Generative AI support system

Abstract

The goal is to provide effective educational support while preventing inappropriate use by students when using generative AI systems. [Solution] The generation AI support system 10 is a generation AI support system that controls whether or not to process a response based on a prompt input by a student to the generation AI, and is equipped with a subject classification unit 12 that performs subject classification for the prompt and calculates a relevance score for the input content, a score determination unit 14 that determines the processing category of the prompt based on the relevance score by comparing it with a predetermined threshold, a teacher approval control unit 16 that displays an approval interface to accept the teacher's approval when the score determination unit 14 determines that the prompt is pending, and an output control unit 18 that outputs the generation AI's response sentence based on the teacher 6's operation on the approval interface.

Description

[Technical Field]

[0001] The present invention relates to a generative AI support system. [Background technology]

[0002] In recent years, the use of generative AI in education has been attracting attention, and it is expected to be a means of supporting students' independent learning. However, there are also concerns about the risks of misuse and non-educational use, and there is a growing need for a system to manage and control the use of generative AI.

[0003] As a technology related to the present invention, for example, Patent Document 1 discloses a reflect generation AI system that is characterized by having the AI ​​memorize and store the thoughts of respondents in advance, and then outputting and answering the questions of inquirers such as voters or product purchasers. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2025-9955 Summary of the Invention [Problem to be solved by the invention]

[0005] In conventional generative AI systems, it was difficult to determine whether the prompts entered by students were appropriate for the learning objectives, making it difficult to respond to inappropriate questions or for teachers to intervene. Therefore, a system was needed that could evaluate the educational appropriateness of student input and, if necessary, control the AI ​​response with teacher approval.

[0006] The object of the present invention is to provide effective educational support while preventing inappropriate use by students when using a generative AI system. [Means for solving the problem]

[0007] The generative AI support system of the present invention is a generative AI support system that controls whether or not to process a response based on a prompt input by a student to the generative AI, and is characterized by comprising: a subject classification unit that performs subject classification on the prompt and calculates a relevance score of the input content; a score determination unit that determines the processing category of the prompt based on the relevance score by comparing it with a predetermined threshold; a teacher approval control unit that displays an approval interface to accept the teacher's approval when the score determination unit determines that the prompt is pending; and an output control unit that outputs the generative AI's response sentence based on the teacher's operation on the approval interface.

[0008] In addition, in the generation AI support system of the present invention, it is preferable that the system further includes an NG word filter unit that analyzes the words contained in the prompt and detects pre-set NG words, and that if the NG word filter unit contains the NG word, it immediately blocks the prompt without executing the response processing of the generation AI.

[0009] In addition, it is preferable that the generative AI support system according to the present invention further comprises a history storage unit that stores the prompt, the response sentence, the relevance score, and a history of the teacher's approval operation in chronological order.

[0010] In addition, it is preferable that the generative AI support system of the present invention further comprises a prompt visualization unit that displays the history information stored in the history storage unit in graph or table format and shows the teacher the input trends and changes in subjective interests of each student.

[0011] In addition, it is preferable that the generative AI support system according to the present invention further comprises an intervention support unit that analyzes the students for whom the teacher should intervene based on score trends based on the history information and presents the results as suggestion information.

[0012] In the generative AI support system according to the present invention, it is preferable that the subject classification unit calculates the similarity between the prompt and the group of subject-specific knowledge vectors using cosine similarity. [Effects of the Invention]

[0013] According to the present invention, the educational appropriateness of prompts entered by students can be automatically evaluated, and the responses of the generating AI can be controlled with teacher approval if necessary, thereby preventing inappropriate use and providing effective educational support. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a diagram showing a configuration of a generation AI support system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram showing an operational flow of a generation AI support system according to an embodiment of the present invention. [Figure 3] FIG. 1 is a diagram showing the relationship between a history storage unit and multiple components linked to it in a generative AI support system according to an embodiment of the present invention. [Figure 4] 1 is an example of an operation screen presented by a teacher approval control unit in a generation AI support system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following, similar elements in all drawings will be designated by the same reference numerals, and duplicate explanations will be omitted. Furthermore, in the description below, previously described reference numerals will be used as necessary.

[0016] FIG. 1 is a diagram showing the configuration of a generative AI support system 10 according to an embodiment of the present invention. FIG. 2 is a diagram showing the operational flow of the generative AI support system 10 according to an embodiment of the present invention. FIG. 3 is a diagram showing the relationship between a history storage unit 22 and multiple components linked thereto in the generative AI support system 10 according to an embodiment of the present invention. FIG. 4 is an example of an operation screen presented by the teacher approval control unit 16 in the generative AI support system 10 according to an embodiment of the present invention.

[0017] The generative AI support system 10 includes a subject classification unit 12, a score determination unit 14, a teacher approval control unit 16, an output control unit 18, an NG word filter unit 20, a history storage unit 22, a prompt visualization unit 24, an intervention support unit 26, and a memory unit 28. The generative AI support system 10 is connected to students 4 and teachers 6 via a network 2.

[0018] As shown in Figure 1, a student 4 uses his or her own terminal to input a prompt to the AI ​​generation support system 10. This prompt is an input sentence to receive a response from the AI ​​generation system. The input prompt is first sent to the subject classification unit 12, where it is classified.

[0019] The subject classification unit 12 evaluates the relevance of the prompt to predefined subject categories (e.g., Japanese, mathematics, science, social studies, etc.) based on the words and context in the prompt, and calculates a relevance score for each category. This relevance score may be calculated using cosine similarity to the knowledge vector group.

[0020] The score determination unit 14 compares the relevance score calculated by the subject classification unit 12 with a predetermined threshold value and determines the processing category of the prompt. Specifically, if the relevance score is equal to or greater than the threshold value, automatic processing is permitted, and if it is less than the threshold value, the prompt is put on hold or blocked.

[0021] If the score determination unit 14 determines that the request is pending, the teacher approval control unit 16 operates to present an approval interface to the teacher 6. The teacher 6 can confirm the contents of the prompt and perform an approval or rejection operation.

[0022] The teacher approval control unit 16 has a function to display an approval interface for use by the teacher 6 when a prompt determined to be "pending" by the score determination unit 14 is entered. The approval interface displays the full text of the prompt entered by the student 4, the subject classification result, the relevance score, and the recommended processing category (e.g., "approval required" or "automatic response not allowed"). The teacher 6 can approve or reject the prompt based on this information, thereby preventing inappropriate responses and ensuring that the prompt is used for educational purposes.

[0023] Based on the approval operation by the teacher 6, the output control unit 18 sends a prompt to the generation AI, obtains a response sentence, and outputs it. On the other hand, if approval is not granted, the generation AI does not respond. Note that if the relevance score of the score determination unit 14 is equal to or greater than the threshold and automatic processing is possible, a prompt is sent to the generation AI, obtains a response sentence, and outputs it.

[0024] The NG word filter unit 20 determines whether or not a prompt contains a preset NG word, and immediately blocks the prompt if one is included, thereby preventing inappropriate prompts from being input to the generation AI.

[0025] The history storage unit 22 records, in chronological order, the history of prompts, the response sentences of the generation AI, the relevance score, approval information from the teacher 6, etc. This makes it possible to later refer to and analyze the usage history and trends of the students 4.

[0026] The prompt visualization unit 24 visualizes the data recorded in the history storage unit 22 in graph or table format, allowing the teacher 6 to check the trends and interest trends of each student 4. Furthermore, the prompt visualization unit 24 has a trend classification display function that can display trends by subject using bar graphs, etc. The prompt visualization unit 24 also has a history management function that allows the teacher 6 to sort and reevaluate the displayed history information, allowing it to be used for individual support and feedback on teaching material design.

[0027] The intervention support unit 26 analyzes the history information, extracts students 4 who are judged to require attention from the teacher 6 based on score transitions and input trends, and presents the extracted information as suggestive information. This makes it possible to clarify the timing and target of instructional intervention by the teacher 6.

[0028] The storage unit 28 stores data and processing programs used by the various components. The storage unit 28 is configured as a remote or local storage medium and supports the operation of the entire generative AI support system 10.

[0029] In the generation AI support system 10, each function other than the storage unit 28 can be realized by either a hardware configuration or a software configuration. For example, when realized by software, these functions can be realized by application software stored in a recording medium such as RAM, ROM, or hard disk running on a server device that is actually configured with a CPU or MPU, RAM, ROM, etc.

[0030] Next, the operation of the generative AI support system 10 configured as above will be described.

[0031] Figure 2 shows the basic processing flow of this system. First, a prompt entered by a student 4 using a student terminal is sent to the NG word filter unit 20, which immediately checks for prohibited words. If an NG word is detected, a blocking process is executed and the prompt is discarded. If no NG word is detected, the prompt is sent to the subject classification unit 12, which calculates a relevance score with the subject based on embedding processing and cosine similarity.

[0032] The calculated relevance score is evaluated by the score determination unit 14 and compared with a predetermined threshold. If the relevance score is determined to be high, a prompt is sent to the generation AI, and a response sentence is created by the generation AI response processing unit. As a result, an output is presented on the display unit from the output control unit 18, and is recorded together with the output history in the history storage unit 22. On the other hand, if the relevance score is below the threshold, the process is put on hold, and the teacher approval control unit 16 transitions to a state where it waits for the teacher's decision.

[0033] The teacher approval control unit 16 presents the teacher 6 with an approval operation interface for the prompt. The teacher 6 can check the entered prompt text, the output content to be generated, the subject name, the relevance score, etc., and select the appropriate operation from "Approve," "Block," or "Re-input." In addition, a field is provided for entering comments, allowing feedback to be left for the student 4. If approved, the prompt is sent to the generation AI, and the output is created and displayed.

[0034] Figure 4 is an example of an operation screen presented by the teacher approval control unit 16. The screen displays the student ID, input prompt, relevance score, subject classification result, and judgment category, and the teacher can select from the options of "Approve," "Block," and "Re-enter." There is also a teacher comment section for feedback and history management, and all records are saved in the history storage unit 22.

[0035] Figure 3 shows the relationship between the history storage unit 22 and the multiple components that are linked to it. The log-saved data is visualized as a line graph or scatter plot by the prompt visualization unit 24, and the trend classification / bar graph display unit displays input trends by subject and category in a graph. In addition, the history table display unit allows a list of assessment statuses to be confirmed, making it easier for the teacher 6 to provide evaluation and instruction support. Based on this information, the intervention support unit 26 provides the teacher 6 with necessary suggestions and notifications.

[0036] The generative AI support system 10 of the present invention is equipped with an integrated function that instantly blocks NG words, evaluates scores based on subject relevance, allows approval by the teacher 6, visualizes the prompt history, and supports intervention, and analyzes and evaluates the input content of the student 4 from multiple angles, while promoting appropriate guidance and control by the teacher 6. This provides a practical and reliable support infrastructure for safely and effectively utilizing generative AI in educational settings.

[0037] With the above configuration, the generative AI support system 10 automatically evaluates whether the prompt entered by the student 4 is educationally appropriate, and by combining automatic judgment based on a threshold with approval from the teacher 6, appropriate control of the generative AI's responses is achieved. Furthermore, the function of immediately blocking prohibited words ensures safety, and the recording and visualization of history allows the teacher 6 to assist in learning guidance. Furthermore, the intervention support unit 26 suggests the need for individual instruction, making it possible to provide effective and detailed learning support. The configuration and effects of the generative AI support system 10 have been described in detail above. Below, examples are presented to further concretely understand the present invention.

[0038] Example 1 The generation AI support system 10 according to this embodiment is designed for educational support purposes for elementary and junior high school students, and is used by accessing a Python application built on Google Colaboratory from a student terminal.

[0039] When a question is input from a student terminal, the subject classification unit 12 performs morphological analysis and embedding-based vectorization, compares the vector with a group of knowledge vectors defined in advance for each subject, and calculates the semantic relevance using cosine similarity.

[0040] The score determination unit 14 controls the output control unit 18 in the following three stages according to the relevance score output by the subject classification unit 12. (1) Score is 0.7 or more: The output control unit 18 automatically approves the generated AI response. (2) Score is 0.4 or more but less than 0.7: Switch to teacher approval control unit 16, and the teacher selects "Approve," "Block," or "Request re-entry." (3) Score less than 0.4: Blocked immediately by the NG word filter unit 20

[0041] The prompt visualization unit 24 automatically adds educational guidance prompts, such as "You are a junior high school science subject advisor," based on the student's grade information. The history storage unit 22 saves prompts, responses, scores, and teacher operations and comments in Excel format to Google Drive. This leads to appropriate AI responses based on the student's 4 input, realizing a support environment in which teachers can safely and flexibly utilize AI in educational settings. Teachers can list, filter, and download this history, and comments added at the time of approval can also be used for students' review learning.

[0042] Example 2 In this embodiment, multiple operation modes are introduced into the generation AI support system 10, and the output control method can be flexibly switched according to the operational policy of the educational site.

[0043] Using an operation unit (not shown), the teacher 6 or the school administrator can select any of the following three operation modes. (1) Full approval mode: A configuration in which the output control unit 18 does not operate unless the teacher approval control unit 16 operates for all student inputs. (2) Inappropriate word filter combined mode: A configuration in which the immediate blocking mechanism by the inappropriate word filter unit 20 and the three-stage judgment by the score judgment unit 14 are combined. (3) Approval mode with time period and number of times limit: A configuration that dynamically switches between control based on the relevance score and control by the teacher approval control unit 16 depending on the usage time or number of inputs.

[0044] Settings are registered by school, grade, and individual student using the student management means, and setting changes are recorded and audited by the history storage unit 22. This allows for the creation of a management system for the use of generative AI that meets the diverse needs of educational settings, achieving both security and flexibility. The full approval mode is effective during information ethics education periods and in situations requiring strict operation in the early stages of implementation, and provides high control through the intervention of teachers 6. On the other hand, the NG word filter combined mode uses NG word dictionaries set by grade and subject, allowing for more flexible control. The time-zone and count-limited approval mode is configured to process some input at the students' 4's own discretion, limited to after-school or self-study periods. This is designed to promote independent learning through natural dialogue with the generative AI while preventing excessive or inappropriate use. Note that this example illustrates differences in the input processing policies of students 4 and does not include a configuration that changes the style or difficulty of the output sentences by the generative AI.

[0045] Example 3 The generation AI support system 10 according to this embodiment is implemented in correspondence with student terminals as mobile terminals such as smartphones and tablets.

[0046] The system was designed using Flutter, a cross-platform development environment, and is configured to operate regardless of the type of student device (iOS / Android). The input screen is composed of an operation unit (not shown), and in addition to a user interface for selecting the grade and purpose of use, it also implements support functions such as voice input, hiragana display, and kanji display with furigana.

[0047] The input question is vectorized by the subject classification unit 12 and compared with a group of knowledge vectors on the cloud. If the communication situation is unstable, simple processing by the NG word filter unit 20 can also be used. The output results can be switched between student and teacher display on the display unit, and if approval is required, a notification is sent from the teacher approval control unit 16, allowing the teacher 6 or parent / guardian to approve the result remotely.

[0048] All operations and inputs / outputs are recorded in the history storage unit 22, and data can be linked to LMSs and other devices. This provides a learning support environment where generative AI can be safely and flexibly utilized in environments such as at home or during supplementary lessons. When Student 4 launches the app, a screen appears prompting them to select their grade and intended use (homework support, self-study, teacher-accompanied mode, etc.), and the corresponding prompt template and score threshold are automatically set. After entering a question, prompts that require approval are not only notified, but also remain in a pending state until approval is complete. In addition to the student / teacher display, the display switching unit allows viewing of the entire history in parent mode, ensuring transparency in learning management at home. Furthermore, the history storage unit 22 has a function for switching between on-device storage and cloud synchronization. When synchronization settings are enabled, it can be integrated with Google Drive or the school's designated LMS for centralized management.

[0049] Example 4 The generation AI support system 10 according to this embodiment is configured to automatically switch the prompt template (role designation and tone guidance sentence) to be assigned to the generation AI depending on the grade of the student 4 and the subject of the question.

[0050] The student management unit associates the grade information of each student 4 with the user at the time of user registration, and automatically acquires this information when the prompt is entered. The subject classification unit 12 analyzes the content of the prompt entered by the student 4 and estimates the corresponding subject based on the relevance score. It is also possible to use a configuration in which the student 4 selects the subject directly from the operation unit.

[0051] The prompt visualization unit 24 has pre-registered prompt templates according to the grade and subject, and a prompt generation engine (not shown) automatically adds these templates to the input content of the student 4. For example, if the subject is science for fourth grade elementary school students, a prompt message such as "As a fourth grade science teacher, please answer in a way that is easy for students to understand" is added. If the subject is social studies for second grade junior high school students, a template such as "As a social studies teacher for second grade junior high school students, please provide a thorough explanation based on the textbook" is added.

[0052] These templates are customized to suit the level of textbook compliance and the style of expression for each subject (e.g., specific examples in science, causal relationships in social studies, and interpretations of expressions in Japanese), and teachers6 can edit and adjust them through the template management screen.

[0053] The template attached to the prompt is processed by the generation AI response control unit, and the generated response content is educationally appropriate and appropriately expressed. In addition, the history of the templates used is recorded in the history storage unit 22, which can be used by the teacher 6 via the intervention support unit 26 to check the effectiveness of the template and analyze areas for improvement.

[0054] With this configuration, even if the question content is the same, it is possible to optimize the response content of the generation AI according to the grade and subject of Student 4, thereby realizing individually optimized educational support.

[0055] Example 5 The generation AI support system 10 of this embodiment is intended to be introduced to multiple schools and classes, and is configured so that upper-level administrators such as boards of education, principals, and information education officers can collectively manage and apply output control policies, prompt settings, etc.

[0056] For this reason, the system has an administrator portal on the cloud that allows the following operations: (1) Setting similarity score thresholds for each school, grade, and subject (e.g., stricter for science, looser for social studies) (2) Editing and sharing settings for the dictionary used in the NG word filter unit 20 (3) Bulk switching of usage modes (e.g., switching the entire school to "Full Approval Mode" during exam periods) (4) Annual update of template prompts (e.g., reflecting compliance with new curriculum guidelines) (5) Select the log output format (CSV / Excel) and save destination (cloud / local)

[0057] These settings can be registered and applied individually for each school, grade, and subject, and changes are recorded and audited with a timestamp. In addition, settings made by a higher-level administrator are immediately reflected on the terminals of teachers 6, preventing deviations in operation at each site.

[0058] Furthermore, various settings are saved in the history storage unit 22 and can be used later for educational policy evaluation and educational research purposes. This configuration enables the formulation and operation of a unified AI educational policy at the local government level, and an educational support infrastructure that balances safety and educational validity can be built. Note that the administrator settings in this embodiment relate to output control and the application of template guidance sentences, and do not include configuration for manipulating the output content itself, such as the vocabulary and syntax of the generation AI.

[0059] Example 6 The generative AI support system 10 of this embodiment is configured to accommodate children and students in special needs classes or who require individual support, and by incorporating visual and linguistic support functions, it provides an inclusive learning environment for learners who have difficulty reading and writing.

[0060] The following support functions are integrated into the display and operation units, and can be enabled in stages according to the characteristics of the student 4. (1) Reading function: The prompt entered by Student 4 and the response from the generating AI are read out using a speech synthesis engine. The speed, voice quality, speech mode (for young children / standard), etc. can be changed from the settings screen. (2) Furigana display mode: In addition to automatically adding furigana to kanji in the output text, it can also convert expressions into "easy Japanese" using a thesaurus based on the level of vocabulary difficulty. This is done in conjunction with a context analysis engine (e.g., BERT-based) to achieve paraphrasing that maintains the naturalness of the expression. (3) Pictogram / illustration support: When the teacher 6 judges through the operation unit that "illustration is recommended," or when the output control unit 18 automatically judges based on the topic classification results, it has a function to automatically insert and display diagrams and icons in the corresponding response sentence.

[0061] In teacher mode, the support function can be enabled / disabled for each individual student and the level of support (e.g., reading "on / natural / emphasis") can be flexibly set through the student management means. The history of switching of the support function is recorded by the history storage unit 22, allowing data feedback for special support plans and individual teaching plans.

[0062] In this way, this configuration makes it possible to provide safe and educationally effective support for the use of generative AI for students who need assistance with visual comprehension, foreign students with limited Japanese language ability, or children who require special consideration for their developmental stage.

[0063] Example 7 The generation AI support system 10 according to this embodiment is configured to analyze usage history data of the student 4 and adaptively adjust the operation of the score determination unit 14 and the output control unit 18. This allows for optimal output control according to the student's learning tendency and maturity.

[0064] Specifically, the student management unit statistically analyzes data such as input history, score trends, approval results, usage time periods, and input frequency recorded in the history storage unit 22. Based on the analysis results, for example, for subjects where the student has a high level of understanding, more flexible prompt guidance sentences are added, while for subjects where the student is recognized to have a tendency to be weak, supplementary introductory sentences (for example, "First, let me explain why that question is important") are added, thereby strengthening individual support.

[0065] The score determination unit 14 automatically adjusts the threshold according to the learning progress and reliability, expanding the range of students for whom approval is not required for highly reliable students, while strengthening the approval requirements when NG words are detected or when a tendency for incorrect answers continues. Furthermore, the teacher approval control unit 16 can dynamically switch the priority of instructor notifications.

[0066] These adaptive control details are visualized to the teacher 6 through the prompt visualization unit 24, and the teacher can manually correct or intervene as necessary. In addition, a dashboard display is provided through parent mode, allowing the teacher to check the growth record and trends, and this also facilitates coordination with home learning.

[0067] In this way, adaptive output control based on history enables safe and flexible use of generative AI tailored to each individual student, thereby improving the quality of educational support.

[0068] Example 8 The generative AI support system 10 according to this embodiment has an educational support configuration that enables teachers 6 to accurately grasp students' learning tendencies and biases in their interests by analyzing and visualizing students' 4 question histories and subject-specific relevance scores, and to intervene in instruction at the appropriate time.

[0069] The relevance scores for each input calculated by the subject classification unit 12 and the score determination unit 14 are recorded as time-series data in the history storage unit 22. The prompt visualization unit 24 visualizes these score trends as line graphs or scatter plots, supporting trend analysis on a student or class basis.

[0070] Furthermore, the recorded prompts are automatically categorized by subject and theme, and the system has a function that can display the frequency of appearance and trend distribution as a bar graph. This allows teachers6 to visually grasp the distribution of question content and temporal bias, and quickly understand each student's learning attitude and areas of interest.

[0071] In addition, all question history is displayed in a table format, with each item showing the score, subject classification, teacher approval status, teacher comments, etc. This allows teachers6 to easily check and evaluate the history using the filter and sort functions, and use it as a teaching record.

[0072] The intervention support unit 26, in accordance with predetermined rules, issues an alert to the teacher 6 if an abnormality is found in the input trends of the student 4 (for example, consecutive blocks in the same subject, a sudden drop in score, etc.). This notification is sent by highlighting on the dashboard, by email, or through collaboration with the school management support system.

[0073] In this way, this configuration allows teachers6 to grasp learning situations objectively and in detail based on AI usage history, enabling them to implement early educational intervention. This improves the quality of learning support using generative AI and enhances the ability to address individual student learning challenges. [Explanation of symbols]

[0074] 2 Network, 4 Students, 6 Teachers, 10 Generative AI Support System, 12 Subject Classification Unit, 14 Score Judgment Unit, 16 Teacher Approval Control Unit, 18 Output Control Unit, 20 NG Word Filter Unit, 22 History Storage Unit, 24 Prompt Visualization Unit, 26 Intervention Support Unit, 28 Memory Unit.

Claims

1. A generation AI support system that controls whether or not to perform a response process based on a prompt input by a student to the generation AI, a subject classification unit that performs subject classification for the prompt and calculates a relevance score for the input content; a score determination unit that determines a processing category of the prompt by comparing the relevance score with a predetermined threshold based on the relevance score; a teacher approval control unit that displays an approval interface for receiving approval from a teacher when the score determination unit determines that the score is pending; an output control unit that outputs a response sentence of the generated AI based on the teacher's operation in the approval interface; A generative AI support system comprising:

2. The generative AI support system according to claim 1, The system further includes an inappropriate word filter unit that analyzes words included in the prompt and detects inappropriate words that have been set in advance, The generation AI support system is characterized in that the NG word filter unit immediately blocks the prompt without executing the generation AI's response processing if the NG word is included.

3. The generative AI support system according to claim 1, A generative AI support system further comprising a history storage unit that stores the prompt, the response sentence, the relevance score, and the history of the teacher's approval operation in chronological order.

4. The generative AI support system according to claim 3, A generative AI support system characterized by further comprising a prompt visualization unit that displays the history information stored in the history storage unit in graph or table format and displays to the teacher the input trends and changes in subjective interests for each student.

5. The generative AI support system according to claim 4, A generative AI support system characterized by having an intervention support unit that analyzes the students in whom the teacher should intervene based on score trends based on the historical information and presents the results as suggested information.

6. The generative AI support system according to claim 1, The subject classification unit calculates the similarity between the prompt and a group of subject-specific knowledge vectors using cosine similarity.

Images