Question creation support device, question creation support method and program
The problem creation support device automates the generation of educational questions using machine learning models, addressing the inefficiencies in manual creation and providing tailored question papers.
Patent Information
- Application Number
- JP2023077937
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-10
- Publication Date
- 2024-11-21
AI Technical Summary
Existing technologies face challenges in efficiently creating problems or questions related to learning content, particularly in educational settings, which are time-consuming and burdensome for educators.
A problem creation support device and method that includes a generation condition acquisition unit and a problem generation unit, utilizing machine learning models like GPT-3 and DALL-E to automatically generate questions based on specified learning elements, difficulty levels, and keywords, generating question papers with corresponding images.
Facilitates easy and efficient creation of questions tailored to learning content, reducing the burden on educators by automating the process and providing customizable question papers.
Smart Images

Figure 2024162437000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to a question creation support device, a question creation support method, and a program. [Background technology]
[0002] With the recent advances in information and communications technology, IT (Information Technology) and DX (Digital Transformation) are being introduced into a variety of industries and business processes.
[0003] For example, information and communication technology is being actively used in the field of education. In schools, lessons using electronic whiteboards and tablets are being introduced. In addition, teaching material creation apps are being used by teachers to create teaching materials, promoting the efficiency of work. For example, Patent Document 1 describes an information processing system for creating question paper images. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] JP 2018-146667 A Summary of the Invention [Problem to be solved by the invention]
[0005] An object of the present disclosure is to provide a question creation support technology for easily creating questions. [Means for solving the problem]
[0006] One aspect of the present disclosure relates to a question creation support device having a creation condition acquisition unit that acquires creation conditions for a question to be generated, and a question generation unit that acquires a question based on the question creation conditions. Effect of the Invention
[0007] According to the present disclosure, a question creation support technology for easily creating questions can be provided. [Brief description of the drawings]
[0008] [Figure 1] FIG. 1 is a schematic diagram showing a question creation support process according to an embodiment of the present disclosure. [Diagram 2] 2A and 2B are schematic diagrams illustrating problem generation conditions and generated problems according to one embodiment of the present disclosure. [Diagram 3] FIG. 3 is a schematic diagram showing a question sheet created by the question creation support process according to an embodiment of the present disclosure. [Figure 4] FIG. 4 is a block diagram illustrating a hardware configuration of a question creation support device according to an embodiment of the present disclosure. [Diagram 5] FIG. 5 is a block diagram illustrating a functional configuration of a question creation support device according to an embodiment of the present disclosure. [Figure 6] FIG. 6 is a diagram showing an input screen for generating conditions according to an embodiment of the present disclosure. [Figure 7] FIG. 7 is a schematic diagram illustrating a question creation support process using a sentence generation model and an image generation model according to an embodiment of the present disclosure. [Figure 8] FIG. 8 is a diagram illustrating training data of a sentence generation model according to an embodiment of the present disclosure. [Figure 9] FIG. 9 is a diagram showing training data of a sentence generation model according to another embodiment of the present disclosure. [Figure 10] FIG. 10 is a diagram illustrating training data for an image generation model according to one embodiment of the present disclosure. [Figure 11] FIG. 11 is a diagram illustrating a correspondence table between learning element IDs and learning element names according to one embodiment of the present disclosure. [Figure 12] FIG. 12 is a schematic diagram illustrating a question creation support process using a sentence generation model and an image generation model according to another embodiment of the present disclosure. [Figure 13] 13A and 13B are diagrams illustrating test paper images according to one embodiment of the present disclosure. [Figure 14] FIG. 14 is a schematic diagram illustrating a question creation support process using a sentence generation model and an image generation model according to another embodiment of the present disclosure. [Figure 15] FIG. 15 is a diagram showing training data of a sentence generation model according to another embodiment of the present disclosure. [Figure 16] FIG. 16 is a flowchart showing a question creation support process according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.
[0010] In the following embodiment, a question creation support device is disclosed.
[0011] [Summary of the Disclosure] In an embodiment described later, as shown in FIG. 1, when question generation conditions such as learning element names, learning element IDs, and keywords are acquired from a user, the question creation support device 100 may generate questions based on the question generation conditions and provide the generated questions to the user in the form of a question paper image or the like. As an example, the question creation support device 100 may generate questions from the acquired generation conditions using a question generation model 10 trained to generate questions from the question generation conditions and provide the generated questions to the user in the form of a question paper image or the like. Alternatively, as another example, the question creation support device 100 may extract questions associated with the question generation conditions from the question database 20 and provide the extracted questions to the user in the form of a question paper image or the like. This makes it possible to automatically generate questions related to the learning content specified by the user, and users such as teachers can efficiently create questions related to the desired learning content using the question creation support device 100.
[0012] Here, the learning element names and learning element IDs were developed primarily by the Japan Association for the Promotion of Educational Information Technology, and are elements that are subdivided at an optimal granularity in order to clearly classify the learning contents in textbooks and teaching materials (for example, see https: / / www.mext.go.jp / a_menu / shotou / zyouhou / detail / __icsFiles / afieldfile / 2018 / 11 / 13 / 1410959_001_1_1.pdf). Each learning element is specified by a learning element name and identified by a learning element ID (consisting of six digits A-Z and 0-9). For example, a learning element name is "area of a triangle (formula: the height is the inside of the figure)," and "T1SUEK" is assigned as the learning element ID for that learning element name.
[0013] For example, as shown in FIG. 2A, when the user provides generation conditions for the problem to be generated, such as "area of a triangle (formula: height is inside the figure)" and "number of problems: 4," the problem creation support device 100 inputs the obtained generation conditions into the problem generation model 10, which will be described in detail below, and generates a problem as shown in FIG. 2B based on the output result from the problem generation model 10, such as "Problem text: Find the area of an equilateral triangle with one side measuring 6 cm. Answer: 15.6 cm." 2 It is possible to generate problem #1 consisting of the text "Problem text: Find the area of a triangle with a base of 10 cm and a height of 6 cm" and a corresponding diagram, problem #2 consisting of the text "Problem text: Find the area of a triangle with a base of 10 cm and a height of 6 cm" and a corresponding diagram, etc.
[0014] When the four questions #1 to #4 are obtained from the question generation model 10 in this manner, the question creation support device 100 may generate a question paper image as shown in Fig. 3 based on the obtained questions #1 to #4. This allows the user to automatically create questions corresponding to the learning content they wish to create using the question generation model 10, and also allows the user to obtain a question paper image for distribution based on the created questions. As a result, it is possible to reduce the burden on users such as teachers involved in the conventional manual task of creating questions.
[0015] The question generation model 10 may be stored in the question creation support device 100, or may be stored in an external server (not shown) connected to the device via a network. A part of the question creation model 10 may be stored in the question creation support device 100, and another part may be stored in an external server.
[0016] Similarly, the question DB 20 may be stored in the question creation support device 100, or may be stored in an external server (not shown) connected to the device via a network. Also, a part of the question DB 20 may be stored in the question creation support device 100, and another part may be stored in an external server.
[0017] Here, question creation support device 100 may be realized by a computing device such as a server, a personal computer (PC), a smartphone, a tablet, etc., and may have a hardware configuration such as that shown in Fig. 4. That is, question creation support device 100 has a drive device 101, a storage device 102, a memory device 103, a processor 104, a user interface (UI) device 105, and a communication device 106, which are interconnected via a bus B.
[0018] The programs or instructions for implementing the various functions and processes of the question creation support device 100 may be stored in a removable storage medium such as a CD-ROM (Compact Disk-Read Only Memory) or a flash memory. When the storage medium is set in the drive device 101, the programs or instructions are installed from the storage medium via the drive device 101 into the storage device 102 or the memory device 103. However, the programs or instructions do not necessarily have to be installed from the storage medium, and may be downloaded from any external device via a network or the like.
[0019] The storage device 102 is realized by a hard disk drive or the like, and stores installed programs or instructions as well as files, data, etc. used in executing the programs or instructions.
[0020] The memory device 103 is realized by a random access memory, a static memory, or the like, and when a program or an instruction is activated, reads and stores the program or instruction, data, or the like from the storage device 102. The storage device 102, the memory device 103, and the removable storage medium may be collectively referred to as a non-transitory storage medium.
[0021] The processor 104 may be realized by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuitry, etc., which may be composed of one or more processor cores, and performs various functions and processing of the question creation support device 100 in accordance with programs, instructions, data such as parameters necessary to execute the programs or instructions, etc. stored in the memory device 103.
[0022] User interface (UI) device 105 may be composed of input devices such as a keyboard, mouse, camera, microphone, etc., output devices such as a display, speaker, headset, printer, etc., and input / output devices such as a touch panel, and realizes an interface between a user and question creation support device 100. For example, a user may operate question creation support device 100 by operating a GUI (Graphical User Interface) displayed on a display or touch panel using a keyboard, mouse, etc.
[0023] The communication device 106 is realized by various communication circuits that execute wired and / or wireless communication processing with an external device, the Internet, a LAN (Local Area Network), a cellular network, or other communication network.
[0024] However, the above-described hardware configuration is merely an example, and the question creation support device 100 according to the present disclosure may be realized by any other appropriate hardware configuration.
[0025] [Question Creation Support Device] Next, a question creation support device 100 according to an embodiment of the present disclosure will be described. Fig. 5 is a block diagram showing a functional configuration of question creation support device 100 according to an embodiment of the present disclosure. As shown in Fig. 5, question creation support device 100 has generation condition acquisition unit 110 and question generation unit 120. Each functional unit of generation condition acquisition unit 110 and question generation unit 120 may be realized by processor 104 executing a computer program stored in memory device 103 of question creation support device 100.
[0026] The generation condition acquisition unit 110 acquires the generation conditions of the question to be generated. For example, the generation conditions may include a learning element name, a learning element ID, or a keyword related to the question. As described above, the learning element name and the learning element ID may be specified in the learning element list organized mainly by the Japan Association for Promotion of Educational Information Technology, or may specify any other learning element. In addition, the keyword may be any description, expression, explanation, etc. that represents the learning content.
[0027] For example, the generation condition acquisition unit 110 may acquire the generation conditions of the problem to be generated from the user via a generation condition input screen as shown in Fig. 6. For example, the learning element name / learning element ID may be selected by a pull-down menu. The generation condition input screen may also have a text box in which keywords can be input.
[0028] The generation conditions may further include the difficulty of the questions. For example, without being limited thereto, three levels of difficulty, low, medium, and high, may be specified as shown in the figure. The generation conditions may further include the number of questions to be generated or the number of questions included on one page of a question paper. As shown in the figure, the number of questions to be generated for each level of difficulty may be specified.
[0029] For example, in the illustrated generation condition input screen, the user specifies the generation conditions as follows: a learning element ID (learning element name) of “T1SUEK (area of a triangle (formula: the height is inside the figure))”, the number of questions to be generated of “low difficulty: 1”, “medium difficulty: 1”, and “high difficulty: 2”, and the number of questions per page of “4”.
[0030] The question generator 120 generates a question based on the question generation conditions. For example, the question generator 120 may input the acquired generation conditions to the question generation model 10 trained to generate a question from the question generation conditions, and acquire a question from the question generation model 10. The question generation model 10 may be realized as any type of machine learning model, such as a language processing model, an image processing model, a video processing model, an audio processing model, and / or a combination thereof. The question generation model 10 may be realized as a trained machine learning model trained by training data consisting of a pair of the question generation conditions and correct answer data (question sentence, answer, figure, etc.), or may be realized as any one or more known machine learning models and / or any combination thereof.
[0031] Regarding an example of a training process of the problem generation model 10, when the problem generation model 10 is realized by a neural network, the generation conditions for training may be input to the problem generation model 10 to be trained, and the parameters of the problem generation model 10 may be adjusted according to an error backpropagation method according to the error between the output result from the problem generation model 10 and the correct answer data (problem text, answer, figure, etc.). When a predetermined termination condition is satisfied, the problem generation model 10 with the adjusted parameters may be used as a trained problem generation model 10. However, the problem generation model 10 according to the present disclosure is not limited to such a neural network model, and may be realized by any other appropriate machine learning model.
[0032] The problem generation model 10 trained in this way may generate a generated problem and a question sheet image as shown in the figure when it receives as input the generation conditions specified by the user as shown in Fig. 6. In the generated problem and question sheet image shown in the figure, four triangle area calculation problems may be generated corresponding to the specified generation conditions, i.e., a learning element ID (learning element name) of "T1SUEK (area of triangle (formula: height is inside the figure))", the number of generated problems of "low difficulty: 1", "medium difficulty: 1" and "high difficulty: 2", and the number of problems per page of "4".
[0033] In the illustrated example, a problem is generated that is composed of a problem statement and a figure, but the problem according to the present disclosure is not limited to this and may be only a problem statement or only a figure. Also, the problem according to the present disclosure may be expressed by data in other formats such as video and audio. For example, the output format of the problem (e.g., text, image, video, audio, etc.) may be specified by the user.
[0034] In one embodiment, the problem generation model 10 includes a sentence generation model 11 trained to generate a problem sentence and an answer from the generation conditions of the problem, and the problem generator 120 may generate a problem sentence and an answer from the generation conditions of the problem to be generated using the sentence generation model 11. The problem generation model 10 may also include an image generation model 12 trained to generate an image related to the problem from the problem sentence of the problem, and the problem generator 120 may generate an image from the problem sentence of the problem to be generated using the image generation model 12.
[0035] For example, as shown in FIG. 7, when a user inputs the learning element name “Area of a triangle (formula: the height is inside the shape)” to the question creation support device 100, the question generation unit 120 adds “Any questions about?” to the end of the learning element name “Area of a triangle (formula: the height is inside the shape)” acquired from the generation condition acquisition unit 110, and inputs the text data “Any questions about the area of a triangle (formula: the height is inside the shape)?” to the sentence generation model 11.
[0036] The sentence generation model 11 receives input text data "What is the problem regarding the area of a triangle (formula: the height is the inside of the figure)?" and generates text data of a question sentence and an answer, "Find the area of a right-angled triangle with a base of 4 cm and a height of 20 cm. [delim] 40 cm 2 The question generator 120 extracts a sentence relating to the figure, “A right-angled triangle with a base of 4 cm and a height of 20 cm,” from the output text data, and inputs the extracted text data to the image generation model 12.
[0037] The image generation model 12 generates an image showing a diagram corresponding to a "right triangle with a base of 4 cm and a height of 20 cm" as shown in the figure. The question generation unit 120 generates a question consisting of a question sentence and an answer generated by the sentence generation model 11 and a figure generated by the image generation model 12 as shown in the figure.
[0038] Here, the sentence generation model 11 may be any known general-purpose language model such as GPT-3 or ChatGPT. Alternatively, the sentence generation model 11 may be any machine learning model trained to generate a question sentence and an answer from a question generation condition. In this case, the sentence generation model 11 may be generated by supervised learning using a training dataset as shown in FIG. 8. As shown in FIG. 8, each training data may be training input data in which a learning element name and a phrase "What is the question about?" are combined, and training output data in which a question sentence and an answer are combined by a connector such as [delim]. The sentence generation model 11 may be generated by inputting each training input data of the training dataset into the sentence generation model 11 to be trained and adjusting the parameters of the sentence generation model 11 according to the error between the output result from the sentence generation model 11 and the training output data. Note that the sentence generation model 11 may be generated from a general-purpose language model by fine-tuning any known general-purpose language model using the training dataset shown in the figure.
[0039] In one embodiment, the sentence generation model 11 may be generated by supervised learning using a training data set as shown in FIG. 9. As shown in FIG. 9, each training data may be training input data containing text data in which a learning element name, a phrase "What about a question?" and the difficulty level of the question are combined, and training output data may be text data in which a question sentence and an answer are combined by a connector such as [delim]. The sentence generation model 11 trained in this way, including the difficulty level, can generate a question sentence and an answer corresponding to the difficulty level. The sentence generation model 11 may be generated from a general-purpose language model by fine-tuning the general-purpose language model using the training data set shown in the figure.
[0040] On the other hand, the image generation model 12 may be any known image generation model that generates an image from text data, such as DALL-E. Alternatively, the image generation model 12 may be any machine learning model trained to generate a figure from a question sentence and an answer. In this case, the image generation model 12 may be generated by supervised learning using a training data set as shown in FIG. 10. As shown in FIG. 10, each training data may be training input data, in which a question sentence and an answer are linked by a connector such as [delim], and image data showing a corresponding figure may be training output data. The image generation model 12 may be generated by inputting each training input data of the training data set into the image generation model 12 to be trained and adjusting the parameters of the image generation model 12 according to the error between the output result from the image generation model 12 and the training output data. Note that the image generation model 12 may be generated from an image generation model by fine-tuning any known image generation model using the training data set shown in the figure.
[0041] In the above-described embodiment, the question generator 120 accepts a learning element name as a generation condition and generates questions based on the learning element name, but the question generator 120 may accept a learning element ID as a generation condition and generate questions based on the learning element ID. In this case, for example, the question generator 120 may have a correspondence table between learning element IDs and learning element names as shown in FIG. 11 and determine a learning element name corresponding to the accepted learning element ID. Then, the question generator 120 may use the determined learning element name as a generation condition and generate questions from the learning element name based on the question generation model 10, the sentence generation model 11, and / or the image generation model 12 as described above.
[0042] Furthermore, the question generator 120 may receive a keyword as a generation condition and generate questions based on the keyword. When the keyword "XXX" is received as a generation condition, the question generator 120 may add "What about questions about ?" to the end of the keyword "XXX", input text data "What about questions about XXX?" to the question generation model 10 and / or the sentence generation model 11, and generate questions as described above. Furthermore, the question generator 120 may estimate a learning element similar to the keyword and generate questions using the estimated learning element as a generation condition. Here, the estimation of the learning element similar to the keyword may be performed using, for example, word2vec.
[0043] In the above-described embodiment, the question generator 120 generates questions by inputting the generation conditions acquired from the user to the question generation model 10, but the question generation process according to the present disclosure is not limited to this. In one embodiment, the question generator 120 may input not only the generation conditions acquired from the user but also the problem example and the generation conditions generated based on the generation conditions to the question generation model 10 to generate questions. It is known that few-shot generation, which provides several input / output examples together with the input, is effective in using general-purpose language models such as GPT-3 and ChatGPT. The question generator 120 may use few-shot generation to generate questions suitable for the generation conditions from the question generation model 10 and / or the sentence generation model 11 realized as a general-purpose language model.
[0044] In generating such few shots, for example, as shown in FIG. 12, when a user inputs the learning element name "Area of a triangle (formula: height is inside the shape)" into question creation support device 100, question generation unit 120 adds "What are the questions about?" to the end of the learning element name "Area of a triangle (formula: height is inside the shape)" acquired from generation condition acquisition unit 110, generating text data saying "What are the questions about the area of a triangle (formula: height is inside the shape)?". Furthermore, question generation unit 120 generates an example question corresponding to the acquired learning element name for use in generating few shots, "What are the questions about the area of a triangle (formula: height is inside the shape)?Find the area of a triangle with a base of 10 cm and a height of 6 cm.[delim]30cm 2 " and "A problem about the area of a triangle (formula: the height is the inside of the figure)? Find the area of a right-angled triangle with a hypotenuse of 8 cm and interior angles of 30° and 60°. [delim]14cm 2" is generated / acquired, and text data consisting of the generated / acquired example problem and "What is the problem regarding the area of a triangle (formula: the height is inside the figure)?" is input into sentence generation model 11 as shown in Figure 12. Thereafter, question generation unit 120 may generate questions in the same manner as the question generation process described with reference to Figure 7. By applying few shot generation in this way and inputting the example problem related to the received generation conditions, along with the generation conditions received from the user, to question generation model 10 or sentence generation model 11, it may be possible to generate questions that are more suited to the generation conditions.
[0045] Here, the problem examples used for generating Few Shots may be generated / obtained by any method, or may be extracted from a training data set such as those shown in FIG. 8 and / or FIG. 9.
[0046] When the question text and the figure are generated from the generation conditions as described above, the question generation unit 120 generates a question paper image based on the question text and the figure. For example, when questions #1 to #4 as shown in FIG. 13A are acquired, the question generation unit 120 may generate a test paper image including questions #1 to #4 as shown in FIG. 13B according to predetermined rules such as the layout and resizing of the test paper image, and provide the test paper image to the user. The rules may be set by default or may be set by the user. The user may save the provided test paper image as it is, or may edit it.
[0047] In the above embodiment, the generation conditions input by the user are text data such as learning element names, learning element IDs, and keywords, but the generation conditions according to the present disclosure are not necessarily limited to such text data. In other embodiments, the generation conditions may include images such as figures, photographs, and drawings. In this case, the question generator 120 may input the images to the question generation model 10 to generate questions.
[0048] For example, as shown in FIG. 14, when a user inputs a graphic image (in the illustrated example, an image showing an equilateral triangle) related to a problem to be generated as a generation condition to the problem creation support device 100, the problem generation unit 120 inputs the graphic image acquired from the generation condition acquisition unit 110 to the sentence generation model 11. Specifically, the problem generation unit 120 may use a convolutional neural network (CNN) or the like to perform size reduction and image feature extraction from the input graphic image, and may input a feature vector obtained by rearranging each pixel of the extracted feature into one dimension as text data to the sentence generation model 11. Here, as shown in the figure, the input graphic image may not include information such as the length of each side and units (e.g., cm). By using a graphic image showing only the shape of a figure without such information, it is possible to diversify the generated question sentences.
[0049] The sentence generation model 11 generates text data of a question sentence and an answer for an inputted graphic image, such as “Find the area of an equilateral triangle whose sides are 6 cm. [delim] 15.6 cm 2 The question generator 120 extracts a sentence relating to the shape, “an equilateral triangle with each side measuring 6 cm,” from the output text data, and inputs the extracted text data to the image generation model 12.
[0050] The image generation model 12 generates an image showing a diagram corresponding to an "equilateral triangle with one side measuring 6 cm" as shown in the figure. The question generation unit 120 can generate a question consisting of a question sentence and an answer generated by the sentence generation model 11 and a corresponding diagram generated by the image generation model 12 as shown in the figure.
[0051] The sentence generation model 11 that generates text from such an image may be any known multi-modality type general-purpose language model such as ChatGPT. Alternatively, the sentence generation model 11 may be any machine learning model trained to generate questions from images of generation conditions. In this case, the sentence generation model 11 may be generated by supervised learning using a training dataset such as that shown in FIG. 15. As shown in FIG. 15, each training data may be training input data of image data, and training output data may be text data in which a question sentence and an answer are linked by a connector such as [delim]. The sentence generation model 11 may be generated by inputting each training input data of the training dataset into the sentence generation model 11 to be trained, and adjusting the parameters of the sentence generation model 11 according to the error between the output result from the sentence generation model 11 and the training output data. Note that the sentence generation model 11 may be generated from a general-purpose language model by fine-tuning any known general-purpose language model using the training dataset shown in the figure. The training dataset shown in the figure may also be used for Few Shot generation. On the other hand, as described above, the image generation model 12 may be any known image generation model that generates images from text data, such as DALL-E.
[0052] In the above-described embodiment, the question generator 120 generates questions (question sentences, answers, diagrams, etc.) based on the generation conditions using the question generation model 10, the sentence generation model 11, and / or the image generation model 12, but the present disclosure is not necessarily limited to this. For example, when one or more questions associated with the generation conditions are stored in the question DB 20, the question generator 120 may acquire the questions associated with the generation conditions from the question DB 20. The questions stored in the question DB 20 may be created and registered by a user or a teaching material creator, or may be created and registered in advance using the question generation model 10, the sentence generation model 11, and / or the image generation model 12.
[0053] [Question creation support process] Next, a question creation support process according to an embodiment of the present disclosure will be described with reference to Fig. 16. Fig. 16 is a flowchart showing the question creation support process according to an embodiment of the present disclosure. The question creation support process may be executed by, for example, the question creation support device 100, and more specifically, may be realized by the processor 104 of the question creation support device 100 executing a computer program or instructions stored in the memory device 103.
[0054] In step S101, question creation support device 100 acquires question creation conditions. Specifically, question creation support device 100 acquires the question creation conditions (e.g., learning element name, learning element ID, keyword, image, number of questions, number of questions per page of question paper, difficulty level, etc.) of the question to be generated from the user.
[0055] In step S102, the question creation support device 100 acquires a question based on the generation condition. Specifically, the question creation support device 100 may add any appropriate wording, such as "What about the question?" to the generation condition and input it to the question generation model 10. Here, the question generation model 10 may be trained using training data consisting of a pair of the question generation condition and the correct answer data (question sentence, answer, figure, etc.), or may be any known general-purpose language model such as GPT-3 or ChatGPT. The question creation support device 100 may create, for example, a question paper image based on the question output from the question generation model 10 and provide it to the user. Alternatively, the question creation support device 100 may extract a question associated with the generation condition from the question DB 20.
[0056] As described above, the question generation model 10 may be composed of the sentence generation model 11 and the image generation model 12. In this case, the question creation support device 100 inputs question generation conditions to the sentence generation model 11, and acquires the question sentence and the answer from the sentence generation model 11. Then, the question creation support device 100 extracts wording related to the image from the question sentence, inputs the extracted wording to the image generation model 12, and acquires an image showing a diagram from the image generation model 12. The question creation support device 100 generates a question paper image from the acquired question sentence and answer and the graphic image, and provides the generated question paper image to the user.
[0057] According to the above-described embodiment, when the creation conditions for the question to be created are obtained from the user, the question creation support device 100 automatically creates a question that corresponds to the creation conditions, thereby reducing the burden of the question creation work.
[0058] Although the examples of the present disclosure have been described in detail above, the present disclosure is not limited to the specific embodiments described above, and various modifications and variations are possible within the scope of the gist of the present disclosure described in the claims. [Explanation of symbols]
[0059] 10 Problem Generation Model 11 Text generation model 12 Image Generation Model 20 Problem Database (DB) 100 Question Creation Support Device 110 Generation condition acquisition part 120 Problem generation part
Claims
1. a generation condition acquisition unit for acquiring generation conditions for a problem to be generated; a question generation unit for acquiring questions based on question generation conditions; A problem creation support device having the above structure.
2. The question creation support device according to claim 1 , wherein the generation conditions include a learning element name, a learning element ID, or a keyword related to the question.
3. 3. The question creation support device according to claim 2, wherein the generation conditions further include the number of questions to be generated or the number of questions included on one page of a question paper.
4. The question creation support device according to claim 2 , wherein the generation conditions further include a difficulty level of the question.
5. The problem creation support device according to claim 1 , wherein the problem includes a problem statement, an answer, and a diagram.
6. The question creation support device according to claim 1 , wherein the question generation unit inputs the acquired generation conditions into a question generation model trained to generate questions from question generation conditions, and acquires questions from the question generation model.
7. The question generation model includes a sentence generation model trained to generate a question sentence for the question from a question generation condition; The question creation support device according to claim 6 , wherein the question generator generates question sentences from generation conditions of the target question by utilizing the sentence generation model.
8. The problem generation model includes an image generation model trained to generate an image related to the problem from a problem sentence of the problem; The problem creation support device according to claim 7 , wherein the problem generator uses the image generation model to generate an image from a problem sentence of the problem to be generated.
9. The question creation support device according to claim 8 , wherein the question generator generates a question paper image based on the question sentence and the image.
10. 7. The problem creation support device according to claim 6, wherein the problem generator inputs the problem example generated based on the generation condition and the generation condition into the problem generation model to generate the problem.
11. The generating conditions include an image, The question creation support device according to claim 6 , wherein the question generator inputs the image into the question generation model to generate the question.
12. The question creation support device according to claim 1 , wherein the question generator acquires the questions associated with the generation conditions from a question database.
13. Obtaining a generation condition for a problem to be generated; Obtaining a problem based on a problem generation condition; A computer-implemented method for assisting in question creation comprising the steps of:
14. Obtaining a generation condition for a problem to be generated; Obtaining a problem based on a problem generation condition; A program that causes a computer to execute the following.
Citation Information
Patent Citations
Program, information processor and information processing method
JP2018146667A
