 140, a control unit 150, and a notification unit 160.

However, the components shown in FIGS. 1 and 2 are only the minimum essential components for the present invention, and an artificial intelligence tutoring system with more additional components can be implemented.

The first terminal 110 is configured to include various known input devices (not shown) such as a keyboard, pad, and mouse. According to a preferred embodiment of the present invention, the first terminal 110 is located at the learner (L). An input environment is provided so that the learning log (X), which will be described later, is input to the web 130, the external server 140, and the control unit 150.

In addition, the second terminal 120 is configured to enable remote communication with the above-described first terminal 110, and may have the same components as the first terminal 110. According to the present invention, It is located to the teacher (T) and provides an input environment where the teacher (T) can input learning management data, which will be described later, into the web 130, the external server 140, and the control unit 150.

Here, the first and second terminals 110 and 120 may be composed of a plurality of terminals, and are preferably tablets, but are not limited thereto and may be PCs, smartphones, etc.

Since the configuration of the first and second terminals 110 and 120 corresponds to a typical terminal device that is generally published, the detailed configuration is not shown in the drawings and a more detailed description thereof will be omitted.

The web 130 is an Internet service configured to support various multimedia such as text, graphics, images, sounds, and videos. According to the present invention, the web 130 is connected to the first and second terminals 110 and 120 to provide the first and second terminals 110 and 120. 2 It can be configured to provide various known educational content to learners (L) and teachers (T) who use the terminals 110 and 120, respectively.

The external server 140 is preferably configured to correspond to the above-described web 130 to form a network with the first and second terminals 110 and 120, the web 130, and the control unit 150, and the web It performs the function of servicing a preset program to the control unit 150 in conjunction with 130.

Here, the program may be an educational program composed of various data related to education, and as this is a known technology, it goes without saying that it can be freely changed and designed by a person skilled in the art.

The control unit 150 establishes and communicates with the first and second terminals 110 and 120, the web 130, and the external server 140, and controls the learning of the learner (L) obtained from the first terminal 110. Diagnosis information for each learner is extracted by analyzing the log ( It generates arbitrary customizing information to guide learning and future teaching of the teacher (T), and performs a function of outputting a preset notification signal according to the customizing information.

Referring to FIG. 3, more specifically, the control unit 150 includes a data generation unit 151 that acquires the learning log (X) of the learner (L) from the first terminal 110 and converts it into data; Recording the data of the learning log (X) generated by the data generator 151, and storing and managing learning management data and arbitrary deep learning algorithm data predetermined by the second terminal 120. database unit 152; The diagnostic sub-module is generated using the deep learning algorithm data stored in the database unit 152, and the data of the learning log (X) is recorded in the database unit 152 through the generated diagnostic sub-module. A comprehension extraction unit 153 that analyzes and extracts diagnostic information for each learner; And matching the diagnostic information for each learner extracted from the understanding extraction unit 153 with the learning management data stored in the database unit 152 to generate the customized information, and to generate the customized information corresponding to the generated customized information. It may be configured to include a notification signal output unit 154 that controls the notification signal to be output.

Here, the diagnostic information for each learner may be data on at least one learning competency among understanding by learning area and understanding by unit, and at this time, the control unit 150 controls the learner ( L) can be configured to learn to reproduce the correct answers to the tests solved in each test and to extract a random understanding that can predict whether the full evaluation questions were answered correctly in future tests.

According to a preferred embodiment of the present invention, the learning log (X) is a response result for T tests, It can be defined as, and the response result for each test ( ) may be composed of responses to E evaluation questions corresponding to the test, and the response result for each evaluation question is a set element (tuple) of questions and incorrect answer results, It can be defined as:

In addition, the control unit 150 can extract the understanding degree θ of the learner (L) when the following learning log (X) exists, and the understanding degree θ is [0,1 for all learning concepts. ] It is a vector with real values in the range. When the numerical value for a specific learning concept is high, it expresses high understanding, and when the numerical value is low, it expresses low understanding, and is proportional to the probability of correct answers for the evaluation questions of the learning concept. Based on this, it may be possible to predict the probability of correct answers for individual evaluation questions.

In addition, the learning management data is configured to correspond to the learning log (X) of the learner (L), which changes in real time, and is data that serves as an indicator for the teacher (T) to manage the learning of the learner (L). As such, it can be pre-specified by being input through an input device (not shown) of the second terminal 120 used by the teacher (T), and can also be input into the web 130 and the external server 140 and be pre-specified. You can.

At this time, the learning management data is data that quantifies learning behaviors and patterns that occur during the learning process, level of subject understanding, total amount and pattern of learning amount by period, statistical distribution and trend of grades, causes, and prescriptions for the same, and three types of learning. (Learning strategy, learning motivation, learning maladaptation), 1 type of career/aptitude (interest, aptitude), 4 types of personality/personality (interpersonal environment, academic stress, self-esteem, emotional behavior) A total of 8 types of diagnostic tests for emotions are provided. It is desirable to include recorded data, etc.

According to the present invention, the learning management data can be quantified based on the international standard (IMS Caliper), and the AI analysis service program "Home Run AI Life Record" currently commercialized on the web 130 or external server 140 " It can be extracted through sharing, and of course, various modifications and designs are possible by those skilled in the relevant field within the technical scope of the present invention.

The Home Run AI Life Record is a learning analysis service provided by Ice Cream Edu Co., Ltd. It has received IMS Caliper Analytics 1.1 standard certification from the IMS Global Learning Consortium and analyzes the learning status, learning patterns, and study habits of the learner (L). This is a technology that summarizes monthly and provides visual materials in the form of an e-portfolio to learners (L) and teachers (T).

Furthermore, the customizing information may include data on at least one customized learning/teaching guide among comprehension visualization, recommended content, recommended evaluation questions (see FIG. 10), and team career, and among the customizing information, the understanding As shown in Figure 8 or Figure 9, the visualization data may be displayed on the screen of the notification unit 160, which will be described later, in the form of a cell connected to the interrelated learning areas of the unit, and each learning area forming the cell may be displayed on the screen of the notification unit 160 in a preset color corresponding to the diagnostic information for each learner.

In addition, according to the present invention, the team career data among the customizing information refers to the number of individually recommended questions for learners who are matched as a team with learning types within a similar range preset in response to the diagnostic information for each learner, with reference to FIG. 11. and first data (D1) displayed on the screen of the notification unit in a ranked form according to the performance level. Second data (D2) displayed on the screen of the notification unit in the form of emoticons that can express mutual praise and encouragement between learners in the first data (D1); Third data (D3) that is displayed on the screen of the notification unit in the form of a unified team mission by combining the number of recommended questions and their performance levels for each learner in the first data (D1); And fourth data (D4) displayed on the screen of the notification unit in the form of a gauge in which the perimeter of each learning area forming the cell changes according to the comprehensive and unified performance level in the third data (D3). It can be configured as follows.

Here, in relation to team career data, the external server 140 according to the present invention may include a product mall server for providing arbitrary products for learners, and the control unit 150 may cooperate with the product mall server. The product and compensation period corresponding to the third data (D3) of the team career data are designated in advance, and when the comprehensive and unified performance level reaches 100% within the compensation period and the team mission is completed, It may further include a product provision unit 155 that provides a designated product corresponding to the third data D3.

Meanwhile, the diagnostic submodule according to the present invention preferably applies a DKT (Deep Knowledge Tracing) AI model based on a Transformer structure for natural language processing among the deep learning algorithms.

Referring to FIG. 4, the diagnostic submodule is, more specifically, an initial comprehension diagnostic module that generates preliminary diagnostic information while providing preset initial comprehension values for each grade and semester based on the learning log (X). ; a comprehension tracking module that updates preliminary diagnosis information generated through the initial comprehension diagnosis module; The pre-diagnosis information updated through the understanding tracking module reflects the learner's (L) guessing probability, probability of making a mistake (Slip), difficulty level of the evaluation item, and discriminative power of the evaluation item, directly improving understanding and Comprehension and item mapping module that extracts interpretable understanding θ by connecting the correct prediction probability of individual evaluation questions; And if it is determined by analyzing the learning log ( It may be configured to include a question-solving mistake prediction module that tracks and allows the probability of one solving mistake that the learner (L) has at each point in time to be reflected as the probability of making a mistake (Slip) when deriving the final correct prediction probability. there is.

As shown in FIG. 5, assuming that there is a learning log ( Divided into tests for use, only the M initial diagnostic tests are used, the transformer structure is used to extract a hidden representation for any one test, and a marker (dummy) indicating the start of the evaluation item is used. As an item, the refined final correlation of the [S0Q] node, which plays the role of summarizing the response results to the evaluation questions within the test, is used as the correlation of the test unit expressing the test result, and is a position indicating the order of the evaluation items. Encoding (positional encoding) may not be applied.

In addition, the initial comprehension diagnostic module utilizes the correlation between the extracted M tests to assign an initial value of memory corresponding to Concept x memory (number of concepts, C=391, × memory dimension, D=48), , In this process, the correlation of the M tests is concatenated and through linear transition. You can find the value of the matrix.

In addition, the initial understanding diagnostic module is In the matrix, the linear transformation and the sigmoid function are used to obtain the degree of understanding θ for each learning concept, and the linear transformation weight, which changes D dimension to 1 dimension, is used equally for all learning concepts, With the above conversion, the the matrix It can be mapped to a comprehension area, and the mapping can be used in the same way in the comprehension tracking module.

With reference to FIG. 6, the understanding tracking module determines which learning concepts will be affected by the current test and how. Decide whether to update the matrix, and then determine the level of understanding at each point in time. can be derived, using the same structure and approach as the initial comprehension diagnosis module and parameters different from the initial comprehension diagnosis module, for each of the N(TM) comprehension tracking tests. Extract correlations between tests and each individual test from can be extracted.

In addition, the comprehension tracking module applies a transformer encoder together with the position encoding through time series modeling (Sequence Modeling), which integrates information from time series tests, when the transformer encoder is performed, it provides information about future points in time. Attention value is masked to prevent future correct answers from being reflected in the present, and in accordance with the characteristics of the transformer encoder, the correlation of each test is harmonized with the correlation of other tests in the time series. go It can be changed to .

In addition, the comprehension tracking module compares the summarized correlation between the assessment questions within the test in the initial comprehension diagnosis module and the response results and encodes the information of the test excluding the response results, and at this time, the initial understanding The same architecture as the diagnostic module is applied, and unlike the initial comprehension diagnostic module that uses incorrect answer results as input, the parameters of the comprehension and item mapping module are used to test the test. Extract relationships that summarize only the evaluation item information, and each individual test above from can be extracted, and the extraction results are used to find out whether the response results of the test can change the understanding of any learning concept, and this means that regardless of the response results, the information on the evaluation questions determines what kind of test the test is. It can be configured to find out whether it is related to the learning concept.

Furthermore, the comprehension tracking module extracts correlations using only response results without the evaluation item information, and from is extracted, which can be used to track the probability of the learner's solution mistake, and the extracted class By utilizing the existing Update the matrix and use the same functions and weights as the comprehension and item mapping module to create the updated From the matrix, the degree of understanding θ for each learning concept can be obtained, and the existing In updating the matrix, first of all, the above What learning concepts and connections does the current test have? After obtaining, for updates using By finding The current P can be updated in proportion to It is possible to select learning concepts that are highly relevant to the current test by utilizing the linear transformation and softmax function.

In the present invention, the function (see FIG. 7) that connects the direct understanding described above in the comprehension and question mapping module with the correct answer prediction probability of individual evaluation questions is named the Proficiency Pallet function.

Through the Proficiency Pallet function, the value of comprehension θ is linked to the difficulty level of the evaluation item and converted into an interpretable value.

For example, if the level of understanding of a specific learning concept is 0.7, it is judged that the evaluation questions with difficulty levels of 0.5, 0.3, and 0.6 can be answered correctly with a probability of more than 50%, and the difficulty level of the evaluation items of 0.8 and 0.9 is judged to be correct with a probability of more than 50%. It can be judged that there is a probability of more than 50% that the correct answer will not be possible for the corresponding evaluation questions.

At this time, it is desirable that the evaluation questions in the same learning concept use the same level of understanding of the learning concept.

According to the present invention, the architecture of the question-solving error prediction module is It is preferably composed of an LSTM (Long Short-Term Memory) utilizing the linear transformation and the sigmoid function to calculate the probability of making a mistake (Slip) at each point in time. .

The diagnostic submodule, which is implemented as described above, uses two learner response result prediction losses and four regularizations for learning stabilization in order to learn the diagnostic module, tracking module, mapping module, and prediction module that constitute it. ) It can be configured to be learned by a combination of a total of 6 Rosses.

The learner response result prediction loss may more preferably include an auto-encoding loss for reproducing the current test result and a Next exam prediction loss for predicting the test result at the next time point.

Regarding the above auto-encoding loss, the trials at each time are finally accumulated and summarized in Concept Memory, which is finally assigned to concept understanding θ. At this time, individual concepts are expressed as comprehension levels of scalar values, and these scalar values are representative comprehension values that express the results of the currently taken test. Therefore, the level of understanding at that point in time θ must well express the level of understanding at that point in time. Using the degree of understanding θ, the Proficiency Pallet function is used to construct the probability of predicting correct answers for each evaluation question at that time and the cross-entropy loss of the incorrect answer results for the evaluation questions at the same time.

Regarding the above Next exam prediction loss, the level of understanding θ at each time point must not only well express the level of understanding of the learning concept of the test at that time, but also be able to predict changes in understanding of other learning concepts. To learn this, a loss that predicts the result of the next test is additionally constructed. In other words, using the current level of understanding θ, the Proficiency Pallet function is used to construct the probability of correct answer prediction for each evaluation question at the “ next time ” and the cross-entropy loss of the incorrect answer results for the evaluation questions at the “ next time ”.

The regularization loss is more preferably Curriculum regularization loss for regularizing understanding at each learning stage, Knowledge map regularization loss for regularizing understanding of knowledge map connection nodes, Proficiency smoothing loss for limiting dramatic changes in understanding, and Noise smoothing loss can be included to limit dramatic changes in the probability of solving mistakes.

Regarding the curriculum regularization loss above, the elementary mathematics course includes a sequential knowledge acquisition process. Accordingly, the generally existing 1st grade → 2nd grade → … → Add restrictions on the relative size of understanding of learning concepts in the order of 6th grade. Loss is composed of ReLU (upper units - lower units), which means that when the understanding of the upper unit is higher than the understanding of the lower unit (when the understanding of the 6th grade unit is higher than the understanding of the 1st grade unit), it lowers the understanding of the upper unit and lowers the understanding of the lower unit. This is a regularized Ross that improves understanding of the unit.

Regarding the above Knowledge map regularization loss, each learning concept has its precedence relationship intertwined by the knowledge map. This is used to construct a loss that encodes the precedence relationship. Loss is composed of ReLU (resulting unit - preceding unit), which is a constraint that the understanding of the succeeding unit must be lower than that of the preceding unit. When learning this diagnostic submodule, we observed a tendency for the diagnostic submodule to significantly overfit the data without the corresponding regularization loss.

Regarding the above Proficiency smoothing loss, the actual learner's (L) understanding improves step by step for each learning area rather than all learning concepts changing dramatically at once. To prevent such dramatic changes and to track understanding stably, a constraint is added to minimize the distance between the current understanding θ and the previous understanding θ. The actual loss consists of absolute loss.

In relation to the above Noise smoothing loss, a regularization term is added to prevent dramatic changes in the probability of slipping (Slip) in order to prevent the probability of solving mistakes from being overfitted to the learning data. This is an absolute loss and is composed of the distance between the probability of making a mistake (Slip) at the previous time and the probability of making a mistake (Slip) at the current time.

The notification unit 160 visually displays preset text and images on an arbitrary screen according to a notification signal output from the notification signal output unit 154 of the control unit 150. Preferably, the learner ( It is desirable to configure it to be provided in the first terminal 110 of L) and the second terminal 120 of the teacher T.

The notification unit 160 may be used for a liquid crystal display (LCD), a thin film transistor-liquid crystal display (TFT LCD), an organic light-emitting diode (OLED), or a flexible display. It may be configured to include at least one of a flexible display and a 3D display.

A detailed description of preferred embodiments of the invention disclosed above is provided to enable any person skilled in the art to make or practice the invention. Although the present invention has been described above with reference to preferred embodiments, those skilled in the art will understand that various modifications and changes can be made to the present invention without departing from the scope of the present invention. For example, a person skilled in the art may use each configuration described in the above-described embodiments by combining them with each other. Accordingly, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The present invention may be embodied in other specific forms without departing from the spirit and essential features of the present invention. Accordingly, the above detailed description should not be construed as restrictive in all respects and should be considered illustrative. The scope of the present invention should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of the present invention are included in the scope of the present invention. The present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein. In addition, claims that do not have an explicit reference relationship in the patent claims can be combined to form an embodiment or included as a new claim through amendment after filing.

100: Artificial intelligence tutoring system
110: first terminal
120: second terminal
130: web
140: external server
150: control unit
151: data generation unit
152: database section
153: Comprehension extraction unit
154: Notification signal output unit
155: Product provision department
160: Notification unit
L: learner
T: teacher

Claims (18)

In an artificial intelligence tutoring system for diagnosing the learner's learning understanding and supporting collaborative learning in a non-face-to-face environment by remotely connecting the learner's first terminal and the teacher's second terminal,
While establishing and communicating with the first and second terminals and the web and external servers, the learner's learning log (X) acquired from the first terminal is analyzed through a diagnostic submodule based on a pre-programmed deep learning algorithm, Diagnosis information is extracted, and the diagnostic information for each learner is matched with pre-designated learning management data to generate arbitrary customization information to guide the learner's future learning and the teacher's future teaching, and is customized in advance according to the customization information. A control unit that outputs a set notification signal; and
A notification unit provided in the first and second terminals and visually displayed on an arbitrary screen according to a notification signal output from the control unit,
The diagnostic information for each learner is,
Characterized by data on at least one learning competency among understanding by learning area and understanding by unit,
The control unit,
Based on the learning log ( Characterized by
The customization information is,
It is characterized by including data on at least one learning/teaching customized guide among comprehension visualization, recommended content, recommended evaluation questions, and team career,
Among the customizing information, the comprehension visualization data is,
Characterized in that the interrelated learning areas of the unit are displayed on the screen of the notification unit in the form of cells connected to each other,
Each learning area that makes up the cell is,
Characterized in that the diagnostic information for each learner is displayed on the screen of the notification unit in a preset color,
Among the customizing information, the team career data is,
On the screen of the notification unit, learners who are matched as a team with learning types within a similar preset range in response to the diagnostic information for each learner are ranked and listed according to the number of individual recommended questions and their performance level. first data displayed on;
second data displayed on the screen of the notification unit in the form of an emoticon capable of expressing mutual praise and encouragement between learners in the first data;
Third data that is displayed on the screen of the notification unit in the form of a unified team mission by combining the number of individual recommended questions and their performance levels of the learners in the first data; and
Characterized in that the perimeter of each learning area constituting the cell includes fourth data displayed on the screen of the notification unit in the form of a gauge that changes according to the comprehensive and unified performance level in the third data,
The external server is,
Characterized by including a product mall server for providing random learner products,
The control unit,
Through partnership with the product mall server, the product and compensation period corresponding to each third data of the team career data are designated in advance, and the comprehensive and unified performance level reaches 100% within the compensation period to perform the team mission. Once this is completed, it includes a product provision department that ensures that designated products corresponding to the relevant third data are provided together,
a data generator that obtains the learner's learning log (X) from the first terminal and converts it into data;
a database unit that records the data of the learning log generated by the data generation unit, and stores and manages learning management data and arbitrary deep learning algorithm data predetermined by the second terminal;
The diagnostic sub-module is generated using the deep learning algorithm data stored in the database unit, and the diagnostic information for each learner is generated by analyzing the data of the learning log recorded in the database unit through the generated diagnostic sub-module. A comprehension extraction unit that extracts; and
Generates the customized information by matching the diagnostic information for each learner extracted from the understanding extraction unit with the learning management data stored in the database unit, and controls the output of a notification signal corresponding to the generated customized information. Characterized by further comprising a notification signal output unit,
The diagnostic submodule is,
Among the deep learning algorithms, it is characterized by applying a DKT (Deep Knowledge Tracing) AI model based on a Transformer structure for natural language processing,
The learning log (X) is the response result for T tests, Characterized by being defined as,
Response results for each test ( ) is characterized by being composed of responses to E evaluation questions corresponding to the test,
The response result for each evaluation question is a set element (tuple) of the question and the incorrect answer result, Characterized by being defined as,
The control unit,
When the following learning log (X) exists, the learner's understanding degree θ is extracted,
The degree of understanding θ is,
For all learning concepts, it is a vector with real values between [0,1]. When the value for a specific learning concept is high, it expresses high understanding. When the value is low, it expresses low understanding. Evaluation questions for the learning concept. It is characterized by being able to predict the probability of correct answers for individual evaluation questions based on this, in proportion to the probability of correct answers for each question.
The diagnostic submodule is,
An initial understanding diagnosis module that generates preliminary diagnosis information while providing preset initial understanding values for each grade and semester based on the learning log (X);
a comprehension tracking module that updates preliminary diagnosis information generated through the initial comprehension diagnosis module;
The learner's guessing probability, probability of making a mistake (Slip), difficulty of the evaluation items, and discriminative power of the evaluation items are reflected in the pre-diagnosis information updated through the understanding tracking module, while providing direct understanding and individual evaluation items. A comprehension and item mapping module that extracts interpretable understanding θ by connecting the correct answer prediction probabilities; and
By analyzing the learning log ( Characterized by comprising a question-solving error prediction module such that the probability of one solving mistake at a time is reflected as the probability of making a mistake (Slip) when deriving the final correct prediction probability,
The initial understanding diagnostic module is,
Assuming that there is a learning log ( , the transformer structure is utilized to extract hidden representations for any one test, and serves as a dummy item indicating the start of the evaluation item to summarize the response results to the evaluation items within the test. The refined final correlation of the performing [S0Q] node is used as the correlation of the test unit expressing the test result, but positional encoding indicating the order of the evaluation items is not applied,
Using the correlation of the extracted M tests, an initial value of memory corresponding to Concept x memory (number of concepts, C=391, × memory dimension, D=48) is assigned, and in this process, Concatenation of relationships and through linear transition Characterized by calculating the value of a matrix,
remind In the matrix, the linear transformation and the sigmoid function are used to obtain the degree of understanding θ for each learning concept, and the linear transformation weight, which changes D dimension to 1 dimension, is used equally for all learning concepts, With the above conversion, the above the matrix It is characterized by being mapped to the comprehension area of,
The mapping is,
Characterized in that it is used in the same way in the comprehension tracking module,
The comprehension tracking module is,
Remind yourself what learning concepts will be affected by the current test and how. Decide whether to update the matrix, and then determine the level of understanding at each point in time. It is characterized by deriving,
Using the same structure and approach as the initial comprehension diagnosis module and different parameters from the initial comprehension diagnosis module, the correlation between each test is extracted for the N(TM) comprehension tracking tests, , each individual test from Characterized by extracting,
When the transformer encoder is performed, the attention value for the future is masked through Sequence Modeling, which integrates information from time series tests by applying a transformer encoder together with the position encoding. (Masking) processing prevents future correct answers from being reflected in the present, and the correlation of each test is harmonized with the correlation of other tests in the time series according to the characteristics of the transformer encoder. go It is characterized by allowing it to change to,
By comparing the summarized correlation between the evaluation questions within the test and the response results in the initial comprehension diagnosis module, the information of the test excluding the response results is encoded, and at this time, the same architecture as the initial comprehension diagnosis module is used. ) is applied, and unlike the initial comprehension diagnosis module that uses incorrect answer results as input, the correlation that summarizes only the evaluation item information of the test while utilizing the parameters of the comprehension and item mapping module Extract and test each individual above from Characterized by extracting,
The extraction result is,
The response results of the test are used to find out whether the understanding of any learning concept can be changed, and regardless of the response results, it is possible to find out which learning concept the test is related to only with the information of the evaluation questions. An artificial intelligence tutoring system that supports diagnosis of learning comprehension and collaborative learning.
delete delete delete delete delete delete delete delete delete delete delete delete delete delete According to claim 1,
The comprehension tracking module is,
Extract correlation using only response results without the above evaluation item information, and extract each individual test above. from is extracted, which is used to track the probability of the learner's solution mistake,
The extracted class By utilizing the existing Update the matrix and use the same functions and weights as the comprehension and item mapping module to create the updated It is characterized by calculating the degree of understanding θ for each learning concept in the matrix,
The existing When updating the matrix,
First of all, the above What learning concepts and connections does the current test have? After obtaining, for updates using By finding Update the current P in proportion to
remind Is,
An artificial intelligence tutoring system that supports diagnosis of learning comprehension and collaborative learning, characterized by selecting learning concepts that are highly relevant to the current test using the linear transformation and softmax function.
According to claim 16,
The architecture of the question-solving error prediction module is,
remind It is composed of LSTM (Long Short-Term Memory) utilizing , and is characterized by being composed of the linear transformation and the sigmoid function to calculate the probability of making a mistake (Slip) at each point in time. An artificial intelligence tutoring system that supports diagnosis of learning comprehension and collaborative learning.
According to claim 17,
The diagnostic submodule is,
In order to learn the diagnostic module, tracking module, mapping module, and prediction module that make up this module, a total of 6 losses are combined, including 2 learner response result prediction losses and 4 regularization losses for learning stabilization. Characterized by being configured to be learned by,
The learner response result prediction loss is,
Includes Auto-encoding loss for reproducing current test results and Next exam prediction loss for predicting test results at the next time.
The regularization loss is,
Curriculum regularization loss to regularize understanding at each learning stage, Knowledge map regularization loss to regularize understanding of knowledge map connection nodes, Proficiency smoothing loss to limit dramatic changes in understanding, and Noise to limit dramatic changes in the probability of solving mistakes. An artificial intelligence tutoring system that supports diagnosis and collaborative learning for learning comprehension, including smoothing loss.