Text generation system

Abstract

The purpose is to improve the efficiency of writing. [Solution] The text generation system is a text generation system that generates observation texts to be included in evaluation documents of students being evaluated in schools, and comprises an evaluation input unit into which evaluation result data of the students being evaluated is input, a storage unit that stores a collection of example observation texts, and a text generation unit that inputs the evaluation result data and the collection of example texts into a large-scale language model and causes the large-scale language model to output observation texts of the students being evaluated.

Description

Technical Field

[0001] The present invention relates to a text generation system for generating texts of documents used in schools.

Background Art

[0002] In primary and secondary schools, teachers evaluate children or students (hereinafter referred to as evaluation targets) and create evaluation documents such as notice sheets or guidance records. For example, a notice sheet notifies the evaluation targets and their guardians of the situation at school of the evaluation targets, along with the grades of each subject. The evaluation document contains "observations" by the teacher. The "observations" are described as texts and are written in a predetermined observation column of the evaluation document. The "observations" described in the notice sheet need to be texts that are accurate and inspiring to the evaluation targets and their guardians, but it was necessary to make the amount of text written in the observation column approximately the same for multiple evaluation targets. Therefore, teachers had to edit the texts for adjusting the number of characters, which sometimes increased the working hours.

[0003] Conventionally, a notice sheet creation support system for reducing the workload of teachers is known. The notice sheet creation support system includes "a means for registering and storing a plurality of example texts of observations in association with the number of characters of each example text of observations, a means for registering and storing the maximum inputable number of characters of an observation text in association with each type of observation, a means for registering and storing the number of characters already input of the inputted observation text in association with each child / student and the type of observation, a means for referring to the maximum inputable number of characters of the observation registered in the above-mentioned means for storing the settings for inputting observations and the number of characters already input of the above-mentioned observation registered in the above-mentioned means for storing the number of characters already input, performing an arithmetic operation of subtracting the number of characters already input from the maximum inputable number of characters, and calculating the inputable number of characters of the observation text for each child / student, a means for extracting an example text of observations within the above-mentioned means for storing example texts of observations that is less than or equal to the inputable number of characters, and a means for outputting the extracted example text of observations to a display unit" (see Patent Document 1).

Prior Art Documents

Patent Documents

[0004] [Patent Document 1] Patent No. 7052946 [Overview of the project] [Problems that the invention aims to solve]

[0005] However, teachers had to write comments for a large number of students, and the task of writing comments for numerous comments sections, as well as adjusting the length of each comment to a certain range, was time-consuming. Furthermore, since the comments needed to be consistent with the evaluation results for each subject, there was a challenge in that creating and checking them accurately required a great deal of effort.

[0006] This invention aims to solve the above-mentioned problems and to enable efficient creation of school documents. [Means for solving the problem]

[0007] According to the present invention, a text generation system is provided for generating observational texts to be included in evaluation documents of students being evaluated in schools, comprising: an evaluation input unit into which evaluation result data of the students being evaluated is input; a storage unit for storing a collection of example observational texts; and a text generation unit that inputs the evaluation result data and the collection of example texts into a large-scale language model and causes the large-scale language model to output observational texts of the students being evaluated.

[0008] According to the present invention, it is possible to efficiently create documents such as report cards produced in schools by utilizing a large-scale language model. [Brief explanation of the drawing]

[0009] [Figure 1] This diagram shows the overall configuration of the text generation system 100 according to Embodiment 1. [Figure 2] This is a block diagram showing an example of the hardware configuration of the information processing device 10 of the text generation system 100 according to the present invention. [Figure 3] This figure shows the detailed configuration of the functional blocks in the information processing device 10 of the text generation system 100 according to the present invention. [Figure 4] This figure shows an example of the input screen for the text generation system 100 according to the present invention. [Figure 5] This figure shows an example of the output screen and editing screen of the text generation system 100 according to the present invention. [Figure 6] This figure shows an example of the input screen when generating observations for "foreign language activities" using the text generation system 100 according to the present invention. [Figure 7] This figure shows an example of the input screen when generating "moral" observations in the text generation system 100 according to Embodiment 1. [Figure 8] This figure shows an example of the input screen when generating observational text related to "integrated learning time" using the text generation system 100 according to Embodiment 1. [Figure 9] This figure shows an example of the input / output screen during the conversion and editing of observation text in the text generation system 100 according to Embodiment 1. [Figure 10] This figure shows an example of the settings screen for the text generation system 100 according to Embodiment 1. [Figure 11] This shows an example of the information processing flow in the text generation system 100 of the present invention. [Figure 12] This figure shows an example of a prompt sent to the LLM server 20 in the text generation system 100 according to Embodiment 1. [Figure 13] This figure shows an example of a prompt sent to the LLM server 20 in the text generation system 100 according to Embodiment 1. [Figure 14] This is an example of a document format created using the output of the document generation system 100. [Figure 15] This is an example excerpt from a collection of example sentences used in the text generation system 100. [Figure 16] This figure shows an example of the input screen when generating a "class / school newsletter" using the text generation system 100 according to Embodiment 2. [Figure 17] An example of an input screen when generating a guidance text by the text generation system 100 according to Embodiment 2 is shown. [Figure 18] An example of an output screen when generating a guidance text by the text generation system 100 according to Embodiment 2 is shown.

Mode for Carrying Out the Invention

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Various characteristic matters shown in the following embodiments can be combined with each other. Also, an invention can be established independently for each characteristic matter.

[0011] Embodiment 1. 1. Configuration of the text generation system 100 1.1 Overall configuration of the text generation system 100 FIG. 1 is a diagram showing the overall configuration of the text generation system 100 according to Embodiment 1. The text generation system 100 includes an information processing device 10, an LLM (Large Language Model) server 20, a user terminal 30, and a network 90 connecting these. The text generation system 100 supports, for example, in elementary, junior high, and senior high schools, when a teacher creates the text of "observations" to be described in a notice form or a guidance record of the evaluation target. Basically, a teacher who is a user of the text generation system 100 connects to the information processing device 10 using the user terminal 30. The user inputs data such as the grades of the evaluation target and setting information for generating an observation text into the information processing device 10. The information processing device 10 generates a prompt based on the data input by the user and inputs it to the LLM server 20. The LLM server 20 outputs an observation text to the information processing device 10 from the input data using a large language model. The user checks the observation text output using the user terminal 30 and appropriately edits it and describes it in an evaluation document such as a notice form or a guidance record. Note that the user can use the text generation system 100 not only by connecting to the information processing device 10 using the user terminal 30 but also by directly operating the information processing device 10.

[0012] The information processing device 10 is a device that performs main processing for generating a findings text. Specifically, it includes an evaluation input unit 511, a storage unit 52, and a text generation unit 513, and plays a role of receiving evaluation result data of an evaluation target person and generating a findings text in cooperation with the LLM server 20.

[0013] The LLM server 20 is a dedicated server for executing a large language model, and generates a findings text based on the data (evaluation result data and example sentence set data) transmitted from the information processing device 10. The generated text is returned to the information processing device 10 via the network 90.

[0014] The input data received by the LLM server 20 is generally referred to as a "prompt". The "prompt" in the present embodiment is input data for giving an arbitrary instruction to the LLM server 20. The prompt can adopt an arbitrary form such as a sentence or a chart, and an arbitrary instruction, command, question, etc. can be set. When the prompt is represented by a sentence, various languages can be used, and any language readable by the LLM server 20 can be adopted. For example, in the embodiment, as the language for describing the prompt, any natural language, any programming language, any database language, any markup language or Markdown language, or any style sheet language (such as CSS) can be adopted.

[0015] The specific configuration of the LLM server 20 is not particularly limited, but any large-scale language model, known as generative AI, can be used. As an example of a large-scale language model, ChatGPT from OpenAI may be used. The LLM server 20 may employ only one type of large-scale language model (for example, only ChatGPT), but is not limited to this, and may employ a combination of multiple large-scale language models. The LLM server 20 may also adopt a form that links with a database, such as RAG (Retrieval Augmented Generation). In this embodiment, program code for building the system desired by the user is efficiently generated by optimizing the prompts to be input to the LLM server 20 (for example, the order in which to input).

[0016] The user terminal 30 is a device for users, such as teachers, to access the text generation system 100. The user terminal 30 can take the form of a PC, tablet, or smartphone, and connects to the information processing device 10 to provide an interface for inputting evaluation result data or for reviewing and editing generated observation texts.

[0017] Network 90 is a communication network connecting the information processing device 10, the LLM server 20, and the user terminal 30, and includes wired or wireless communication means. Data is exchanged bidirectionally via Network 90.

[0018] In the document generation system 100 according to Embodiment 1, evaluation result data of the person being evaluated is input from the user terminal 30 to the information processing device 10, and the information processing device 10 transmits this data to the LLM server 20. The observation document generated by the LLM server 20 is then displayed on the user terminal 30 via the information processing device 10. This series of processes enables the user to generate observation documents quickly and efficiently.

[0019] 1.2. Hardware configuration of the text generation system 100 Figure 2 is a block diagram showing an example of the hardware configuration of the information processing device 10 of the text generation system 100 according to the present invention. The information processing device 10 comprises at least a control unit 51, a storage unit 52, and a communication unit 53. The information processing device 10 may also comprise an input unit 54 and an output unit 55.

[0020] The control unit 51 is the main component that controls the entire information processing device 10 and is responsible for executing functions including the evaluation input unit 511 and the text generation unit 513. For example, the control unit 51 acquires evaluation result data and example sentence data stored in the storage unit 52 and generates an opinion text based on this data.

[0021] The storage unit 52 stores evaluation result data of the person being evaluated, sample data of observation statements, and various data necessary for the operation of the system. The storage unit 52 may include, for example, a hard disk, SSD, or memory device. Alternatively, the storage unit 52 may be an external storage unit (for example, an external storage medium, cloud storage, etc.).

[0022] The communication unit 53 provides an interface for communicating with other devices via the network 90. ​​For example, the communication unit 53 sends and receives data with the LLM server 20 and is responsible for generating observation documents and receiving results.

[0023] The input unit 54 is an interface for receiving evaluation result data and setting information from an external device such as a user terminal 30. The input unit 54 may include a keyboard, touch panel, or dedicated input device. The output unit 55 is an interface for presenting the generated observation document to the user. The output unit 55 includes means for transmitting data to the user terminal 30 via a display, printer, or network.

[0024] As shown in Figure 2, the information processing device 10 smoothly executes the process of generating observation documents by coordinating these components. In this embodiment, it is possible to streamline the entire process from inputting evaluation result data to outputting observation documents.

[0025] 1.3. Functional Block Configuration of Text Generation System 100 Figure 3 shows the detailed configuration of the functional blocks in the information processing device 10 of the text generation system 100 according to the present invention. The information processing device 10 is composed of the following parts, centered around the control unit 51 and the storage unit 52.

[0026] The control unit 51 includes an evaluation input unit 511, a setting unit 512, a text generation unit 513, a text output unit 514, an editing unit 515, and a conversion unit 516. The above functional units may be implemented by software or by hardware. When implemented by software, various functions can be realized by the CPU executing a computer program. The program may be stored on a non-temporary recording medium that can be read by the computer, provided for download from an external server, or implemented by so-called cloud computing, which reads a program stored in an external memory unit to realize the function. When implemented by hardware, it can be implemented by various circuits such as ASICs, FPGAs, or DRPs. In this embodiment, various information and concepts encompassing it are handled, which are represented by high or low signal values ​​or qubits as a set of binary bits composed of 0s or 1s, and communication and calculations can be performed by the above-mentioned software or hardware configuration. The software may be a general-purpose OS or a dedicated OS.

[0027] The evaluation input unit 511 has the function of inputting evaluation result data of the person being evaluated and comment data of the teacher. This data is used as the main input data for generating the observation document. The evaluation input unit 511 is also responsible for receiving data from the user terminal 30.

[0028] The settings unit 512 has a function for inputting the number of characters in the generated text, common phrases, and specific setting information. The settings unit 512 receives the information necessary for generating the prompt and reflects it in the processing of the text generation unit 513.

[0029] The text generation unit 513 has the function of generating prompts based on evaluation result data and configuration information and sending them to the LLM server 20. Specifically, as shown in Figures 12 and 13, it generates prompts structured in JSON format and inputs them to the LLM server 20. In addition, when extracting appropriate individual example sentences from the example sentence database (example sentence DB) 521, AWS Lambda may be used. Lambda provides a serverless, event-driven execution environment and can execute code only when needed. This mechanism allows for efficient and flexible extraction of example sentences and prompt generation, resulting in the rapid and appropriate generation of observational texts.

[0030] The text output unit 514 displays the text output from the LLM server 20 on the user terminal 30. It can also display a list of the generated texts.

[0031] Editorial Unit 515 provides a function to re-edit the generated observation text. In addition to manual editing, Editorial Unit 515 also supports interactive editing using the LLM Server 20 again.

[0032] The conversion unit 516 has the function of converting the writing style of the generated text. For example, it can convert from a polite style ("desu / masu") to a more formal style ("dearu"), depending on the intended use. The conversion unit 516 may also send the text data to the LLM server 20 and convert the writing style using LLM.

[0033] The storage unit 52 includes a sentence example database (sentence example DB) 521 and a prompt database (prompt DB) 522. The sentence example DB 521 stores sentence examples used for creating observational texts. The prompt DB 522 stores prompts corresponding to various modes and provides them to the text generation unit 513.

[0034] In this embodiment, the data input in the evaluation input unit 511 is optimized for generating observation texts via the setting unit 512 and sent to the LLM server 20 as a prompt by the text generation unit 513. The generated text is presented to the user by the text output unit 514 and modified as needed by the editing unit 515. Furthermore, by using the conversion unit 516, conversion according to writing style and purpose becomes possible.

[0035] This configuration allows the present invention to reduce the workload of teachers and efficiently generate appropriate commentary texts based on evaluation results.

[0036] 2. Operation of the text generation system 100 2.1. Input screen for generating the "Overall Findings" report. Figure 4 shows an example of the input screen of the text generation system 100 according to the present invention. This screen is provided as an interface for users, such as teachers, to input data on the person being evaluated and generate an evaluation text. In the input screen shown in Figure 4, the mode for generating an "overall evaluation" text is selected in the mode selection field 40. The area for specifying generation conditions and the area for inputting student evaluations are configured to allow input of setting information for generating an "overall evaluation" text.

[0037] As shown in Figure 4, the input screen has a mode selection field 40 on the left side. The user selects one of the modes from the mode selection field 40 to switch the mode for generating the text, and then inputs the necessary setting information and evaluation results. The modes for generating the text are "General Observations," "Moral Education," "Foreign Language Activities," and "Conversion / Editing" from the mode selection field 40, and "Settings" is used to configure common settings for each text generation mode. The following describes the screen, operation, and processing displayed by the text generation system 100 when generating "General Observations," based on Figure 4.

[0038] The area for specifying generation conditions is an input area for enabling the functions of the text generation unit 513. This area includes fields for inputting the purpose of use of the observation text 31, selecting the grade level 32, setting the number of people to be evaluated 33, setting the character count range 34, and setting common text 35a, and is used to set the conditions necessary for text generation. When the user clicks the generation START button 25, the entered conditions are sent to the LLM server 20, and text generation begins.

[0039] The input field 31 for the purpose of use of the observation document allows selection, for example, whether it is for report cards or student records. Generally, report cards are evaluation documents intended for children or students and their guardians to see, and the observations are written in a polite, formal style. Student records are evaluation documents mainly referenced within the school, and the observations are written in a more formal, objective style. The setting unit 512 sends setting information regarding evaluation documents such as report cards or student records to the document generation unit 513, and the document generation unit 513 generates prompts regarding the style of the observation document to be generated.

[0040] The grade selection field 32 is a field for specifying the grade level of the person being evaluated, and information such as elementary school grades 1-6 and junior high school grades 1-3 can be selected. The setting unit 512 sends the information set in the grade selection field 32 to the text generation unit 513, and the text generation unit 513 extracts example sentences related to the target grade level from the example sentence collection DB 521 and generates a prompt based on those example sentences.

[0041] The number of people to be evaluated setting field 33 is for setting the number of people to be evaluated. Specifically, it is used to set the number of entries for "people to be evaluated" in the evaluation data input area 41, which will be described later.

[0042] The character count range setting field 34 is a field for setting the number of characters corresponding to the comments section of evaluation documents such as report cards. For example, in report cards, the amount of comments should generally be as equal as possible for each of the multiple people being evaluated, so the character count range for the comments can be set. The character count range setting field 34 is configured to be set in the form of "~ characters" as an example, but it may also be configured to set upper and lower limits for the number of characters, or to set only a lower limit, or to set only an upper limit. The character count range setting information is sent from the setting unit 512 to the document generation unit 513, and the document generation unit 513 generates a prompt regarding the setting of the character count range for the generated comments.

[0043] The common text setting field 35a is a field for setting terms or sentences that are to be inserted in common into the observation documents of multiple evaluators. As an example, in the common text setting field 35a shown in Figure 4, you can specify text that you want to insert in common to a class, such as "At the sports day, the class worked together as one." This setting information regarding the common text is sent from the setting unit 512 to the document generation unit 513, and the document generation unit 513 generates a prompt so that the common text is included in the generated observation documents. In relation to the common text setting field 35a, there is an insertion position setting field 35b. The insertion position setting field 35b is configured so that, for example, "beginning" or "end" can be selected. However, the insertion position setting field 35b may be configured to allow settings other than "beginning" and "end". The settings in this insertion position setting field 35b are also reflected in the prompts of the document generation unit 513.

[0044] The above settings are input fields for enabling the functions of the setting unit 512 shown in Figure 3, and are used to finely customize the text generation based on the evaluation results. For example, it is possible to set terms to be inserted in common in the generated text or to set specific text formats. The set information is reflected in the prompt generation by the text generation unit 513 and in the input to the LLM server 20.

[0045] The evaluation data input area 41 is a table located below the label "<Student Evaluation Input>" in Figure 4. The evaluation data input area 41 is an input area that enables the evaluation input unit 511 shown in Figure 3 to function, and is used to input evaluation data and comments for the person being evaluated (student). Specifically, evaluation results for behavior, lifestyle, and each subject (e.g., Japanese language, mathematics, science, etc.) can be entered. This data is used as basic data for generating observation documents. The evaluation data input area 41 is a spreadsheet-type table, and is configured so that users can enter comments or evaluations in each cell as appropriate. The data entered by the user in the evaluation data input area 41 is sometimes called "evaluation result data". In addition, among the data entered by the user in the evaluation data input area 41, evaluation result data for, for example, only a specific subject or field may be called "individual evaluation result data".

[0046] Furthermore, the evaluation data input area 41 can be switched between displaying each item in a collapsed state and displaying each item in an expanded state. For example, the evaluation data input area 41 shown in Figure 4(a) is shown in a collapsed state. The evaluation data input area 41 shown in Figure 4(b) is shown in an expanded state. For example, in Figure 4(a), the fields where teachers can freely enter comments are not displayed in the "Student's Lifestyle" and "Study Status" sections of the evaluation data input area 41, but in Figure 4(b), the comment entry fields 38a and 39a are displayed.

[0047] The evaluation data input area 41 includes, for example, a number field 36, a field for recording the child's behavior 37, a field for recording their daily life 38, and evaluation fields 39 for each subject. The fields for recording the child's behavior 37, the field for recording their daily life 38, and the evaluation fields 39 for each subject each have a field for entering evaluation results expressed in a graded format such as "○", "×", "A", "B", and "C", and a field where the user can freely write comments. These fields can be folded to hide them, expanded to show them, or their display can be switched.

[0048] The number field 36 is a field for distinguishing the individuals being evaluated using numbers. The number field 36 may be configured to allow input of the names of the individuals being evaluated, but from the perspective of protecting personal information, it is desirable to configure it to distinguish individuals using numbers or other identifiers. For example, the user identifies the individuals being evaluated by associating the attendance numbers of the students in the class with the numbers in the number field 36. The number field 36 is created in accordance with the number of individuals to be evaluated set in the number of individuals to be evaluated setting field 33 in the generation condition specification area. For example, if the number of students is set to 23 in the number of individuals to be evaluated setting field 33, rows from No. 1 to 23 will be generated in the evaluation data input area 41. In Figure 4, only rows No. 1 to 3 are displayed, but the evaluation data input area 41 is configured to allow scrolling to display any portion of No. 1 to 23.

[0049] The evaluation result data entered in the evaluation data input area 41 is sent from the evaluation input unit 511 to the document generation unit 513, and the document generation unit 513 generates a prompt to be input to the LLM server 20 based on the evaluation result data.

[0050] As described above, each area of ​​the input screen shown in Figure 4 is linked to a specific functional block (evaluation input unit 511, setting unit 512, text generation unit 513, conversion unit 516), enabling efficient and flexible generation of observation texts. By operating the generation START button 25, each setting information and evaluation data is sent to the LLM server 20 in real time, realizing a rapid text generation process.

[0051] 2.2. Output and editing screen for generating the "Overall Findings" report. Figure 5 shows an example of the output screen and editing screen of the text generation system 100 according to the present invention. This screen displays the generated observation text to the user and provides an interface for editing as needed.

[0052] The output screen shown in Figure 5(a) provides interfaces for implementing the following functions.

[0053] The list display area 42 for example observations is a display unit for realizing the functions of the text output unit 514, and displays the observation sentences generated by the LLM server 20 in a list format. In the list display area 42 for example observations, the generated observation sentences corresponding to the numbers in the number field 36 of the evaluation data input area 41 shown in Figure 4 are displayed in field 42a. Next to the list display area 42 is a character count display area 43, which displays the number of characters in the generated observation sentences. In addition, field 42a of the list display area 42 where the observation sentences are displayed is provided with a scroll bar 42b for scrolling the display when the entire observation sentence cannot be displayed. This scroll bar 42b does not need to be displayed when the entire observation sentence can be displayed.

[0054] The copy operation field 44 provides a function to individually copy the observation text generated for each person being evaluated. For example, it is used when transferring observation text generated by a teacher to report card data.

[0055] The regeneration operation field 45 is for performing the operation of regenerating the observation text generated for each person being evaluated. For example, if you want to regenerate the observation text generated for person No. 1, click the regeneration button corresponding to person No. 1 in the regeneration operation field 45. The text generation unit 513 will then extract the necessary data from the evaluation input unit 511, the setting unit 512, the example sentence collection DB 521, and the prompt DB 522 to generate a prompt, which will then be sent back to the LLM server 20 to retrieve the observation text generated by the large-scale language model. The retrieved observation text is processed by the text output unit 514 and displayed in field 42a of the list display field 42.

[0056] The output operation field 48 is an operation button for outputting the observation text displayed in the list display field 42 as data. When the user clicks it, for example, an Excel data file is output that contains the contents displayed in the number field 36 and the list display field 42 in a table format. In this embodiment, the output operation field 48 is a button that outputs Excel data, but it may also be set to output CSV data, Word data, etc.

[0057] The batch copy operation field 49 provides a function that allows users to copy and transcribe the observation text displayed in the list display field 42. When a user clicks this, the observation texts displayed in the list display field 42 are copied to the clipboard in bulk.

[0058] The list display area 42 may have checkboxes corresponding to the number area 36. Each checkbox has a function that links to the "regeneration operation area 45" and the "bulk copy operation area 49". The checkboxes are for the user to select the observation documents they wish to regenerate, and are configured so that only the checked observation documents can be regenerated by the regeneration operation area 45. For example, if the user wishes to regenerate the observation documents for evaluation subjects No. 1, No. 3, and No. 5, they check the corresponding checkboxes and then operate the regeneration button in the regeneration operation area 45 to regenerate the checked observation documents all at once.

[0059] The checkboxes may also be linked to the batch copy operation field 49. The checkboxes provide a function for users to select the observation texts they wish to copy for transcription into report cards, student records, etc. For example, to copy only the observation texts for students No. 2 and No. 4, the user checks the corresponding checkboxes and then operates the batch copy operation field 49 to copy the selected observation texts. This function allows users to efficiently copy and transcribe only the necessary observation texts.

[0060] Furthermore, checkboxes for individual selection and checkboxes for selecting all items at once may be provided. Operating the checkbox for selecting all items at once will change the status of individual selections simultaneously. For example, after selecting all items by clicking the checkbox for selecting all items, if a user does not wish to regenerate data for a specific evaluation subject, they can adjust this by individually unchecking the corresponding items in the checkboxes for individual selection. This configuration allows users to flexibly and efficiently process both batch and individual evaluation documents.

[0061] The observation texts displayed in the list display area 42 are editable. The editing unit 515 of the information processing device 10 implements the editing function for the generated observation texts. The editing unit 515 provides a function for the user to manually edit the observation texts displayed in the list display area 42. In addition, the editing unit 515 provides a function for the user to directly input prompts in an interactive manner with the LLM server 20 to have the observation texts regenerated in the LLM. The chat operation area 46 is for switching to a mode in which the observation texts generated for each person being evaluated are regenerated in the LLM.

[0062] Figure 5(b) is an example of a screen displayed after operating the Chat operation field 46 corresponding to the evaluation subject's comments. When the Chat operation field 46 is operated, the dialogue screen 60 is displayed, showing the LLM's response and the instructions entered by the user for the LLM. The editorial department 515 first sends the selected evaluation subject's comments to the text generation unit 513, extracts the necessary data from the evaluation input unit 511, the settings unit 512, the example sentence database DB 521, and the prompt DB 522 to generate prompts, and sends them to the LLM server 20. The text generation unit 513 sends the generated comments and comments from the LLM back to the editorial department 515, and the comments are displayed on the dialogue screen 60. The user refers to the comments and comments and enters requests, etc., into the instruction input field 62. The content entered into this instruction input field 62 is sent from the editorial department 515 to the text generation unit 513 and then sent to the LLM server 20. The text generation unit 513 obtains the LLM's response based on the content entered in the instruction input field 62 and sends it to the editorial unit 515, which then displays the response on the dialogue screen 60. This process may be repeated until the user obtains the desired result, or it may be limited to a certain number of attempts.

[0063] If the user obtains the desired observation text in the dialogue screen 60 shown in Figure 5(b), they can click button 61 (the button labeled "Apply") to reflect the new observation text in the corresponding cell of the list display area 42 in Figure 5(a). The user can also discard the observation text obtained in the dialogue screen 60 by operating button 63 (the button labeled "Close"). If button 63 is operated, the observation text in the list display area 42 will remain in the state it was in before the operation of the Chat operation area 46.

[0064] As described above, the output and editing screen configuration shown in Figure 5 allows users to efficiently review, select, and edit the generated observation documents. Furthermore, the real-time document generation and adjustment function utilizing LLM enables highly flexible document creation.

[0065] 2.3. Regarding the input screen for generating observations about "Foreign Language Activities" Figure 6 shows an example of the input screen when generating observations for "Foreign Language Activities" using the text generation system 100 according to the present invention. Figure 6(a) shows the state in which the "Foreign Language Activities" mode is selected in the mode selection field 40, and Figure 6(b) shows the state in which the evaluation data input area 67 is expanded.

[0066] As shown in Figure 6(a), the screen has a mode selection field 40 on the left side, and by selecting "Foreign Language Activities," the user can input evaluation items corresponding to this mode. At the top of the screen, there is an area 64 for specifying generation conditions, where the type of evaluation document (for report cards or student records), grade level, number of students, character limit for the observation text, and common phrase settings can be entered. These input fields are the same as the input field 31 for the purpose of use of the observation text, the grade level selection field 32, the number of students to be evaluated setting field 33, the character limit setting field 34, and the common phrase setting field 35a shown in Figure 4. For example, when generating an observation text related to "Foreign Language Activities," a specific example sentence such as "Through foreign languages, students deepened their understanding of language and culture" can be entered in the common phrase setting field 35a.

[0067] The information set in the generation condition specification area 64 is sent to the document generation unit 513 via the evaluation input unit 511 and the setting unit 512, and is input to the LLM server 20 as a prompt. Through this process, an appropriate observation document is generated according to the evaluation results regarding the foreign language activity.

[0068] In Figure 6(a), the evaluation data input area 67 is displayed in a simplified form, allowing users to check basic items related to each child. On the other hand, as shown in Figure 6(b), when expanded, it displays a list of detailed evaluation items for each child. The evaluation data input area 67 includes a number field 36, an evaluation item field 65, and a special notes input field 66.

[0069] The number field 36 is used to identify the person being evaluated, and for example, the student number or identification number is displayed. As shown in Figure 6(b), the evaluation item field 65 has evaluation items such as "listening," "speaking (interaction)," and "speaking (presentation)" for each of the following categories: "knowledge and skills," "knowledge and judgment (thinking, judgment, and expression)," and "active learning attitude (attitude towards learning)," and each can be evaluated using a scale such as a circle (〇) or checkmark. In the special notes input field 66, teachers can enter specific comments in a free-form text format. This allows for detailed recording of the characteristics and areas for improvement in the children's foreign language activities.

[0070] As shown in Figure 6(b), in the expanded display state, all evaluation items for each child are visible, making it possible to efficiently input evaluations for multiple children. In addition, the information entered in each input field is sent to the text generation unit 513 via the setting unit 512 and the evaluation input unit 511, and is used to generate the observation text.

[0071] This input screen configuration allows teachers to easily and accurately input evaluation results related to "Foreign Language Activities." Furthermore, based on the information entered on this screen, it is possible to quickly generate observation documents using LLM, thereby streamlining the observation document creation process.

[0072] The user inputs information using the input screen shown in Figure 6. Based on this input, the text generation unit 513 generates prompts to be entered into the LLM server 20 and obtains the observation text. The display and editing screens for the generated observation text are the same as those shown in Figure 5, so their explanation is omitted.

[0073] 2.4. Regarding the input screen for generating opinion texts on "morality" Figure 7 shows an example of the input screen when generating "moral" observations in the text generation system 100 according to Embodiment 1. Figure 7(a) shows the state in which the "moral" mode is selected in the mode selection field 40.

[0074] As shown in Figure 7(a), a mode selection field 40 is provided on the left side of the screen, and the user can switch to the mode for generating observation texts based on moral education lessons by selecting "Moral Education". In the generation condition specification area 72 at the top of the screen, it is possible to set the grade level, number of students, character count range, common phrases, etc. These input fields are the same as the input field 31 for the purpose of use of the observation text, the grade level selection field 32, the number of students to be evaluated setting field 33, the character count range setting field 34, and the common phrase setting field 35a shown in Figure 4. These settings are used as basic data for generating observation texts, similar to the input screens for other modes shown in Figures 4 and 6.

[0075] The generation condition specification area 72 has a field for entering common text, where you can set an example sentence such as, "Through moral education lessons, students developed fair judgment and cooperation." This common text will be used as content that is inserted for all students.

[0076] The evaluation data input area 71 includes a number field 36, an evaluation item input field 68, a content item selection field 69, and a special notes input field 70. The number field 36 contains the identification information of the person being evaluated, and the evaluation item input field 68 lists evaluation items such as "multi-faceted and multi-perspective" and "relationship with oneself." These items are entered using simple symbols such as circles (〇).

[0077] In the content item selection field 69, users can select content related to the evaluation items. For example, items such as "friendship, trust," "pursuit of truth," and "diligence, public duty" can be selected from a drop-down list. This selection information will be reflected in the individual child's commentary during the commentary generation process described later.

[0078] The special notes input field 70 is a field for teachers to freely describe each student's behavior during moral education lessons. For example, specific comments such as "The student actively engaged in the pursuit of truth and solved problems in cooperation with others" can be entered. This information is used in combination with the content selected in the content item selection field 69 to generate an appropriate commentary.

[0079] The input screen in Figure 7 is linked to the evaluation input unit 511, the setting unit 512, and the text generation unit 513, and the entered information is immediately reflected in the text generation process. This configuration allows teachers to efficiently and accurately generate observational texts based on moral education lessons. The generated observational texts can be viewed and modified on a screen similar to the output and editing screen shown in Figure 5.

[0080] 2.4. Regarding the input screen for generating observations about "Integrated Studies" Figure 8 shows an example of the input screen when generating observational text related to "Comprehensive Learning Time" in the text generation system 100 according to Embodiment 1. As shown in Figure 8, a mode selection field 40 is located on the left side, and the user can use the input screen corresponding to this mode by selecting "Comprehensive Learning Time".

[0081] In the generation conditions specification area 73 at the top of the screen, it is possible to set the type of evaluation document (for report cards or student records), grade level, number of students, character limit for the comments, and common text. These input fields are the same as the input field 31 for the purpose of use of the comments, the grade level selection field 32, the number of students to be evaluated setting field 33, the character limit setting field 34, and the common text setting field 35a shown in Figure 4. For example, a common text such as "Regarding integrated learning time..." can be entered, and based on this, common content can be inserted into the comments for all students.

[0082] Below the area 73 for specifying generation conditions, there is an area where fields 74 for setting learning activities and evaluation criteria are displayed. These fields 74 allow teachers to enter information about the content of learning activities and evaluation criteria. The information entered in the fields 74 is sent to the text generation unit 513 via the setting unit 512, and the text generation unit 513 generates a prompt based on that information. The prompt generated by the text generation unit 513 is configured to generate an observational text by comparing the graded evaluations entered in the evaluation data input area 77 for "Knowledge and Skills," "Thinking, Judgment, and Expression," and "Proactive Learning Attitude" with the information in the fields 74. Alternatively, as an example, if the evaluation entered in the evaluation data input area 77 shows that "Knowledge and Skills" is rated A, "Thinking, Judgment, and Expression" is rated C, and "Proactive Learning Attitude" is rated C, the text generation unit 513 may generate a prompt that instructs the LLM to generate an observational text based on the content entered in the "Knowledge and Skills" section of the evaluation criteria in the input field 74.

[0083] In addition, the number of characters is displayed next to the evaluation criteria column (in Figure 8, 12 characters are displayed next to the learning activity column, 27 characters next to the knowledge / skill column, etc.), allowing users to check and adjust the number of characters in the sentences they enter in the evaluation criteria column. The LLM server 20 generates the observation document based on these evaluation criteria, so users should check the number of characters in the sentences they enter in the evaluation criteria column to ensure that the observation document is generated within the character limit.

[0084] The evaluation data input area 77 includes a number field 36, an evaluation input field 75, and a special notes input field 76. The identification number of the person being evaluated is entered in the number field 36, and the evaluation result of the child is entered in the evaluation input field 75. Evaluations can be entered simply using symbols such as "A," "B," and "C," which helps to reduce the burden on teachers.

[0085] The special notes input field 76 allows for free-form descriptions of each child's activities. For example, specific activity details such as "Interviewed a Bingo Kasuri craftsman and deepened their understanding of local culture" can be recorded. This description is reflected in the generated observation text, enabling the creation of personalized reports.

[0086] The screen configuration shown in Figure 8 is closely linked to the evaluation input unit 511, the setting unit 512, and the text generation unit 513, reflecting the input data in real time and supporting the process of efficiently generating observation texts. The generated text is displayed on an output and editing screen similar to that shown in Figure 5, and can be further adjusted.

[0087] 2.5. About the screen for converting and editing observation documents Figure 9 shows an example of the input / output screen during the conversion and editing of observation texts in the text generation system 100 according to Embodiment 1. This screen is provided as an interface for modifying or converting already created observation texts. A mode selection field 40 is provided on the left side of the screen, and this screen can be accessed by selecting the "Conversion / Editing" mode.

[0088] In the input area 78 in the center of the screen, users can enter observation texts that they wish to modify or convert. For example, when entering observation texts for multiple evaluators, each observation text can be entered corresponding to a number in the number field 36.

[0089] A "Convert and Edit" button 79 is located at the bottom of the screen. When the user clicks this button, the input commentary text is converted and edited. At the top of the screen, there is a setting field 78a for the conversion mode of the commentary text entered in the input area 78, and a setting field 33 for the number of people to be evaluated. Below that, the processing content to be performed by the text generation system 100 is briefly described, with specific explanations such as "Converts text for report cards (polite style) to text for student records (plain style)" and "Checks for typos and corrects / adds errors."

[0090] The conversion unit 516 generates a prompt based on the observation text and processing content entered in the input area 78, and retrieves the observation text generated by the LLM server 20 based on that prompt. Once the conversion and editing are complete, the converted observation text is reflected in the input area 78. If the conversion result is not appropriate, it is also possible to adjust the conversion conditions or manually edit the text by clicking the "Convert and Edit" button again.

[0091] This screen works in conjunction with the conversion unit 516 and supports the process of appropriately converting the style and expression based on the input text. This makes it possible to efficiently generate commentary texts suitable for use in report cards and student records. In addition, the function to correct typographical errors contributes to reducing the burden on teachers. These functions are reflected in the prompts and sent to the LLM server 20 upon user instruction.

[0092] As shown in Figure 9, this screen allows users to quickly and accurately correct entered text. Furthermore, they can make further adjustments while viewing the correction results in real time. This system configuration improves both the quality of the text and the efficiency of the work.

[0093] Additionally, a paste operation field 78b is provided in the upper right corner of the input area 78. The paste operation field 78b is a button for pasting data copied from a list of observation documents created by the user in an Excel file or similar to the PC's clipboard into the input area 78.

[0094] 2.6. About the settings screen Figure 10 shows an example of the settings screen for the text generation system 100 according to Embodiment 1. In the settings screen shown in Figure 10, the setting items such as the number of children and the character count range also correspond to the area for specifying generation conditions in the screens shown in Figures 4, 6, 7, and 8. By setting these items all at once in the settings screen of Figure 10, these settings are automatically reflected in each screen. The character count range setting item is configured to be set for each mode ("General Comments," "Moral Education," etc.) used to generate the commentary text.

[0095] With this configuration, when a teacher enters settings common to the entire class (such as the number of students and character limit) in the settings screen shown in Figure 10, these settings are automatically reflected in the respective screens shown in Figures 4 through 8. This eliminates the need for teachers to repeatedly set the same content on each screen, preventing setting errors and increased workload. Furthermore, if settings are changed, the changes made in the settings screen shown in Figure 10 are reflected in each screen in real time, making it easy for teachers to unify the settings across the entire system.

[0096] The settings screen shown in Figure 10 includes an NG word setting field 81. NG words entered here serve to control the generation of the commentary text, preventing their inclusion. This setting prevents inappropriate words and expressions from being included in the commentary texts generated by teachers, thereby improving the quality of the writing.

[0097] For example, if words such as "trouble," "bullying," and "violence" are entered in the NG word setting field 81, the system will control the generation of observation texts on each screen (Figure 4 ("Overall Observation"), Figure 6 ("Foreign Language Activities"), Figure 7 ("Moral Education"), and Figure 8 ("Integrated Learning Time") to prevent these words from being included in the generated results. Specifically, the NG word setting information is sent from the setting unit 512 to the text generation unit 513 and is taken into consideration when generating prompts.

[0098] Furthermore, the system is designed to allow users to easily register prohibited words by simply entering multiple words separated by commas. This simple input method enables teachers to quickly set up the exclusion of specific expressions.

[0099] Furthermore, since the NG word settings are applied all at once, the settings configured in Figure 10 are applied consistently to all screens from Figures 4 to 8. This batch application eliminates the need to individually configure NG words in multiple modes and ensures consistency in settings.

[0100] The settings in Figure 10 are sent to the text generation unit 513 via the setting unit 512. At this time, the batch settings such as the character count range are taken into consideration when prompt generation, so that the prompts sent to the LLM server 20 are controlled to maintain consistency with the individual settings of each screen. This process ensures that the generated observation text is always based on the latest settings, improving accuracy.

[0101] 3. Information processing of the text generation system 100 3.1. Information Processing Flow of Text Generation System 100 Figure 11 shows an example of the information processing flow in the document generation system 100 of the present invention. This flow illustrates a mechanism in which the user terminal 30, the information processing device 10, and the LLM server 20 work together to efficiently execute the processes of generating, editing, and regenerating the observation document. The steps in Figure 11 will be described in detail below.

[0102] First, the user inputs evaluation result data about the person being evaluated using the user terminal 30 (step S1). This evaluation data includes graded evaluations for each subject, graded evaluations for predetermined evaluation items such as learning status and behavioral evaluation, and evaluation comments by the user, and is transmitted to the information processing device 10. The evaluation result data is used as basic data for generating an observation document. The input evaluation result data may be stored in the storage unit (local storage, etc.) of the user terminal 30, or in the storage unit 52 of the information processing terminal 10.

[0103] Next, the information processing device 10 generates a prompt based on the input configuration information and evaluation data (step S2). The prompt is an instruction to the LLM that reflects the content of the evaluation data input to the information processing device 10, and is important information for the LLM server 20 to generate an appropriate observation document. The configuration information reflected in the prompt includes the observation document generation mode specified in the mode selection field 40, as well as information set in the generation condition specification areas in Figures 4, 6, 7, and 8, such as the observation document usage purpose input field 31 and the grade selection field 32, and the content entered in the configuration input area 80 shown in Figure 10 and the NG word setting field 81 shown in Figure 10.

[0104] Next, the information processing device 10 extracts relevant example sentences from the example sentence database (example sentence DB) 521 based on the setting information such as the observation document generation mode specified in the mode selection field 40, the purpose of use input field 31 for the observation document, and the grade selection field 32, and reflects them in the prompt (step S3).

[0105] Next, the information processing device 10 sends the generated prompt to the LLM server 20 (step S4). The LLM server 20 generates an observation document using a large-scale language model based on the received prompt (step S5), and sends the generated document back to the information processing device 10 (step S6).

[0106] Next, the information processing device 10 transmits the returned findings document to the user terminal 30 and displays it on the screen (step S7). The displayed document can be used as is by the user, or edited as needed (step S8). Editing can be done manually, or regeneration using LLM can be used. The output findings document data may be stored in the storage unit of the user terminal 30 (such as local storage), or in the storage unit 52 of the information processing terminal 10.

[0107] In the editing operation, if the user wishes to regenerate the observation document using LLM, the user operates the regeneration operation field 45 shown in Figure 5(a) (indicated as "Yes" in step S9), and a prompt is sent to LLM (step S10). The LLM server 20 generates a new observation document (step S11) and sends it to the information processing device 10 (step S12). On the other hand, if the user does not instruct regeneration (indicated as "No" in step S9), the document manually edited by the user is used as is.

[0108] Finally, the user reviews the outputted findings document on the user terminal 30 (step S13), and uses it by copying it as needed, exporting it as a spreadsheet, etc. This flow allows the user to create findings documents quickly and accurately, and to flexibly adjust them by manual editing or regeneration as needed.

[0109] 3.2. Prompts generated by the information processing device 10 Figure 12 shows an example of a prompt sent to the LLM server 20 in the text generation system 100 according to Embodiment 1. The configuration of the prompt and how the prompt is used in generating the observation text will be explained below.

[0110] The prompt consists of two main parts: system content and user content. These indicate the data structure sent to the LLM server 20 and contain the information necessary for generating the report text.

[0111] The system content is the section that describes the basic instructions and constraints for the LLM. In this example, it includes the following information: (1) Explanation that evaluation data will be provided in JSON format. (2) The tone (style) of the document to be written (for example, writing in plain language) (3) Requirements to be included in the generated text (e.g., "Everyone did a great job at the music recital.") (4) If the content includes prohibited words (e.g., "trouble," "bullying," "violence"), it is necessary to rephrase them appropriately. (5) Specific examples of output formats (e.g., create a child's individual assessment report of approximately 200 characters) (6) Setting the LLM role (Example: You are an elementary school teacher) This clarifies the rules that LLM should follow when generating text. The specific details of (1) to (5) above will be reflected in the content entered in the generation condition specification area shown in Figure 4, etc. For example, (2) above will reflect the content of the purpose of use input field 31 shown in Figure 4(a). (3) above will reflect the content of the common text setting field 35a shown in Figure 4(a). (4) above will reflect the content of the NG word setting field 81 shown in Figure 10. (5) above will reflect the content of the character count range setting field 34 shown in Figure 4(a).

[0112] The user content section describes the input data and specific information to be provided to the LLM. For example, it is written in JSON format and includes specific evaluation data. This is created based on the data input to the information processing device 10 in steps S2 and S3 shown in Figure 11. The main structure is described below. (7) Hierarchical structure of evaluation data The top-level key (e.g., 1) indicates the identification information of the person being evaluated (student). In this case, "1" corresponds to "1" in the number column 36 shown in Figure 4(a), etc. The contents enclosed in "{}" after "1" are listed hierarchically for each evaluation item. In the example in Figure 12, the contents enclosed in "{}" after "1" are classified into each evaluation item (e.g., "Daily Life," "Subjects," "Learning"). Values ​​corresponding to each evaluation item are obtained from the evaluation data and reflected. For example, the evaluation item "Daily Life" includes the items "Special Notes" and "Unit." The value "The student has learned about 'From Jomon Villages to Kofun Lands' and has deepened their interest in history." is entered for the item "Special Notes." The item "Unit" contains the items "Subject," "Unit Name," and "Evaluation," with corresponding values ​​described for each (for "Subject," it is "Social Studies," for "Unit Name," it is "From Jomon Villages to Kofun Lands," and for "Evaluation," it is "Students have a good understanding of the social conditions and changes from the Jomon to Kofun periods, and have acquired the skills to research and express them."). In this way, the prompts are described hierarchically, corresponding to each evaluation item included in the evaluation data entered by the user into the text generation system 100. (8) Content of evaluation items The "Daily Life" section includes special notes and the students' activities based on the learning unit. The special notes section describes the learning content "From Jomon Villages to Kofun Lands," and the evaluation of "Social Studies" is shown as the result of this learning. The "Subject" section includes grades for science, Japanese language, and mathematics (e.g., "Evaluation: B," "Evaluation: C"), as well as specific comments. The "Learning Activities" section includes special notes on the results of studying the unit "The Shape of the Moon and the Sun," as well as the fact that the children developed an interest in natural phenomena. (9) Basic information on specific examples The contents of "Special Notes," "Unit Name," and "Evaluation" (the content written after the value ":" corresponding to "Special Notes," etc., in the JSON format description) are important basic data for generating observation documents in the LLM server 20, and LLM creates appropriate documents based on this data.

[0113] As explained in (7) to (9) above, user content is written in JSON format, and its structure consists of a 'key' and a 'value' as follows: The 'key' represents the name of the evaluation data item, such as "special notes" or "unit name". On the other hand, the 'value' indicates the specific content corresponding to the 'key', such as "The person is punctual and has good behavior" or "From Jomon village to Kofun land". By utilizing this relationship between 'key' and 'value', the LLM (Large-Scale Language Model) understands the information for each person being evaluated and generates an observation document.

[0114] A key feature of this prompt format is that the evaluation data for each evaluator is organized in JSON format, and the evaluation items are structured. This structuring allows the LLM to efficiently process each piece of data and generate natural-sounding sentences. For example, if the results of learning "From Jomon Villages to Kofun Lands" are listed as a special note, that content will be incorporated into the prompt, generating a sentence such as "Interest in history deepened."

[0115] Even if the evaluation data contains prohibited words, appropriate paraphrasing will be performed according to the instructions in the prompt. This mechanism ensures that the generated observation text is expressed appropriately.

[0116] The prompts shown in Figure 12 illustrate a specific example of how the text generation system 100 utilizes teacher input data and communicates it to the LLM server 20. By using this prompt format, natural and appropriate observation texts based on evaluation results can be efficiently generated.

[0117] The prompt shown in Figure 12 is generated based on the information entered on the screens in Figures 4 and 5. Specifically, the teacher enters evaluation data on the input screen shown in Figure 4 and specifies setting conditions based on that data, which generates a prompt like the one in Figure 12 in the information processing device 10. This prompt is then sent to the LLM server 20 to generate the observation document that can be viewed on the output / editing screen shown in Figure 5.

[0118] In the generation condition specification area shown in Figure 4, the grade level, number of students, character limit, common text, etc., are set. This information is reflected in the system content section of the prompt. The evaluation data input area shown in Figure 4 contains specific evaluation data regarding children's behavior and learning outcomes. This evaluation data is structured and described as the user content portion of the prompt. For example, if a sentence such as "Let's work together as a class at the sports day" is entered in the "Common Sentence Setting Field 35a" shown in Figure 4, it will be added to the instructions in the system content. Furthermore, when evaluation results or special notes for each subject are entered in the "Evaluation Data Input Area 41," they are incorporated into the prompt as the corresponding key (e.g., "Subject") in the user content.

[0119] Furthermore, the output and editing screen shown in Figure 5 displays the observation text generated based on Figure 12. As an example, the observation text generated based on Figure 12 is output as follows: "In social studies, students are learning about the transition from Jomon villages to Kofun burial mounds, deepening their interest in and understanding of history. Their interest in cultural and social changes is growing, and their inquisitive spirit is being nurtured. In Japanese language kanji learning, they are paying attention to strokes, hooks, and flourishes, making an effort to write characters accurately. In mathematics, they are demonstrating data organization skills and improving their understanding using charts and graphs. In the unit on "The Shape of the Moon and the Sun," their interest in natural phenomena has increased, and their understanding of the moon's changes has deepened. Everyone did a great job at the music recital."

[0120] Furthermore, the prompt in Figure 12 operates based on the NG words (e.g., "trouble," "bullying," "violence") set in the settings screen shown in Figure 10. If an NG word is included in the evaluation data, the LLM server 20 will perform an appropriate rephrasing according to the instructions specified in the prompt's system content. This prevents the generation of expressions that the teacher did not intend.

[0121] Figure 13 shows an example of a prompt sent to the LLM server 20 in the text generation system 100 according to Embodiment 1. The prompt shown in Figure 13 is an example of a prompt generated by the text generation system 100 based on simplified evaluation data of the person being evaluated. This prompt is, for example, the result when, in the input screen for generating the "Overall Observation" observation text shown in Figure 4, the evaluation fields (e.g., A, B, C / C / ○×) for the "Lifestyle" and "Learning" sections (Lifestyle entry field 38 and evaluation entry field 39 for each subject) are not filled in, and only the comment entry fields 38a and 39a are filled in when the observation text generation is performed. The text generation system 100 can generate an observation text based on the evaluation result data entered by the user, even if not all of the evaluation result data is entered. If the evaluation fields are not filled in and the observation text generation is performed with only comments, the text generation unit 513 can extract only the entered data and generate a simplified prompt as shown in Figure 13.

[0122] In the prompt in Figure 13, the user content includes, as an example, evaluation items for "daily life" and "learning," with corresponding special notes for each item. Under "daily life," it says, "The person is punctual and manages their daily life properly," and under "learning," it says, "The person has a solid understanding of basic learning content." The descriptions "The person is punctual and manages their daily life properly" and "The person has a solid understanding of basic learning content" reflect the content entered by the user, the teacher, in the evaluation data input area 41 shown in Figure 4.

[0123] The prompt shown in Figure 13 contains less information than the one in Figure 12, but the text generation system 100 generates an observational text according to the character count setting. Such prompts are useful when the evaluation result data entered in the evaluation input unit 511 is limited. In elementary schools, it is time-consuming for teachers to input a large amount of information for all students, but the text generation system 100 has the advantage that LLM will generate a text simply by inputting things like everyday observations about students in a memo-like manner. Note that the user content shown in Figure 13 is entered into LLM together with the system content shown in Figure 12.

[0124] As an example of an observation text generated based on the prompts shown in Figure 13, a relatively concise sentence can be given: "The student is punctual and maintains a regular daily routine. Their behavior is stable, and they have a good understanding of basic learning content. In particular, they show a conscious effort to interact with others and to value time. I hope they will continue in this manner and aim for further growth. Everyone did a great job at the music recital." Thus, the prompt format in Figure 13 provides a foundation for efficiently generating natural and appropriate observation texts, even when evaluation result data is limited.

[0125] 4. Operation of Embodiment 1 The first embodiment of the document generation system 100 is a document generation system 100 that generates observation texts to be included in evaluation documents of students being evaluated at a school, and comprises an evaluation input unit 511 into which evaluation result data of the students being evaluated is input, a storage unit 52 that stores example sentence data for observation texts, and a document generation unit 513 that inputs the evaluation result data and example sentence data into a large-scale language model (LLM) and causes the large-scale language model to output observation texts of the students being evaluated. This configuration allows users to have the LLM generate and utilize observational texts based on the evaluation results they input.

[0126] Figure 14 shows an example of the format of a document created using the output of the document generation system 100. The output of the document generation system 100 is applied to column X shown in Figure 14. The document shown in Figure 14 is a report card, which is given to each third-grade elementary school student and is also viewed by their parents. Teachers need to create appropriate comments for each student, but generally, teachers are responsible for a large number of students, such as 20 or more, and creating comments while aligning them with the graded evaluation results for each subject was a time-consuming task.

[0127] According to the text generation system 100, users can have the LLM create opinion pieces based on evaluation results and example sentences, and it has the advantage of reducing the burden on users to check the consistency between evaluation results and opinion pieces.

[0128] Furthermore, the text generation system 100 may be configured as follows in addition to the above.

[0129] In the above-described text generation system 100, the evaluation result data includes individual evaluation result data, and the example sentence data includes individual example sentence data corresponding to the individual evaluation result data. The text generation unit 513 extracts the individual evaluation result data and individual example sentence data corresponding to the type of observation text to be generated and inputs them into the large-scale language model. With this configuration, the text generation system 100 can input individual example sentence data corresponding to some of the individual evaluation result data from the input evaluation result data into the LLM, enabling the LLM to efficiently generate observational texts.

[0130] The above-described text generation system 100 includes a setting unit 512 into which setting information corresponding to the observation text to be generated is input. The text generation unit 513 generates prompts that reflect the setting information, evaluation result data, or example sentence collection, and inputs the prompts into a large-scale language model. With this configuration, the text generation system 100 can give instructions to the LLM using a predetermined prompt format, thereby efficiently passing user-inputted data to the LLM and enabling the LLM to understand the data more accurately. In particular, by describing the evaluation results in JSON format, a wide range of evaluation result data and corresponding example sentences can be efficiently passed to the LLM.

[0131] Figure 15 shows an example of a portion of the example sentence collection used in the text generation system 100. The example sentence collection stored in the example sentence collection DB521 is created so that "subject," "evaluation," and "example sentence" correspond to each other, as shown in Figure 15. The table shown in Figure 15 is created for each "grade level," and also corresponds to the "grade level" of the subject of the generated observation text. Therefore, the text generation system 100 extracts example sentences belonging to the "grade level," "subject," and "evaluation" corresponding to the observation text to be generated and generates a prompt. As another example, the example sentence collection may be in a format where each example sentence is listed corresponding to "grade level," "item," and "classification." In this case, the text generation system 100 extracts example sentences belonging to the "grade level," "item," and "classification" corresponding to the observation text to be generated and generates a prompt. The example sentence collection shown in Figure 15 only lists a portion of the example sentences for each "subject" and "evaluation," but the example sentence collection actually used has many more example sentences corresponding to "subject" and "evaluation" accumulated and stored. In Figure 15, the information attached to each example sentence, such as "grade," "subject," and "evaluation," is called "extraction parameters." The text generation unit 513 obtains the extraction parameters from the evaluation result data input to the text generation system 100, extracts example sentences based on the extraction parameters, and inputs the extraction parameters and example sentences into the LLM. The extraction parameters are not limited to the above-mentioned "grade," "subject," "evaluation," "item," and "classification," but can also be other, such as the evaluation items set in the evaluation data input fields 41, 67, 71, and 77 in Figures 4, 6 to 8.

[0132] In the above-described text generation system 100, the configuration information is at least one of the following: attribute information of the person being evaluated, the number of people being evaluated, the number of characters in the generated observation text, and a term that is commonly inserted into the observation text for the person being evaluated. With this configuration, the text generation system 100 can receive specific instructions from the user, convert them into prompts, and pass them to the LLM.

[0133] In the above-described text generation system 100, the subjects of evaluation include multiple subjects, and the evaluation input unit is configured to allow input of the evaluation results and evaluation comments for each of the multiple subjects as a list. With this configuration, the text generation system 100 can accept data on a large number of children or students in the teacher's class as a single list, thereby improving user convenience.

[0134] The above-described text generation system 100 includes a text output unit that displays a list of opinion texts about multiple evaluators output from a large-scale language model. With this configuration, the text generation system 100 displays multiple output observation texts together as a list, making it easy for the user to review, copy the data, and generally more convenient.

[0135] The above-described text generation system 100 includes an editing unit 515 that individually edits the observation texts output from the large-scale language model for each of the multiple evaluators. The editing unit 515 is configured to input the observation texts of a specific evaluator and prompts entered by the user into the large-scale language model. With this configuration, the text generation system 100 can have only a portion of the multiple observation texts output as a list regenerated by the LLM, thus providing high user convenience.

[0136] The above text generation system 100 further includes a conversion unit 516 that performs a batch style transformation on the opinion texts about the subjects of evaluation output from a large-scale language model. With this configuration, the text generation system 100 can have only a portion of the multiple observation texts output as a list regenerated by the LLM, thus providing high user convenience.

[0137] Embodiment 2. 5. Configuration of the text generation system 100 Embodiment 2 describes an example in which the document generation system 100 according to Embodiment 1 is applied to other school documents.

[0138] The basic functions of the document generation system 100 according to Embodiment 2 are the same as the functional block configuration shown in Figure 3, but the evaluation input unit 511 is omitted because the document is not about the person being evaluated.

[0139] Figure 16 shows an example of the input screen when generating "classroom / school communication" documents using the document generation system 100 according to Embodiment 2. This screen is provided as an interface for generating communication documents used within a school. A mode selection field 85 is provided on the left side of the screen, and the user can use the input screen shown in Figure 16 by selecting the "school / school communication" mode. In the generation condition specification area 82 at the top of the screen, generation conditions such as grade (5th grade), date (September), and character count range (200-250 characters) are entered. These conditions function as basic setting information that is reflected in the generated document.

[0140] The information entered in the generation condition specification area 82 is similar to the information entered into the setting unit 512 using the input screen shown in Figures 4, 6 to 8 in Embodiment 1, and is reflected, for example, in the system content of the prompt shown in Figure 12. For example, information such as "Grade: 5th grade" and "Character count range: 200 to 250 characters" is described in the system content as follows.

[0141] { "Grade": "5th grade", "Character count range": "200-250 characters" }

[0142] When generating a "class / school newsletter," the content entered in the generation condition specification area 86 and the guidance text display area 87 shown in Figure 15 is generated as a prompt and input into the LLM. Note that this prompt generation may be configured to refer to the example sentence database (example sentence DB) 521. This allows the use of example sentences such as school newsletters issued in the past, enabling the generation of consistent newsletters that reflect past records and expression methods.

[0143] In the center of the screen, there is an input field 83 for entering the content of the message. This field allows users to input specific information that will form the basis of the message, such as "Teacher XX participated in the 'Sports Day Plan' and did their best" or "Announcement of additional events." This information, like the information entered into the setting unit 512 using the input screens shown in Figures 4, 6-8, is sent to the text generation unit 513 via the setting unit 512 and reflected in the prompt. Alternatively, the example sentence database DB 521 can be used to supplement the entered content by utilizing similar past examples.

[0144] A START button 84 is located at the bottom of the screen. When the user clicks this button, a prompt based on the entered information is generated and sent to the LLM server 20. This process automatically generates the content of the school communication document, and the generated document can be reviewed and edited by the user. The information set in the generation condition specification area 82 is reflected in the system content of the prompt, and by combining appropriate example sentences from the example sentence database as needed, the system efficiently generates documents in a unified format.

[0145] Figure 17 shows an example of the input screen for generating an announcement text in the text generation system 100 according to Embodiment 2. In the generation condition specification area 86 shown in Figure 17, the conditions necessary for the announcement text can be set. This area has input fields for setting the recipient of the announcement text (e.g., parents), title, whether or not to include seasonal greetings, the content of the announcement text, and whether or not to include attendance confirmation. This configuration allows the user to easily input basic information and detailed content of the announcement text. For example, it is possible to input a title such as "Announcement of XXXXX Harvest Festival at XX Elementary School" or specific content such as "We will be making sweets using vegetables harvested from the school garden." The information entered in the generation condition specification area 86 is similar to the information entered into the setting unit 512 using the input screens shown in Figures 4, 6 to 8 in Embodiment 1, and is reflected in the system content of the prompt shown in Figure 12. Furthermore, on this screen, templates and example sentences for the generated announcement text may be obtained from the example sentence collection DB 521 stored in the storage unit 52 and used for prompt generation.

[0146] Figure 18 shows an example of the output screen when generating an announcement text for the text generation system 100 according to Embodiment 2. This screen displays the announcement text generated based on the conditions entered in the generation condition specification area 86 in Figure 17. The announcement text display area 87 specifically describes the generated announcement text, and the announcement text reflects the content entered by the user, such as "Announcement of the Harvest Festival "XXXXX" at XX Elementary School" or "Date and time: Month X, Day X, Location: XXXXXXX," which were entered in the input screen shown in Figure 17. Furthermore, if additional editing of the announcement text is required, the user can give instructions to the LLM through dialogue, and the LLM can adjust the text by regenerating it. The user can enter instructions for the LLM in the instruction input field 88 and regenerate the announcement text in a dialogue format. In addition, this screen may be provided with a button to return to the settings screen, making it easy to reset the input conditions and regenerate. With this configuration, the generated announcement text can be flexibly modified according to the user's needs, supporting efficient announcement text creation.

[0147] The document generation system 100 according to Embodiment 2 is a document generation system 100 for generating documents used in schools, and comprises a setting unit 512 for inputting document generation conditions, and a document generation unit 513 for inputting generation conditions and example sentence data into a large-scale language model and causing the large-scale language model to output documents. The generation conditions include at least one of the type of document to be generated, the purpose of the document, the character count range of the document, and terms that are commonly inserted. With this configuration, the text generation system 100 can use a large-scale language model to create school documents such as class newsletters and notices. Users only need to input some information into the settings section to generate the content of school documents, thus reducing their workload.

[0148] Furthermore, in the above-described text generation system 100, the example sentence data may include example sentences from documents generated in the past, and the text generation unit may extract example sentence data suitable for the text to be generated and input it into the large-scale language model. With this configuration, the document generation system 100 can generate documents by referring to example documents created around the same time in the past, for example, thus efficiently generating suitable documents.

[0149] Furthermore, in the above-described document generation system 100, the example sentence data is classified according to the type of document to be generated, and an appropriate example sentence may be selected based on the generation conditions. With this configuration, the text generation system 100 can efficiently provide the necessary example sentences to the LLM, enabling efficient generation of suitable texts.

[0150] Furthermore, the above-described text generation system 100 may include an editing unit for individually editing the output text. The editing unit can re-input the generation conditions reflecting the edited content into the large-scale language model. With this configuration, the text generation system 100 can have only a portion of the generated text regenerated by the LLM. Furthermore, the user can input requests for specific parts of the text as prompts to the LLM, making it easier to improve the quality of the text.

[0151] Furthermore, the above-described text generation system 100 may also include a conversion unit that performs a batch style conversion on texts output from a large-scale language model. With this configuration, the document generation system 100 can convert the writing style of the generated documents all at once. This is because documents created in schools are for various purposes, such as for students, parents, or school staff, and having a function to convert them all at once improves user convenience. Furthermore, the conversion of writing style also includes conversions such as replacing kanji in the generated documents with hiragana or other characters to match the kanji learned in each grade level.

[0152] While embodiments have been described above, these are presented as examples only and are not intended to limit the scope of the invention. The novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications are permitted. These embodiments and their variations are included within the scope and essence of the invention, as well as within the scope of the claims and their equivalents.

[0153] Furthermore, the text generation system 100 described above may also include combinations of the features shown in the following appendices 1 to 8. These combinations are shown below.

[0154] [Note 1] A text generation system that generates observational text to be included in evaluation documents for students being evaluated in schools, An evaluation input unit into which the evaluation result data of the person being evaluated is entered, A memory unit that stores a collection of example sentences for observational texts, The system includes a text generation unit that inputs the aforementioned evaluation result data and the aforementioned example sentence data into a large-scale language model and causes the large-scale language model to output the observational text of the person being evaluated. Text generation system. [Note 2] The text generation system described in Appendix 1, The aforementioned evaluation result data includes individual evaluation result data, The aforementioned example sentence data includes individual example sentence data corresponding to the individual evaluation result data, The aforementioned text generation unit, The individual evaluation result data and the individual example sentence data corresponding to the type of observation text to be generated are extracted and input into the large-scale language model. Text generation system. [Note 3] The text generation system described in Appendix 1 or 2, It includes a setting section into which setting information corresponding to the generated observation text is input, The aforementioned text generation unit, A prompt is generated that reflects the aforementioned setting information, the aforementioned evaluation result data, or the aforementioned example sentence collection data. The prompt is input to the large-scale language model. Text generation system. [Note 4] The text generation system described in Appendix 3, The aforementioned configuration information is At least one of the following: attribute information of the person being evaluated, the number of people being evaluated, the number of characters in the generated observation document, and a term that is commonly inserted in the observation document for the person being evaluated. Text generation system. [Note 5] A text generation system described in any one of the appendices 1 to 4, The aforementioned individuals subject to evaluation include multiple individuals subject to evaluation, The evaluation input unit is, The system is configured to allow input of the evaluation results and evaluation comments for each of the multiple individuals being evaluated, as a list. Text generation system. [Note 6] The text generation system described in Appendix 5, The system includes a text output unit that displays a list of observational texts for the multiple evaluators output from the large-scale language model. Text generation system. [Note 7] The text generation system described in Appendix 6, The system includes an editing unit that individually edits the observation text output from the large-scale language model for each of the multiple evaluators. The aforementioned editorial department, The system is configured to allow input into the large-scale language model of observation texts output from the aforementioned large-scale language model, specifically the observation texts of a particular evaluator and prompts entered by the user. Text generation system. [Note 8] A text generation system described in any one of the appendices 1 to 7, The system further includes a conversion unit that performs a batch style transformation on the opinion texts about the evaluated individuals output from the aforementioned large-scale language model. Text generation system. [Explanation of symbols]

[0155] 10: Information Processing Device 20: LLM Server 25: Start generation button 30: User terminal 31: Input field for purpose of use 32: Grade Selection Field 33: Setting field for the number of people to be evaluated 34: Range setting field 35a: Common text settings field 35b: Insertion position setting field 36: Number field 37: Section for recording the child's condition 38: Section for describing daily life 38a: Comment section 39: Evaluation field 39a: Comment section 40: Mode selection field 41: Evaluation data input area 42: List display field 42a: Column 42b: Scroll bar 43: Character count display field 44: Copy operation field 45: Regeneration operation field 46:Chat operation field 48: Output operation field 49: Batch copy operation field 51: Control Unit 52: Storage section 53: Communications Department 54: Input section 55: Output section 60: Dialogue screen 61: Button 62: Instruction input field 63: Button 64: Designated area 65: Evaluation items section 66: Special Notes Input Field 67: Evaluation Input Area 68: Input field for evaluation items 69: Content item selection field 70: Special Notes Input Field 71: Evaluation data input area 72: Designated area 73: Designated area 74: Entry field 75: Evaluation input field 76: Special Notes Input Field 77: Evaluation data input area 78: Input Area 78a: Setting field 78b: Paste operation field 79: Button 80: Settings input area 81: NG Word Setting Field 82: Designated area 83: Entry field 84: Start generation button 85: Mode Selection Field 86: Designated area 87: Information display area 88: Instruction input field 90: Network 100: Text generation system 511: Evaluation Input Section 512: Settings section 513: Sentence generation section 514: Text output section 515: Editorial Department 516: Conversion section 521: Example Sentences Database 522: Prompt DB

Claims

1. A text generation system that generates observational text to be included in evaluation documents for students being evaluated in schools, An evaluation input unit into which the evaluation result data of the person being evaluated is entered, A memory unit that stores a collection of example sentences for observational texts, The system includes a text generation unit that inputs the aforementioned evaluation result data and the aforementioned example sentence data into a large-scale language model and causes the large-scale language model to output the observational text of the person being evaluated. Text generation system.

2. A text generation system according to claim 1, The aforementioned evaluation result data includes individual evaluation result data, The aforementioned example sentence data includes individual example sentence data corresponding to the individual evaluation result data, The aforementioned text generation unit, The individual evaluation result data and the individual example sentence data corresponding to the type of observation text to be generated are extracted and input into the large-scale language model. Text generation system.

3. A text generation system according to claim 1 or 2, It includes a setting section into which setting information corresponding to the generated observation text is input, The aforementioned text generation unit, A prompt is generated that reflects the aforementioned setting information, the aforementioned evaluation result data, or the aforementioned example sentence collection data. The prompt is input to the large-scale language model. Text generation system.

4. A text generation system according to claim 3, The aforementioned configuration information is At least one of the following: attribute information of the person being evaluated, the number of people being evaluated, the number of characters in the generated observation document, and a term that is commonly inserted in the observation document for the person being evaluated. Text generation system.

5. A text generation system according to claim 1 or 2, The aforementioned individuals subject to evaluation include multiple individuals subject to evaluation, The evaluation input unit is, The system is configured to allow input of the evaluation results and evaluation comments for each of the multiple individuals being evaluated, as a list. Text generation system.

6. A text generation system according to claim 5, The system includes a text output unit that displays a list of observational texts for the multiple evaluators output from the large-scale language model. Text generation system.

7. A text generation system according to claim 6, The system includes an editing unit that individually edits the observation text output from the large-scale language model for each of the multiple evaluators. The aforementioned editorial department, The system is configured to allow input into the large-scale language model of observation texts output from the aforementioned large-scale language model, specifically the observation texts of a particular evaluator and prompts entered by the user. Text generation system.

8. A text generation system according to claim 1 or 2, The system further includes a conversion unit that performs a batch style transformation on the opinion texts about the evaluated individuals output from the aforementioned large-scale language model. Text generation system.

Images