Model generation system, method, and program

Abstract

The objective is to provide a model generation system, method, and program that can generate a base model that produces appropriate output according to the purpose. [Solution] The model generation system according to the present invention is characterized by comprising: a storage means for storing learning data used for learning and target information including the classification of the learning data in association with each other; and a generation means for generating a base model that has learned the degree of association between question information indicating a question and answer information indicating an answer to the question, based on the learning data and target information stored by the storage means.

Description

Technical Field

[0001] The present invention relates to a model generation system, method, and program.

Background Art

[0002] Currently, research is being conducted on the automatic generation of text using AI (Artificial Intelligence). In particular, by using a natural language processing model trained on a large amount of text data, for example, it is possible to classify the input text, analyze the sentiment from the input text, summarize the input text, or generate a response to the input question.

[0003] Patent Document 1 discloses a system that uses a natural language processing model to generate a natural language question using a first sentence representing a cause or result, extracts a candidate second sentence from candidate texts, and selects the candidate second sentence having the highest probability as one having a causal relationship with the first sentence, thereby generating a combination of a question and an answer.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] According to the generation system disclosed in Patent Document 1, by using a natural language processing model, a natural language question using a first sentence representing a cause or result is generated, a candidate second sentence is extracted from candidate texts, and the candidate second sentence having the highest probability is selected as one having a causal relationship with the first sentence, thereby generating a combination of a question and an answer.

[0006] On the other hand, in order to train foundational models such as natural language processing models with knowledge in a specific field, it is necessary to collect a large amount of data that is appropriate for that purpose. Furthermore, this data must be appropriate and consistent with the purpose. For this reason, data collection requires a lot of manpower and specialized skills, and it cannot be easily done with information that is updated daily. For this reason, it is necessary to collect accurate and high-quality data that is appropriate for the purpose.

[0007] However, the generation system disclosed in Patent Document 1 does not take into consideration, for example, the storage of training data suitable for the purpose of the base model. Therefore, the generation system disclosed in Patent Document 1 has the problem that it cannot generate a base model that produces an appropriate output according to the purpose.

[0008] Therefore, the present invention was devised in view of the above-mentioned problems, and its objective is to provide a model generation system, method, and program that can generate a base model that generates an appropriate output according to the purpose. [Means for solving the problem]

[0009] The model generation system according to the first invention is characterized by comprising: a storage means for storing learning data used for learning and target information including the classification of the learning data in association with each other; and a generation means for generating a base model that has learned the degree of association between question information indicating a question and answer information indicating an answer to the question, based on the learning data and target information stored by the storage means.

[0010] The model generation system according to the second invention is characterized in that, in the first invention, the storage means stores the learning data in association with purpose information including two or more classifications.

[0011] The model generation system according to the third invention is characterized in that, in the second invention, the storage means stores location information indicating the position of each of two or more classifications included in the target information in association with the learning data.

[0012] The model generation system according to the fourth invention is characterized in that, in the first invention, the storage means stores evidence information indicating the basis of the learning data in association with the learning data.

[0013] The model generation system according to the fifth invention further comprises an extraction means for extracting the target information based on user information relating to the user, and the storage means is characterized in that it stores the target information extracted by the extraction means in association with the learning data.

[0014] The model generation system according to the sixth invention further comprises a modification means for modifying the learning data based on the objective information in the first invention, and the storage means is characterized in that it stores the learning data modified by the modification means and the objective information linked together.

[0015] The model generation system according to the seventh invention is characterized in that, in the first invention, the generation means generates a large language model that learns the degree of association between the question information and the answer information based on the learning data and target information stored by the storage means.

[0016] The model generation method according to the eighth invention is characterized by causing a computer to perform a storage step of associating and storing learning data used for learning with objective information including the classification of the learning data, and a generation step of generating a base model that has learned the degree of association between question information indicating a question and answer information indicating an answer to the question, based on the learning data and objective information stored in the storage step.

[0017] The model generation program according to the ninth invention is characterized by causing a computer to execute a storage step of associating and storing learning data used for learning with objective information including the classification of the learning data, and a generation step of generating a base model that has learned the degree of association between question information indicating a question and answer information indicating an answer to the question, based on the learning data and objective information stored in the storage step. [Effects of the Invention]

[0018] According to the first to ninth inventions, the model generation system, method, and program of the present invention store training data and target information in association. This makes it possible to generate a base model that has learned appropriate training data for the target. Therefore, it is possible to generate a base model that produces appropriate output according to the target.

[0019] In particular, according to the second invention, the model generation system of the present invention associates and stores objective information, including two or more classifications, with training data. This makes it possible to generate a base model that produces output more specific to the objective.

[0020] In particular, according to the third invention, the model generation system of the present invention stores location information and training data in association. This makes it possible to automatically create a map of training data suitable for an objective by, for example, associating the training data with each objective in a multidimensional space where each objective is one dimension. Therefore, it becomes possible to automatically assign objective information to training data with high accuracy.

[0021] In particular, according to the fourth invention, the model generation system of the present invention stores rationale information linked to training data. This makes it possible to present the rationale for the output to the user when generating the output.

[0022] In particular, according to the fifth invention, the model generation system of the present invention extracts target information based on user information. This makes it possible to automatically add target information suitable for the user to the training data.

[0023] In particular, according to the sixth invention, the model generation system of the present invention modifies the learning data based on the target information. As a result, it becomes possible to automatically obtain learning data suitable for the target.

[0024] In particular, according to the seventh invention, the model generation system of the present invention uses a large language model. As a result, it becomes possible to learn a model from a vast amount of data. Therefore, it becomes possible to generate support information more flexibly.

Brief Description of the Drawings

[0025] [Figure 1] FIG. 1 is a schematic diagram showing an example of the model generation system in the present embodiment. [Figure 2] FIGS. 2(a) to 2(b) are schematic diagrams showing an example of the configuration of the model generation system in the present embodiment. [Figure 3] FIG. 3 is a schematic diagram showing an example of the detailed configuration of the model generation system in the present embodiment. [Figure 4] FIG. 4 is a flowchart showing an example of the operation of generating a generation model of the model generation system in the present embodiment. [Figure 5] FIG. 5 is a schematic diagram showing how target information is imparted to the learning data of the model generation system in the present embodiment. [Figure 6] FIG. 6 is a schematic diagram showing an example of the operation of the generation model that generates support information of the model generation system in the present embodiment. [Figure 7] FIG. 7 is a flowchart showing an example of the operation of generating support information of the model generation system in the present embodiment.

Modes for Carrying Out the Invention

[0026] The following describes in detail an example of a model generation system 100, a model generation method, and a model generation program as embodiments of the present invention, with reference to the drawings. Note that the configurations in each figure are schematically represented for illustrative purposes, and the actual dimensions of each component, such as size comparisons between components, may differ from those shown in the figures.

[0027] An example of the configuration of the model generation system 100 in this embodiment will be described with reference to Figures 1 to 3.

[0028] The model generation system 100 includes, for example, a user terminal 1, a server 2, and a communication network 9, as shown in Figure 1. The model generation system 100 generates a base model such as a large-scale language model. The model generation system 100 generates a base model that generates support information for supporting user U's learning, for example, based on learning status information regarding user U's learning status. The model generation system 100 also receives input from user U via the user terminal 1 and obtains queries to be executed against a database that stores pre-stored base models. Subsequently, by executing the obtained queries, the model generation system 100 can generate answer information indicating the answer to the queries based on the model such as the large-scale language model.

[0029] User U is a user of the model generation system 100. User U may also be a learner who is the target of learning support provided by the generated model 71 generated using the model generation system 100.

[0030] In this embodiment, a database for storing models such as large-scale language models is stored on server 2, and the model generation system 100 generates models on server 2 is used as an example for explanation, but the system is not limited to this configuration. The model generation system 100 may also have a database for storing large-scale language models stored on user terminal 1, and generate models on user terminal 1, in which case server 2 may not be provided.

[0031] <User Terminal 1> User terminal 1 is the terminal that controls the generation of the model. User terminal 1 is the terminal operated by user U. User terminal 1 communicates with server 2 via communication network 9.

[0032] User terminal 1, as shown in Figure 2(a), for example, comprises a housing 10, a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a storage unit 104, and I / Fs 105 to 107. Each component 101 to 107 is connected by an internal bus 110. As user terminal 1, known electronic devices such as laptop PCs, desktop PCs, tablet terminals, and smartphones can be used.

[0033] The CPU 101 controls the entire user terminal 1. The ROM 102 stores the operating code for the CPU 101. The RAM 103 is a working area used when the CPU 101 is operating. The storage unit 104 stores various information such as backups of the data stored in the ROM, databases, and learning target data. As the storage unit 104, a data storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive) can be used. For example, the user terminal 1 may have a GPU (Graphics Processing Unit) which is not shown.

[0034] I / F105 is an interface for sending and receiving various information with Server 2 via the communication network 9 as needed. I / F106 is an interface for sending and receiving information with Input Unit 108. For example, a keyboard or mouse can be used as Input Unit 108, and User U inputs various information via Input Unit 108. I / F107 is an interface for sending and receiving various information with Display Unit 109. Display Unit 109 displays various information stored in Storage Unit 104, or generation results, etc. A display can be used as Display Unit 109, and if it is a touch panel type, it is provided together with Input Unit 108. Display Unit 109 displays various information generated by each configuration of the model generation system 100.

[0035] The display unit 109 displays various information such as answer information. The display unit 109 may also display learning materials as support information on a monitor or the like. The display unit 109 may also display learning materials converted into a format that user U can understand, for example, using a display format pre-stored in the storage unit 104 or the like.

[0036] User terminal 1, as shown in Figure 3 for example, comprises an acquisition unit 11, a determination unit 12 connected to the acquisition unit 11, a generation unit 13 connected to the determination unit 12, and a presentation unit 14 and a storage unit 15 connected to the generation unit 13. Each configuration of user terminal 1 is realized by the CPU 101 executing programs stored in ROM 102, storage unit 104, etc., using RAM 103 as a working area.

[0037] <Acquisition part 11> The acquisition unit 11 acquires various types of information. The acquisition unit 11 transmits the information to, for example, the server 2. The acquisition unit 11 transmits the information acquired, generated, evaluated, or stored by, for example, the user terminal 1 to the server 2 via the communication network 9. The acquisition unit 11 receives the information acquired, generated, or stored by, for example, the server 2 via the communication network 9.

[0038] Furthermore, the acquisition unit 11 acquires generation requests, which are queries (query data) that request the generation of models such as large-scale language models. The acquisition unit 11 acquires queries that request the generation of models such as large-scale language models by, for example, receiving input from user U via the input unit 108.

[0039] <Judgment part 12> The decision unit 12 adds target information to the training data of the generative model 71.

[0040] <Generation part 13> The generation unit 13 executes generation requests acquired or generated by each component of the model generation system 100. The generation unit 13 also refers to the database 7 where the training data is stored and generates the generated model 71. Furthermore, the generation unit 13 executes generation requests acquired by the acquisition unit 11 to the database 7, which contains pre-stored models such as large-scale language models for generating support information, as described later.

[0041] <Presentation part 14> The display unit 14 displays various types of information. The display unit 14 may, for example, display response information on the display unit 109.

[0042] <Storage section 15> The memory unit 15 stores various types of information. It may also store the database 7 and the generation model 71.

[0043] <Server 2> Server 2 stores a database containing pre-stored models, such as large-scale language models. Server 2 communicates with user terminal 1 via a communication network 9. Server 2 performs information generation, storage, and transmission / reception in response to processing requests received from user terminal 1 via the communication network 9.

[0044] Server 2 comprises, for example, a chassis 20, a CPU 201, a ROM 202, a RAM 203, a storage unit 204, and an I / F 205, as shown in Figure 2(b). Each component 201-205 is connected by an internal bus 210. As Server 2, a well-known electronic device such as a laptop PC or a desktop PC can be used.

[0045] The CPU 201 controls the entire server 2. The ROM 202 stores the operating code for the CPU 201. The RAM 203 is a working area used when the CPU 201 is operating. The storage unit 204 stores various information such as backups of the data stored in the ROM, databases, and learning target data. As the storage unit 204, a data storage device such as an HDD or SSD can be used. For example, the server 2 may also have a GPU (not shown). The I / F 205 is an interface for sending and receiving various information with the user terminal 1 via the communication network 9 as needed.

[0046] <Communication Network 9> The communication network 9 is, for example, the Internet network to which user terminal 1 and server 2 are connected via a communication circuit. The communication network 9 may consist of a so-called optical fiber communication network. In addition, the communication network 9 may be implemented using known communication technologies such as wired communication networks or wireless communication networks.

[0047] <Database 7> Database 7 is a collection of data pre-stored in the model generation system 100. Database 7 is pre-stored, for example, in a storage unit 204 of server 2. Database 7 stores, for example, the generated model 71.

[0048] Next, we will describe the generated model 71 generated using the model generation system 100 in this embodiment.

[0049] <Generative Model 71> The generative model 71 is a foundational model that takes question information as input and outputs answer information indicating the answer to the question. The generative model 71 is a type of natural language processing model used, for example, by the model generation system 100 to automatically generate support information. The generative model 71 may also be a known large-scale language model (LLM) that has been pre-machine-trained using a large amount of text data.

[0050] A large-scale language model (LLM) is a deep learning model that pre-trains a language model, which models human spoken language based on its probability of occurrence, using a vast amount of data. In other words, a large-scale language model is a natural language processing model trained using a large amount of text data, taking a sentence as question information and outputting a sentence as answer information. When a large-scale language model is applied to a question-and-answer system, a question sentence is input to the LLM as question information, and the LLM outputs an answer sentence as answer information. The large-scale language model statistically estimates the probability of generating the next word from the sentence contained in the received prompt and sends the estimation result to the requester.

[0051] When a large-scale language model receives text data (a prompt), it uses the model to statistically estimate the probability of generating the next word from the text contained in the received prompt, and outputs a response based on the estimation result. As a large-scale language model, publicly known technologies can be used, for example, as described on internet sites such as "https: / / chatgpt-lab.com / n / n418d3aa56f0b" and "https: / / agirobots.com / chatgpt-mechanism-and-problem / ".

[0052] Specifically, the generative model 71 may include GPT-3, GPT-3.5, or GPT-4 related to "GPT (Generative Pre-trained Transformer) (registered trademark)", "BERT (Bidirectional Encoder Representations from Transformers)", "LaMDA (Language Model for Dialogue Applications)", "PaLM (Scaling Language Modeling with Pathways)", "LLaMA (Large Language Model Meta AI)", etc.

[0053] Furthermore, the generative model 71 may be a base model. The base model may be a natural language model. The base model may be a model generated by unsupervised learning. The base model may also be a generative AI. The natural language model may be an interactive, chat-type or conversational model that alternately receives question information such as instructions and generates answer information such as responses.

[0054] Furthermore, the generative model 71 may be generated using machine learning, for example, a neural network model. The generative model 71 is, for example, an AI neural network. The generative model 71 is trained using machine learning, for example, a neural network model such as a CNN (Convolutional Neural Network), or any arbitrary model may be used. In addition, the generative model 71 may be generated using methods such as Retrieval-Augmented Generation (RAG), Seq2Seq (Sequence To Sequence) linear discrimination, support vector machines, k-nearest neighbors, random forests, deep learning, etc.

[0055] In such cases, the generative model 71 stores a correlation, which is, for example, the degree of correlation between the input data, which is question information, and the output data, which is answer information. In such cases, morphemes or words included in the question information may be used as input data. The degree of correlation indicates the degree of connection between the input data and the output data; for example, a higher degree of correlation indicates a stronger connection between the data. The degree of correlation may be expressed as a percentage or a value of three or more values ​​or a value of three or more levels, or it may be expressed as a value of two values ​​or a value of two levels.

[0056] Question information is information indicating a question or instruction, and may be in text or audio format. It may also be a prompt. Furthermore, it may be learning status information indicating the learning progress. It may also be image or video information input by the user. It may also be information obtained by morphological analysis of a question. It may also be audio or image format converted into text format using speech recognition or image recognition. It may also be image format converted into text format, such as the names of objects or places, based on features calculated using R-CNN (Region Based Convolutional Neural Networks), YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), etc. Furthermore, it may contain queries requesting processing.

[0057] The answer information is information that shows the answer or response to the question contained in the question information. The answer information may also be support information related to learning support. In addition, the answer information may include information that shows the basis for the answer. Information that shows the basis for the answer may be, for example, the contents of a paper or patent document. In addition, the answer information may be information obtained by morphological decomposition of the answer.

[0058] Next, we will describe the details of the process for generating the generative model 71. Figure 4 is a flowchart showing the process for generating the generative model 71.

[0059] <Training data acquisition step S1> First, in the learning data acquisition step S1, the acquisition unit 11 acquires learning data. The learning data may be information about knowledge related to an event. The learning data may also include, for example, information on past learning status and support information. The learning data may also be information related to natural science. The learning data may be a tool that performs the aforementioned simulations related to natural science based on the input parameters. As for simulations related to natural science, tools capable of simulating physical phenomena, chemical phenomena, and even all kinds of natural sciences may be used, such as a FEM simulator that analyzes stress fields and strains based on the finite element method, chemical reaction simulations, chemical engineering simulations including plants, electrical circuit simulations, fluid dynamics simulations, quantum mechanics simulations, etc. The learning data may also consist of keyword information indicating keywords and related information related to those keywords. The keyword information may be, for example, information indicating a word, but it may not be limited to words and may include information indicating any word such as a person, place name, and event. The related information is information related to the word indicated by the keyword information. The related information may be, for example, an explanation and description of the word indicated by the keyword information, or questions and answers related to the word indicated by the keyword information. The acquisition unit 11 may store the acquired learning data in any memory, such as the storage unit 15.

[0060] <Step S2 for obtaining objective information> Next, in the objective information acquisition step S2, the acquisition unit 11 acquires objective information including the classification of the learning data. The objective information is information including the classification of the learning data, and may include, for example, information to identify the classification of the learning data, such as the subject, the name of the qualification or examination, the field, the era, the person, the question format, the answer format (presence or absence of choices, presence or absence of explanations, etc.), the number of questions, the difficulty level, and educational classifications such as primary education, secondary education, and higher education under the school system corresponding to the curriculum guidelines. The objective information may also be words such as "second year of junior high school" or "mathematics". Furthermore, the objective information may also be information indicating the purpose of the learning data. The purpose of the learning data may be one or more words such as "education", "business", "politics", "tourism", or "thesis" that identify the final goal of learning. In this case, the objective information may be a string of characters such as, for example, "to acquire basic knowledge for taking a qualification exam," "to be able to cover frequently asked questions on qualification exams," "to deepen understanding of subjects that are easy to stumble over through repeated study," "to teach history to junior high school students," "to write a paper," "to pass on lessons to future generations," "to apply to business," "to apply to politics," or "to plan a guided sightseeing tour."

[0061] Furthermore, the target information may include location information for each classification. Location information indicates the location of the training data within a single classification, and may, for example, be information indicating coordinates in a multidimensional space where each of two or more classifications is one dimension. Also, location information may indicate each of two or more classifications, such as "Junior High School 2nd Year" and "Mathematics." Furthermore, location information may indicate the range within each classification. For example, location information may indicate the range within each classification, such as "Junior High School 1st Year" to "Junior High School 3rd Year" or "Mathematics" to "Natural Sciences."

[0062] Furthermore, in the objective information acquisition step S2, the acquisition unit 11 may acquire user information relating to user U who uses the generated model 71 generated by the model generation system 100 to which an embodiment of the present invention is applied. User information may also be information indicating the user's attributes, such as the user's age, gender, and affiliation.

[0063] Furthermore, in the objective information acquisition step S2, the acquisition unit 11 may acquire evidence information that indicates the basis for the learning data. Evidence information is information that indicates the source, publication, etc., of the learning data.

[0064] <Purpose information assignment step S3> Next, in the objective information assignment step S3, the judgment unit 12 assigns objective information to the learning data. For example, as shown in Figure 5, the judgment unit 12 assigns objective information such as "3rd year junior high school" and "mathematics" to the learning data A. The judgment unit 12 may also assign location information in a two-dimensional space consisting of grade and field to the learning data A. The judgment unit 12 may also assign location information indicating the range of each classification, such as "1st year junior high school" to "3rd year junior high school" and "mathematics" to "natural science," to the learning data A. The judgment unit 12 associates the objective information assigned to the learning data and stores it in the memory unit 15. The judgment unit 12 may also further associate the rationale information with the learning data and store it in the memory unit 15.

[0065] Furthermore, the decision unit 12 may extract target information from the learning data based on user information. For example, the decision unit 12 may extract educational classifications such as primary education, secondary education, and higher education under the school system that correspond to the curriculum guidelines of the target information, according to the age included in the user information, and assign the target information to the learning data. In such a case, the decision unit 12 may extract the target information based on the user information using a relationship table that shows the relationship between user information and target information acquired in advance. Furthermore, the decision unit 12 may correct the learning data according to the target information. In such a case, the decision unit 12 may refer to a modified model that has been trained using modified learning data, which takes reference learning data and target information as inputs and outputs the corrected learning data, and correct the learning data in the learning data acquisition step S1 based on the acquired learning data and target information.

[0066] Furthermore, the decision unit 12 may refer to a classification model trained using pre-trained data that takes reference training data as input and target information as output, and in the training data acquisition step S1, it may add target information to the acquired training data. Reference training data is training data acquired in advance for use in training the classification model. Furthermore, the decision unit 12 may refer to a classification model trained using pre-trained data that takes reference training data and user information as input and target information as output, and based on the training data acquired in the training data acquisition step S1 and the user information acquired in the target information acquisition step S2, it may add target information to the training data. The decision unit 12 stores the training data with the added target information in the database 7. Furthermore, the decision unit 12 may store the training data in the database 7 separately for each type of target information.

[0067] <Generation Step S4> Next, in generation step S4, the generation unit 13 refers to the database 7 in which the learning data associated with the target information in the target information assignment step S3 is stored, and generates a generation model 71. The generation unit 13 stores the generated generation model 71 in the database 7, for example. The generation unit 13 may also generate a generation model 71 for each piece of target information. For example, the generation unit 13 may refer to the database 7 in which the learning data with "first year of junior high school" assigned as target information is stored, and generate a generation model 71, and so on, and generate multiple generation models 71 for each piece of target information, such as the generation model 71 generated by referring to the database 7 in which the learning data with "third year of junior high school" assigned is stored, and so on.

[0068] The above steps complete the process of generating the generative model 71. Furthermore, the generative model 71 is not limited to the above configuration; it may be generated using any equipment. This allows for the generation of the generative model 71 using training data linked to the target information, making it possible to generate a generative model 71 that produces appropriate output depending on the objective.

[0069] Next, an example of the generated model 71 generated by the model generation system 100 will be described. As an example, an example of the operation in which support information is generated as answer information based on learning status information input as question information using the generated model 71 will be described, but the system is not limited to this, and answer information may be generated based on any question information. The model generation system 100 is executed, for example, via a learning support program installed in the user terminal 1 and server 2.

[0070] First, the various types of information associated with the operation of the model generation system 100 to generate response information in this embodiment will be described. The various types of information handled by the model generation system 100 include, for example, query D1 and generated data D71, as shown in Figure 6. The model generation system 100 has a database 7 stored in advance.

[0071] <Query D1> Query D1 is information that includes a processing request to a database 7 in which generative models 71, such as large-scale language models, are stored. Query D1 includes, for example, support instruction information D11 and learning status information D12. Support instruction information D11 and learning status information D12 may be linked to each other, for example.

[0072] Query D1 is a prompt containing instructions for processing, for example, for generative model 71, and consists of a tag specifying the objective and natural language specifying the content of the instructions. A possible combination of objective and content is, for example, objective " <system>Education for junior high school students. The instructions were: <user>The string contains the combination of "Please explain the Battle of Sekigahara" and "...". By executing query D1 containing this combination, it is possible to generate support information including questions, options, answers, and explanations related to Japanese history, for example.

[0073] Furthermore, the instructions may include a request for assistance in generating a problem set consisting of multiple problems, and may also include the title of the problem set, the chapter structure of the problem set, the goals or objectives of each chapter, keywords for each chapter, the number of problems in each chapter, and the output format (such as HTML format).

[0074] <Support instruction information D11> Support instruction information D11 is information indicating a request for the generation of support information among the processing requests. Support instruction information D11 constitutes, for example, part of the string of query D1. Support instruction information D11 corresponds to, for example, "Please explain." in the prompt example sentence above.

[0075] <Learning Status Information D12> Learning status information D12 is information about user U's learning status. Learning status information D12 may include, for example, one or more types of information such as learning content information about what user U has learned, performance information about user U's grades, and history information about user U's learning history. Learning status information D12 may also be user information about user U's attributes, such as age, gender, and hobbies. Learning status information D12 may also be information indicating the purpose of the support information requested by support instruction information D11. Learning status information D12 may constitute part of the string of query D1, for example. Learning status information D12 consists of one or more words that specify the purpose of the support information, such as the subject, the name of the qualification or examination, the field, the era, the person, the question format, the answer format (presence or absence of choices, presence or absence of explanations, etc.), the number of questions, the difficulty level, and the educational classification in the school system, such as primary education, secondary education, and higher education, corresponding to the curriculum guidelines. Learning status information D12 includes, for example, "education for junior high school students" and "Battle of Sekigahara" in the prompt example sentences mentioned above.

[0076] Furthermore, the learning status information D12 may be, for example, one or more words that identify the ultimate learning goal. In this case, the learning status information D12 may be a string of characters such as "acquire the basic knowledge necessary to take the qualification exam," "be able to cover frequently asked questions on the qualification exam," or "deepen understanding of subjects that are difficult to study through repeated learning."

[0077] The learning status information D12 may use, for example, one or more words input by user U via the input unit 108. The support instruction information D11 may use, for example, a syntax pre-stored in user terminal 1. In other words, the model generation system 100 can obtain query D1 and generate support information simply by having user U input one or more words. In this case, even user U with limited expertise regarding query D1 can automatically generate support information. This improves the efficiency of generating support information.

[0078] Learning content information includes information about the subjects studied by user U, the scope of study, and the content of the learning materials. Grade information includes information about user U's grades, such as test scores. History information includes information about user U's learning history, such as the usage history and conversation history of the application and the model generation system 100 to which the present invention is applied, and the content that user U has learned.

[0079] <Generated Data D71> Generated data D71 shows the data generated based on query D1. Generated data D71 includes, for example, support information D711.

[0080] Support information D711 is information related to learning support. Support information D711 includes one or more types of information from the following: learning material information, problem information, evaluation information regarding user U, and explanation information regarding explanations to user U. Learning material information refers to information about learning materials such as workbooks and tests.

[0081] Learning materials refer to tests that assess user U's knowledge, including either questions or answers, and include question formats such as written questions, multiple-choice questions, and matching questions. Learning materials may also be texts that provide explanations, such as textbooks. Learning materials may also be texts that include keywords representing words such as names of people, places, and events, and explanations related to those keywords. Learning materials also include any educational materials that are output to user U as visual or auditory information. Examples of learning materials include learning material data generated by the model generation system 100, images and videos displayed on a display based on the learning material data, 3D images and 3D videos displayed via a hologram display based on the learning material data, printed materials printed based on the learning material data, and audio played back via audio equipment based on the learning material data.

[0082] Next, we will describe in detail the operation of generating support information as response information using the generated model 71 generated by the model generation system 100 in this embodiment. Figure 7 is a flowchart of the operation of generating support information. The operation of the model generation system 100 to generate support information includes, for example, a learning status acquisition step S11, a type discrimination step S12, a support generation step S13, and a presentation step S14, as shown in Figure 7.

[0083] <Preparation> Before the model generation system 100 performs the operation to generate support information, user U confirms that the database 7 containing the generated model 71 is saved.

[0084] <Learning status acquisition step S11> In the learning status acquisition step S11, the acquisition unit 11 receives input from user U via the input unit 108 and acquires a query D1 that includes support instruction information D11 and learning status information D12. In addition, in the learning status acquisition step S11, the acquisition unit 11 may automatically acquire learning status information D12 transmitted from, for example, server 2, and generate support instruction information D11 based on the acquired learning status information D12. In the learning status acquisition step S11, the acquisition unit 11 may automatically generate support instruction information D11 if, for example, the performance indicated by the performance information included in the automatically acquired learning status information D12 falls below a threshold, or if the period during which user U has not been learning, as indicated by the history information, exceeds a standard value. In addition, in the learning status acquisition step S11, the acquisition unit 11 may acquire user information.

[0085] <Type determination step S12> In the type determination step S12, the determination unit 12 determines the type of information to be included in the support information to be generated, based on the learning status information D12 acquired in the learning status acquisition step S11. The determination information is information relating to the type of information to be included in the support information. The determination information is information indicating the type of information, such as teaching material information, problem information, evaluation information, and explanation information. The determination information may also be information indicating the type of format, such as text, audio, or image. In the type determination step S12, for example, the determination unit 12 may determine the determination information based on the learning status information, using a relationship table that shows the relationship between the previously acquired learning status information and the determination information.

[0086] Furthermore, the decision unit 12 may refer to a decision model trained using decision learning data that takes reference learning status information as input and decision information as output, and make a decision on the decision information based on the learning status information acquired in the learning status acquisition step S11. The reference learning status information is learning data acquired in advance for use in training the decision model. Furthermore, the decision unit 12 may refer to a decision model trained using decision learning data that takes reference learning status information and user information as input and decision information as output, and make a decision on the decision information based on the learning status information acquired in the learning status acquisition step S11 and the user information. In addition, the process of determining this decision information may be performed by the generation unit 13 using the generation model 71.

[0087] <Support generation step S13> In the support generation step S13, the generation unit 13 inputs the learning status information obtained in the learning status acquisition step S11 into the generation model 71, which has been trained using learning data including the learning status information and support information, and generates support information. In the support generation step S13, the generation unit 13 executes the query D1 obtained in the learning status acquisition step S11 against the database 7. The generation unit 13 may automatically execute the query D1 obtained by the acquisition unit 11 when the acquisition unit 11 obtains the query D1, or it may execute the query D1 according to input from the user U. The generation unit 13 may also perform a query execution step separately before the type determination step S12, distinct from the type determination step S12.

[0088] When query D1 is executed against database 7, the generation unit 13 refers to the generation model 71 stored in database 7 and generates generated data D71, including support information D711, based on the learning status information D12. In this case, the generation model 71 can automatically generate support information data D711 based on the learning status information D12. This makes it possible to generate appropriate support information according to the learning status.

[0089] Furthermore, the generation unit 13 may input the judgment information determined in the type determination step S12 and the learning status information acquired in the learning status acquisition step S11, and generate support information. In this case, the types of information included in the generated support information will include the types of information indicated by the judgment information. This makes it possible to generate more appropriate types of information according to the learning status of user U.

[0090] Furthermore, the generation unit 13 may extract objective information based on the learning status information and determine which generation model 71 to reference according to the extracted objective information. In this case, the generation unit 13 may extract objective information from user information, words that identify the final goal, etc., included in the learning status information, and determine which generation model 71 to use for generating support information from among the multiple generation models 71 generated in the generation step S4 according to the extracted objective information. This makes it possible, for example, if user U is a third-year junior high school student, to reference a generation model 71 trained with learning data that has objective information for a third-year junior high school student attached, and generate support information. This makes it possible to generate support information that is more suitable for the user. Furthermore, the generation unit 13 may generate support information based on the learning status information and user information. In this case, the generation unit 13 may input the learning status information obtained in the learning status acquisition step S11 into a generation model 71 trained using learning data that includes learning status information, support information, and user information, and generate support information.

[0091] Furthermore, the generation unit 13 may generate generated data D71 that includes support information D711 and justification information D712. This makes it possible to generate justification for the support, enabling support that is more tailored to the user U. Justification information D712 is information that shows the justification for the support in support information D711, and may be information that shows, for example, the source and publication location of the explanatory information in support information D711. The generation unit 13 outputs the generated generated data D71 to the presentation unit 14.

[0092] <Presentation Step S14> In presentation step S14, the presentation unit 14 presents to the user U via the display unit 109 or the like when the user terminal 1 receives the generated data D71 generated by the generation unit 13 via the communication network 9.

[0093] Furthermore, in presentation step S14, the presentation unit 14 may present a warning to user U based on the support information of the generated data D71 generated by the generation unit 13. For example, if the evaluation information included in the support information indicates an evaluation below a threshold, the presentation unit 14 may present a warning to user U. This enables proactive support for user U.

[0094] After performing each of the steps described above, the operation of generating support information using the generated model 71 generated by the model generation system 100 in this embodiment is completed. Note that the model generation system 100 may, for example, repeat each of the steps described above. This makes it possible to generate appropriate types of information according to the learning status of user U.

[0095] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. [Explanation of symbols]

[0096] 100 Model Generation Systems 1 User terminal 10 cabinets 101 CPU 102 ROM 103 RAM 104 Preservation Department 105~107 I / F 108 Input section 109 Display section 110 Internal bus 11 Acquisition Department 12 Judgment Department 13 Generation part 14 Presentation section 15 Storage section 2 servers 20 cabinets 201 CPU 202 ROM 203 RAM 204 Preservation Department 205 I / F 210 Internal bus 7 Databases 71 Generative Models 9. Communication Network U User S1 Training data acquisition step S2 Step to obtain objective information S3 Purpose Information Assignment Step S4 Generation Step S11 Steps to acquire learning status S12 Type determination step S13 Support Generation Step S14 Presentation Step D1 Query D11 Support instruction information D12 Learning Status Information D13 Judgment Information D71 Generated Data D711 Support Information D712 Supporting Information< / user> < / system>

Claims

1. A storage means for storing learning data used for learning and target information including the classification of the learning data, The system comprises a generation means that generates a base model that learns the degree of correlation between question information indicating a question based on the learning data and target information stored by the storage means, and answer information indicating the answer to the question. A model generation system characterized by the following:

2. The storage means stores the learning data by associating it with target information that includes two or more classifications. A model generation system according to claim 1, characterized by the following:

3. The storage means stores location information indicating the location of each of the two or more classifications included in the target information, linked with the learning data. The model generation system according to claim 2, characterized by the following:

4. The storage means stores evidence information that indicates the basis of the learning data, linked to the learning data. A model generation system according to claim 1, characterized by the following:

5. The system further comprises extraction means for extracting the aforementioned target information based on user information relating to the user, The storage means stores the learning data in association with the target information extracted by the extraction means. A model generation system according to claim 1, characterized by the following:

6. The system further comprises a correction means for correcting the learning data based on the aforementioned objective information, The storage means stores the learning data corrected by the correction means in association with the target information. A model generation system according to claim 1, characterized by the following:

7. The generation means generates a large language model that learns the degree of association between the question information and the answer information based on the learning data and target information stored by the storage means. A model generation system according to claim 1, characterized by the following:

8. A memory step in which learning data used for learning is linked to target information including the classification of the learning data and stored; The computer is to perform a generation step that generates a base model that learns the degree of correlation between question information indicating a question and answer information indicating the answer to the question, based on the learning data and target information stored in the memory step. A model generation method characterized by the following.

9. A memory step in which learning data used for learning is linked to target information including the classification of the learning data and stored; The computer is to perform a generation step that generates a base model that learns the degree of correlation between question information indicating a question and answer information indicating the answer to the question, based on the learning data and target information stored in the memory step. A model generation program characterized by the following.

Images