Interactive learning system, interactive learning method, and interactive learning program

Abstract

To facilitate a learner to improve his / her academic capability through immersive interactions between the learner and a system.SOLUTION: In a server device which is an interactive learning system, an artificial intelligence unit has a first trained model having a first message as output and a second trained model that has machine-learned contents included in one or more teaching materials. The first trained model, where fine-tuning is performed using an input message, includes: an input step in which an input message input by a user is accepted; a response step in which a first instruction is given to an artificial intelligence to generate a first message in accordance with the first trained model based on the input message; a scoring step in which the correctness of a content for the input message is scored based on the second trained model; a response step in which a second instruction is given to the artificial intelligence to generate a second message based on at least a scoring result in the scoring step; and an output step in which the second message output by the artificial intelligence is output.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an interactive learning system, an interactive learning method, and an interactive learning program.

Background Art

[0002] In the learning of learners, it is known that knowledge becomes more firmly established by teaching others such as classmates who study together. In addition, learners can also improve their language expression ability and communication ability by teaching others. Therefore, there is a need for a technology that can improve the ability of learners by ostensibly giving them an opportunity to teach others.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] Patent Document 1 discloses a dialogue system for effectively improving the foreign language skills, particularly speaking skills, of learners through dialogue with virtual characters. Patent Document 2 relates to two-way interaction training, and presents a system in which a user can intervene in and stop a pre-recorded scenario to identify errors when dealing with scenarios and the like.

[0005] An object of the present invention is to provide a system that can promote the improvement of the academic ability of users through a dialogue full of a sense of presence with the system.

Means for Solving the Problems

[0006] To achieve the above object, a dialogue learning system according to one aspect of the present invention is a dialogue learning system that receives an input message from a user who is a learner and displays a first message generated by artificial intelligence as a reply to the user from a virtual learner. The artificial intelligence has a first learned model that takes the input message as input and outputs the first message, and a second learned model that has learned the content included in one or more teaching materials by machine learning. The first learned model is fine-tuned by the input message and includes a processor. In the input step, the processor receives the input message input by the user. In the response step, based on the input message, the processor gives a first instruction to the artificial intelligence to generate the first message according to the first learned model. In the scoring step, the processor scores the correctness of the content of the input message based on the second learned model. In the response step, based on at least the scoring result in the scoring step, the processor gives a second instruction to the artificial intelligence to generate a second message. In the output step, the processor outputs the first message output by the artificial intelligence based on the input message and the second message output by the artificial intelligence based on the second instruction.

[0007] In the scoring step, an answer instruction to input a question to the fine-tuned first learned model and generate an answer message including an answer may be given to the artificial intelligence, and the correctness of the answer may be scored based on the second learned model.

[0008] In the scoring step, the correctness may be scored by matching the history of the input message with the second learned content stored in the second learned model.

[0009] In the input step, further, information regarding the learning field is received. Further, in the question - posing step, at least the information regarding the learning field is used as an input, and a third instruction is given to the artificial intelligence to generate a third message including a question regarding the learning field for the user according to the first learned model. In the output step, the third message output by the artificial intelligence based on the third instruction may be outputted.

[0010] In the input step, further, information regarding the learning field is received. The processor further, in the lesson - requirement step, gives a fourth instruction to the artificial intelligence to generate a fourth message for teaching knowledge to the user as a lesson by a virtual teacher according to at least the second learned model based on the information regarding the learning field. In the output step, the fourth message output by the artificial intelligence based on the fourth instruction may be outputted.

[0011] In the response step, further, a fifth instruction is given to the artificial intelligence to generate a fifth message for the virtual teacher based on the fourth message according to the fine - tuned first learned model. In the output step, the fifth message output by the artificial intelligence based on the fifth instruction may be outputted.

[0012] To achieve the above object, a dialogue learning method according to another aspect of the present invention is a dialogue learning method that receives an input message from a user who is a learner and displays a first message generated by artificial intelligence as a reply to the user from a virtual learner, wherein the artificial intelligence has a first learned model that takes the input message as an input and outputs the first message, and a second learned model that has learned the content included in one or more teaching materials, the first learned model is fine-tuned by the input message, and the computer executes an input reception step of receiving the input message input from the user, a response step of giving an instruction to the artificial intelligence to generate the first message based on the input message, a scoring step of scoring the correctness of the content of the input message based on the second learned model, a response step of giving an instruction to the artificial intelligence to generate a second message based on at least the scoring result in the scoring step, and an output step of outputting the input message output by the artificial intelligence based on the first instruction and the second message output by the artificial intelligence based on the second instruction.

[0013] To achieve the above object, a dialogue learning program according to still another aspect of the present invention is a dialogue learning program that receives an input message from a user who is a learner and displays a first message generated by artificial intelligence as a reply to the user from a virtual learner. The artificial intelligence has a first learned model that takes the input message as an input and outputs the first message, and a second learned model that has learned the content included in one or more teaching materials by machine learning. The first learned model is fine-tuned by the input message, and the computer is caused to execute an input reception step of receiving the input message input from the user, a response step of giving a first instruction to the artificial intelligence to generate the first message based on the input message, a scoring step of scoring the correctness of the content of the input message based on the second learned model, a response step of giving a second instruction to the artificial intelligence to generate a second message based on at least the scoring result in the scoring step, and an output step of outputting the input message output by the artificial intelligence based on the first instruction and the second message output by the artificial intelligence based on the second instruction. Note that the computer program can be stored and provided in various data-readable recording media, or can be provided so as to be downloadable via a network such as the Internet.

Effects of the Invention

[0014] According to the present invention, it is possible to promote the improvement of the user's academic ability through an immersive dialogue with the system.

Brief Description of the Drawings

[0015]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Mode for Carrying Out the Invention

[0016] Hereinafter, embodiments of the dialogue learning system according to the present invention will be described with reference to the drawings.

[0017] ● Dialogue learning system (1) ● FIG. 1 shows the configuration of the server device 10 according to an embodiment of the present invention. The server device 10 is an example of a dialogue learning system, and is a device that realizes improvement of the learner's academic ability by having a virtual dialogue with the learner who is the user. The server device 10 imitates a virtual learner who has an evaluation value of learning equal to or lower than that of the learner, that is, a so-called low grade, and talks to the learner by text or voice. The virtual learner asks questions about the learning field to be learned to the learner, and responds to the learner's answer, so that the virtual learner behaves as if it were a classmate studying together and keeps pace with the learner in learning. The server device 10 acquires the text or voice utterance from the learner to the virtual learner and uses it as an input to the virtual learner.

[0018] In addition, the server device 10 pretends to be a virtual teacher and talks to the learner and the virtual learner in words or voice. The virtual teacher presents knowledge that the learner and the virtual learner may remember, or poses questions to the learner and the virtual learner. The virtual learner may answer the questions posed by the virtual teacher. Further, the server device 10 has a text or voice conversation between the virtual teacher and the virtual learner, and allows the learner to confirm the state of this conversation. By grasping such a conversation, the learner can further solidify knowledge through a vivid experience as if he / she has heard the answers, questions, and explanations of the teacher of another learner studying in the same classroom.

[0019] As shown in FIG. 1, the server device 10 is configured to be communicable with a learner terminal 30 used by a learner via a network NW. The server device 10 may be configured by a hardware device, or some or all of its functions may be realized by a cloud computer. Further, each configuration of the server device 10 may be realized by an API (Application Programming Interface). In this embodiment, the mutual communication between the server device 10 and the learner terminal 30 is wireless, but some or all of the connections may be wired. Furthermore, the server device 10 may be configured by a plurality of hardware configurations. In this case, the plurality of hardware configurations may be connected by wire or wirelessly, and information may be transmitted and received between them.

[0020] ● Learner Terminal 30 The learner terminal 30 is a terminal operated by a learner, such as a smartphone, a tablet terminal, or a personal computer. The learner terminal 30 mainly constitutes a functional block including an output unit 31, an operation reception unit 32, and a communication processing unit 33 by a CPU (Central Processing Unit), a computer program executed by the CPU, a RAM (Random Access Memory) and a ROM (Read Only Memory) that store the computer program and predetermined data, etc.

[0021] The output unit 31 is a so-called display unit realized by, for example, a display for displaying data. A message for the user who is the learner received from the server device 10 is displayed on the output unit 31. Note that the output unit 31 may be, instead of or in addition to the above, a speaker that outputs sound, etc. In this case, the output unit 31 outputs the message for the user as sound. In the following description, the message may be not only a displayed character string but also sound. Further, the message includes not only a problem in the learning field that the learner intends to learn but also greetings and responses for creating a sense of presence in the dialogue.

[0022] The operation reception unit 32 is realized by a touch panel, a keyboard, a mouse, a microphone, etc. for inputting data. The operation reception unit 32 can input a designated learning field and a message for the virtual learner or the virtual teacher.

[0023] The communication processing unit 33 is a processing unit that can execute data transmission and reception processing according to a predetermined protocol with the server device 10 via a network NW such as the Internet, and is realized by an app or a web browser, etc.

[0024] ● Server device 10 Here, the configuration of the server device 10 will be described with reference to FIG. 1. The server device 10 is an information processing device that executes virtual dialogue processing while exchanging data with the learner terminal 30 via the network NW. The server device 10 stores a teaching material information database DB1, a virtual learner database DB2, and a virtual teacher database DB3.

[0025] The teaching material information database DB1 stores knowledge information referred to in questions posed by a virtual learner or a virtual teacher, which will be described later. The teaching material information database DB1 also has a learned model (an example of the "second learned model" in the claims) obtained by machine learning one or more teaching materials. A teaching material includes content related to a predetermined learning field and is a so-called textbook or reference book. A teaching material is composed of, for example, any one or a combination of text data, image data, video data, and audio data. A teaching material may be provided by an administrator or the like, or may include information collected from the Internet or the like. Further, the learned model may perform machine learning on a question set presented to a learner. Instead of or in addition to this configuration, the artificial intelligence unit 15 may generate questions from the machine learning data of the teaching materials.

[0026] The teaching material information database DB1 stores the content of teaching materials pre-registered by an administrator or the like. The server device 10 has a function of artificial intelligence (AI: Artificial Intelligence) and selects and outputs questions posed by a virtual learner or a virtual teacher using the information stored in the teaching material information database DB1. Further, the server device 10 may select and output knowledge information to be displayed on the screen using the information stored in the teaching material information database DB1 by the function of artificial intelligence.

[0027] The hypothetical learner database DB2 has a learned model (an example of the "first learned model" in the claims) that takes as input messages from the learner or the virtual teacher and outputs messages issued by the hypothetical learner. Further, the hypothetical learner database DB2 learns knowledge acquired through interaction with the learner and is fine-tuned. Note that if the learner teaches incorrect knowledge, it may be fine-tuned as is, and information conflicting with the information stored in the teaching material information database DB1 or the virtual teacher database DB3 may be accumulated.

[0028] The hypothetical learner database DB2 and the virtual teacher database DB3 are configured independently of each other and store the words and knowledge used in the messages issued by the hypothetical learner and the virtual teacher, respectively. The hypothetical learner database DB2 behaves as if it has less memory than the teaching material information database DB1, at least for the learning fields that the learner plans to learn. Also, the hypothetical learner database DB2 may be configured such that knowledge in the learning field is not stored in the first place.

[0029] The hypothetical learner database DB2 and the virtual teacher database DB3 may store the words used in the problem generation process. In this case, the artificial intelligence generates questions to be asked from the hypothetical learner to the learner using the information stored in the teaching material information database DB1 and the hypothetical learner database DB2. Also, questions to be posed from the virtual teacher to the learner and the hypothetical learner are generated using the information stored in the teaching material information database DB1 and the virtual teacher database DB3.

[0030] Furthermore, the virtual learner database DB2 stores the degree of understanding in each learning area of the virtual learner. The degree of understanding is determined based on, for example, the knowledge accumulated through interactions with the learner. The degree of understanding is stored in a subdivided manner for each of a plurality of units or categories included in the learning area. Also, the degree of understanding may be determined according to the correct answer rate calculated by comparing the accumulated knowledge with the data in the teaching material information database DB1, or according to the scoring result. That is, if the learner teaches correct knowledge to the virtual learner, the degree of understanding of the virtual learner also improves, and if the taught knowledge is incorrect or the learner cannot answer the questions from the virtual learner, the degree of understanding of the virtual learner does not improve. That is. According to such a configuration, the degree of understanding of the learner can be estimated through the degree of understanding of the virtual learner. Therefore, by determining the correctness of the learning content of the first learned model of the virtual learner possessed by the virtual learner database DB2, the degree of understanding of the user who is the real learner can be estimated. That is, even without conducting a test on the user, the degree of understanding of the user can be appropriately estimated.

[0031] Note that the knowledge stored in the virtual learner database DB2 may be associated and stored with the correct / incorrect results obtained by comparing with the data in the teaching material information database DB1. Further, the knowledge stored in the virtual learner database DB2 may be associated and stored with the correct knowledge information in the case where the content is incorrect. According to this configuration, an appropriate administrator or the guardian of the learner can specifically grasp the points where the learner's understanding is incorrect by referring to the virtual learner database DB2. When the content taught from the learner to the virtual learner is incorrect, it is possible to convey in the message from the virtual learner to the learner that the knowledge is incorrect. The learned model has obtained appropriate learning data in advance and can also perform additional learning as appropriate.

[0032] The virtual teacher database DB3 stores the words and phrases used in the teacher's speech. The virtual teacher database DB3 may have a learned model that takes a message from the learner or the virtual learner as input and outputs a message issued by the virtual teacher.

[0033] As shown in FIG. 1, the server device 10 mainly includes functional blocks such as a display control unit 11, a storage control unit 12, an AI control unit 13, a scoring unit 14, an artificial intelligence unit 15, and a communication processing unit 20, which are composed of a CPU (Central Processing Unit, an example of a processor in the claims), a computer program executed by the CPU, a RAM (Random Access Memory) or a ROM (Read Only Memory) that stores the computer program and predetermined data.

[0034] The display control unit 11 executes processing for displaying a screen related to the interactive learning system on the learner terminal 30. The display control unit 11 performs processing such as generating and transmitting an HTML (Hyper Text Markup Language) file, and causes the learner terminal 30 to display a web page as a screen for the learner to view when learning in this system. Note that the display control unit 11 may perform processing such as generating and transmitting display data for an application for using the interactive learning system.

[0035] The display control unit 11 causes the learner terminal 30 to display a chat screen on which messages from the virtual learner or the virtual teacher are displayed. The chat screen may be provided with a plurality of chat screens of different natures, such as a chat screen with the virtual learner, a chat screen with the virtual teacher, and a group chat screen where messages can be exchanged among three or more people including the virtual learner and the virtual teacher. A message recipient is associated with each of the plurality of chat screens, and the learner can switch the display of the received messages by switching the plurality of chat screens. At the same time, the learner can specify the message sender by switching the plurality of chat screens.

[0036] In addition, the display control unit 11 may cause the learner terminal 30 to display an understanding degree screen that displays the understanding degree of the virtual learner in a predetermined learning field. Further, the display control unit 11 may display the dialogue history with the virtual learner or the virtual teacher. The dialogue history may be, in addition to the content of specific messages, the category of the content of the dialogue, the learning field, the unit, or the summary of the message, etc.

[0037] The memory control unit 12 is a functional unit that controls an appropriate storage device provided in the server device 10 and performs writing and reading of data. The memory control unit 12 may store, for example, the dialogue history with the virtual learner or the virtual teacher, the history of questions presented to the learner and the answer history, and the history of correct and incorrect answers, in association with the identification information of the learner in the storage device.

[0038] The AI control unit 13 is a functional unit that controls the input to the artificial intelligence unit 15. The AI control unit 13 mainly generates an instruction representing outputting a message to be displayed on the learner terminal 30 as a statement of the virtual learner or the virtual teacher, using the information input by the learner as an input. The instruction may be, for example, a prompt, or a programming language, or may be in an appropriate format acceptable to the artificial intelligence unit 15. Note that the prompt is information that instructs the artificial intelligence unit 15 about the generated content, and is, for example, a character string described in natural language. With such a configuration, the virtual learner and the virtual teacher can make appropriate statements in response to the input of the learner.

[0039] The AI control unit 13 discriminates the destination of the input message of the learner from the type of the input chat screen, and instructs the artificial intelligence unit 15 to generate a message according to different learned models according to the destination. That is, the AI control unit 13 discriminates that the message input to the chat screen with the virtual learner is destined for the virtual learner, and generates an instruction representing outputting a message according to the first learned model. The AI control unit 13 discriminates that the message input to the chat screen with the virtual teacher is destined for the virtual teacher, and generates an instruction representing outputting a message according to the second learned model.

[0040] Further, the AI control unit 13 may further generate an instruction using time information as an input. With this configuration, the virtual learner and the virtual teacher transmit messages to the learner at appropriate timings even at timings when there is no utterance from the learner.

[0041] Furthermore, the AI control unit 13 may generate an instruction indicating that it outputs a message from the virtual learner to the virtual teacher (an example of the "fifth message" in the claims) using time information or a message uttered by the virtual teacher as an input. In this case, the AI control unit 13 generates an instruction for the artificial intelligence unit 15 to output a message according to the first learned model. Further, the AI control unit 13 uses, as an input, a message from the virtual learner to the virtual teacher, and generates an instruction indicating that it outputs a message from the virtual teacher to the virtual learner according to the second learned model.

[0042] The scoring unit 14 is a functional unit that scores the correctness of the content of the input message based on the second learned model. For example, the scoring unit 14 scores the correctness by comparing the history of the input message with the second learned content stored in the second learned model. According to this configuration, the scoring process can be executed regardless of the timing of fine-tuning. Further, the scoring unit 14 may give an answer instruction including a question to the artificial intelligence unit 15 to generate an answer message including an answer for the first learned model fine-tuned based on the input message, and score the correctness of the answer based on the second learned model. The question to be input may be given by an administrator or the like, or may be a question stored in advance in the server device 10 in association with the learning range. According to such a configuration, it is possible to easily set the learning range to be scored. Furthermore, the scoring unit 14 may score the correctness by comparing at least a part of the learned content of the first learned model stored in the virtual learner database DB2 with the learned content stored in the learned model stored in the teaching material information database DB1. The scoring result is stored in the virtual learner database DB2. Also, the scoring result is reflected in the degree of understanding stored in the virtual learner database DB2 for each of a plurality of units or categories included in the learning field.

[0043] The artificial intelligence unit 15 is an AI equipped with a language model such as a Transformer including BART (Bidirectional and Auto-regressive Transformer), BERT (Bidirectional Encoder Representations from Transformers), or GPT (Generative Pretrained Transformer, including GPT-1, GPT-2, GPT-3), especially a learning model such as a large language model (LLM). The learning model (also referred to as a machine learning model) refers to a learning model based on a machine learning algorithm. Specific algorithms for machine learning include the nearest neighbor method, the naive Bayes method, decision trees, and support vector machines. Also, deep learning (deep neural networks) that uses neural networks to generate its own feature quantities and connection weight coefficients for learning is included. The artificial intelligence unit 15 can appropriately apply the above algorithms.

[0044] The artificial intelligence unit 15 takes information related to the information in the designated learning field as input and generates a message (an example of the "third message" in the claims) containing questions related to the learning field for the user according to the first trained model. By the learner answering this message output to the output unit 31, the first trained model of the hypothetical learner is fine-tuned. Note that the timing of fine-tuning is not limited to immediately after obtaining a message from the learner and may be performed at any timing. The first trained model may be fine-tuned according to the history of questions asked to the learner and the answer history recorded in the memory control unit 12.

[0045] The artificial intelligence unit 15 generates a message of a hypothetical learner based on an instruction generated by the AI control unit 13, taking as input a message from the learner. The artificial intelligence unit 15 generates a message of the hypothetical learner (an example of the "first message" in the claims) according to the first learned model stored in the hypothetical learner database DB2. With such a configuration, the learner can communicate messages with the hypothetical learner, and through this communication, enhance the learning effect in a predetermined learning field. Also, when interacting with an actual friend, there may be a fear that one's own lack of understanding will be revealed. However, according to the interactive learning system, since it is an interaction with a machine, shame is reduced and active interaction is possible.

[0046] Also, the artificial intelligence unit 15 generates a message (an example of the "fourth message" in the claims) for teaching knowledge to the user as a lesson by a virtual teacher based on information related to the learning field, according to at least the second learned model stored in the teaching material information database DB1. The artificial intelligence unit 15 may also refer to the learned model stored in the virtual teacher database DB3. With such a configuration, the learner can receive a lesson from a virtual teacher who has machine-learned the content of the teaching material, and through this interaction, improve the academic ability in a predetermined learning field. Also, the artificial intelligence unit 15 generates a message of the virtual teacher based on an instruction generated by the AI control unit 13, taking as input the input message from the learner. According to this configuration, the learner can also communicate messages with the virtual teacher.

[0047] Also, the artificial intelligence unit 15 outputs a message (an example of the "second message" in the claims) based on the scoring result for the first learned model scored by the scoring unit 14. This message may be output only when the learning content of the first learned model is incorrect. Also, this message may be, for example, a text indicating to the user who is the learner that the taught content is incorrect, or a text containing correct content.

[0048] Note that the message based on the scoring result is not limited to the mode of being immediately output to the output unit 31 for the input message from the user who is the learner, and it is desirable to output it at an appropriate time. For example, it is better to output the second message after a few hours to a few days after the hypothetical learner receives instructions from the user, so that the hypothetical learner can notice the mistakes through tests at school or cram school, and a sense of presence as if teaching the user can be realized. The time from the reception of the input message to the output of the second message may be determined in advance to be constant, or may vary according to the field or difficulty level for which the correct answer is to be taught. It may also vary according to machine learning.

[0049] The artificial intelligence unit 15 generates, according to the instruction from the AI control unit 13, a message from the hypothetical learner to the virtual teacher and a message from the virtual teacher to the hypothetical learner, and displays them on the output unit 31. This message is, in other words, the dialogue between the virtual teacher and the hypothetical learner. According to such a configuration, the learner can observe while the hypothetical learner with insufficient understanding is asking questions, and can obtain an opportunity to confirm his or her own knowledge and understanding. Furthermore, a sense of presence as if participating in an actual class can be obtained. Also, since the hypothetical learner learns through interactions with the user and asks questions based on what has been learned, the learner also asks questions about points that were unclear or ambiguous. Furthermore, the hypothetical learner is highly likely to ask questions about the content that has been taught by the learner but for which the learning is insufficient. Therefore, the dialogue between the hypothetical learner and the virtual teacher becomes content tailored to each user, and can further promote the improvement of the user's academic ability.

[0050] The communication processing unit 20 is a processing unit that can execute data transmission and reception processing according to a predetermined protocol with the learner terminal 30 via a network NW such as the Internet. The communication processing unit 20 receives from the learner terminal 30 answers to the learning field and questions, messages to the hypothetical learner or messages to the virtual teacher, etc. Also, the communication processing unit 20 transmits the message uttered by the hypothetical learner or the virtual teacher, generated by the artificial intelligence unit 15, to the learner terminal 30.

[0051] ●Processing Flow (1): Processing flow when the learner is addressed from the virtual learner As shown in FIG. 2, first, the server device 10 refers to the virtual learner database DB2, refers to the understanding level of the virtual learner (step S101), determines units or the like to be the theme of the dialogue according to the understanding level, and then generates a message including a problem by the artificial intelligence unit 15 (step S102). Next, the server device 10 transmits this message to the learner terminal 30 as a message of a question from the virtual learner (step S103). The learner terminal 30 displays this message on the output unit 31 and accepts an input from the learner (step S104). The server device 10 records the information acquired via the learner terminal 30 (step S105). Next, the server device 10 transmits a reaction to the input message from the learner to the learner terminal 30 (step S106). When the conversation continues (N in step S107), the process returns to step S102, and questions conforming to the conversation from the virtual learner are continued. Note that the input from the user (step S104) and the transmission of the reaction from the virtual learner (step S106) may be performed at appropriate timings, and the process may be omitted. Also, the determination of whether the conversation continues in step S107 may be appropriately determined from the presence or absence of an input from the user, the time elapsed without an input from the user after the message is transmitted from the virtual learner, or the content of the conversation or the like.

[0052] When the server device 10 determines that the conversation ends in step S107 (Y in step S107), the server device 10 transmits, for example, a message notifying the end of the conversation to the learner terminal 30 (step S108), and ends the process. Fine-tuning of the first learning model may be performed at an appropriate timing during the above processing flow. For example, it may be performed while recording the information acquired via the learner terminal 30 in step S105, or may be performed in step S108.

[0053] ● Processing Flow (2): A processing flow that outputs the dialogue between the virtual learner and the virtual teacher This processing flow shows the state where the user as a learner is taking a lesson from the virtual teacher together with the virtual learner. As shown in FIG. 3, first, the server device 10 refers to the virtual learner database DB2, refers to the understanding level of the virtual learner (step S111), determines a unit or the like as the theme of the dialogue according to the understanding level, and then generates a message according to the second learned model by the artificial intelligence unit 15 (step S112). Next, the server device 10 transmits this message to the learner terminal 30 as a message from the virtual teacher as a lesson (step S113).

[0054] Also, taking the message from the virtual teacher as an input, a message according to the first learned model is generated by the artificial intelligence unit 15 and transmitted to the learner terminal 30 as a message from the virtual learner to the virtual teacher (step S114). Here, for example, via the server device 10, the user transmits a message to the virtual teacher (step S115). Next, when the conversation continues (N in step S116), the process returns to step S113 or step S114 as appropriate.

[0055] When the conversation ends (N in step S116), the server device 10 transmits, for example, a message notifying the end of the conversation to the learner terminal 30 (step S117). Or, the process may shift to step S112 or step S113, return to generating or transmitting a message as a lesson, and continue the lesson. In this way, a group lesson by the virtual teacher, as well as an immersive dialogue among the virtual teacher, the virtual learner, and the user can be carried out. Note that the order of speaking is an example and is not limited thereto.

[0056] ● Display Example FIG. 4 is an example of a screen displayed on the learner terminal 30, and is a screen showing the understanding level of the virtual learner. This screen is generated based on the understanding level information stored in the virtual learner database DB2. In the example of this figure, regarding the learning field of "history", knowledge about ancient Greece is out of 100 points, while knowledge about medieval Europe is 56 points. Also, regarding units such as "French Revolution" and "Napoleon" further subdivided in the category of medieval Europe, conversation topics with the virtual learner are displayed.

[0057] FIG. 5 is a diagram showing a specific example of a chat screen where the virtual learner and the learner are communicating. The balloon coming out from the icon of "AI" is the message sent by the virtual learner, and the balloon coming out from the icon of "You" is the message input by the learner. A specific chat interaction at a certain point in time is as follows, for example. AI "Napoleon came out in the test but I don't understand~. Who is Napoleon...?" User "According to my dictionary, he's the one who said 'There is no impossible letter for me'. From France." AI "He seems cool. His lines are so king-like." User "I don't think he was a king." AI "What kind of person?" User "A warrior-like person." AI "Is he a person who fights?" AI "So he was involved in wars and such, and he's not a modern person, right? What era was he from?" User "I saw it on a quiz show the other day. It said he became emperor in 1804." AI "Wow! So that's it."

[0058] After the above conversation, inside the dialogue-based learning system, a test to examine the understanding level of the virtual learner is conducted. Based on the results, the message content from the virtual learner is generated and sent to the learner at a later date. In this case, the chat with the learner is as follows, for example. AI: "I got it wrong when I wrote 'Napoleon's occupation: soldier'! But there was a time when he became emperor." User: "Did you write exactly'soldier' like that? ww" AI: "Because I said'soldier'." User: "So, what was the answer?" AI: "The correct answers were'military man' and'revolutionary'." User: "A revolutionary, huh."

[0059] The virtual learner presents a frank message as if it were a friend who is keeping up in learning as described above. Also, based on the instructions from the learner, the virtual learner learns that Napoleon's occupation is a soldier and that Napoleon became emperor in 1804. Then, it determines right or wrong and presents the result of right or wrong to the learner. Moreover, during the conversation, the virtual learner instructs the learner that the correct answer regarding Napoleon's occupation is a military man or a revolutionary. Through such a series of exchanges, the learner can obtain an opportunity to output by teaching the virtual learner, can fix knowledge, and can cultivate the ability to explain clearly. Also, by teaching the virtual learner, the learner can confirm the accuracy of their own knowledge, etc., and can obtain an opportunity to acquire correct knowledge. Furthermore, since learning can be continued while interacting with the virtual learner, motivation can be maintained. Additionally, since the virtual learner accumulates knowledge and grows through the conversation with the learner, attachment to the virtual learner and thus to the dialogue-based learning system is formed through the dialogue for cultivating the virtual learner. That is, according to this system, effective learning can be enjoyed and continued pleasantly.

[0060] As described above, according to the dialogue-based learning system according to the present invention, the user's academic ability improvement can be promoted through a dialogue full of a sense of presence with the system.

Explanation of Signs

[0061] 10 Server device (dialogue-based learning system) 11 Display control unit 12 Memory control unit 13 AI control unit 14 Scoring section 15 Artificial intelligence section 20 Communication processing section DB1 Teaching material information database DB2 Virtual learner database DB3 Virtual teacher database 30 Learner terminal

Claims

1. A dialogue learning system that receives an input message from a user who is a learner and displays a first message generated by artificial intelligence as a reply to the user from a virtual learner, wherein the artificial intelligence has a first trained model that uses the input message as an input and outputs the first message, and a second trained model that has learned the content included in one or more teaching materials through machine learning, the first trained model is fine-tuned by the input message, comprises a processor, and the processor in an input step, receives the input message input by the user, in a response step, gives a first instruction to the artificial intelligence to generate the first message according to the first trained model based on the input message, in a scoring step, scores the correctness of the content of the input message based on the second trained model, and in the response step, gives a second instruction to the artificial intelligence to generate a second message based on at least the scoring result in the scoring step, in an output step, outputs the first message output by the artificial intelligence based on the input message and the second message output by the artificial intelligence based on the second instruction, a dialogue learning system.

2. In the scoring step, an answer instruction to input a question to the fine-tuned first trained model and generate an answer message including an answer is given to the artificial intelligence, and the correctness of the answer is scored based on the second trained model. The dialogue learning system according to claim 1. The dialogue learning system according to claim 1.

3. In the scoring step, the correctness is scored by matching the history of the input message with the second learned content stored in the second trained model. The dialogue learning system according to claim 1. The dialogue learning system according to claim 1.

4. In the input step, further receives information related to the learning field, further, in a question posing step, gives a third instruction to the artificial intelligence to generate a third message including a question related to the learning field for the user according to the first trained model using at least the information related to the learning field as an input, in the output step, outputs the third message output by the artificial intelligence based on the third instruction. The dialogue learning system according to claim 1. The dialogue learning system according to claim 1.

5. In the input step, further, information regarding the learning field is received, In the class requirement step, the processor further issues a fourth instruction to the artificial intelligence to generate a fourth message for teaching knowledge to the user as a class by a virtual teacher based on the information regarding the learning field, at least according to the second learned model, In the output step, the fourth message output by the artificial intelligence based on the fourth instruction is output, The interactive learning system according to claim 1.

6. In the response step, further, a fifth instruction to generate a fifth message for the virtual teacher based on the fourth message is issued to the artificial intelligence according to the fine-tuned first learned model, In the output step, the fifth message output by the artificial intelligence based on the fifth instruction is output, The interactive learning system according to claim 5.

7. An interactive learning method for receiving an input message from a user who is a learner and displaying, as a reply to the user from a virtual learner, a first message generated by an artificial intelligence, The artificial intelligence has a first learned model that takes the input message as an input and outputs the first message, and a second learned model that has learned the content included in one or more teaching materials by machine learning, The first learned model is fine-tuned by the input message, By a computer, An input reception step of receiving the input message input from the user, A response step of issuing a first instruction to the artificial intelligence to generate the first message based on the input message, A scoring step of scoring the correctness of the content of the input message based on the second learned model, A response step of issuing a second instruction to the artificial intelligence to generate a second message based on at least the scoring result in the scoring step, An output step of outputting the input message output by the artificial intelligence based on the first instruction and the second message output by the artificial intelligence based on the second instruction, is executed, Interactive learning method.

8. A dialogue learning program that receives an input message from a user who is a learner and displays a first message generated by artificial intelligence as a reply to the user from a virtual learner, wherein the artificial intelligence has a first learned model that takes the input message as input and outputs the first message, and a second learned model that has learned the content included in one or more teaching materials through machine learning, the first learned model is fine-tuned by the input message, a computer, an input reception step of receiving the input message input from the user, a response step of giving a first instruction to the artificial intelligence to generate the first message based on the input message, a scoring step of scoring the correctness of the content of the input message based on the second learned model, a response step of giving a second instruction to the artificial intelligence to generate a second message based on at least the scoring result in the scoring step, an output step of outputting the input message output by the artificial intelligence based on the first instruction and the second message output by the artificial intelligence based on the second instruction, to execute, a dialogue learning program.

Images