Minutes creating device, minutes creating method, and recording medium

Abstract

To provide a minutes generation device, a minutes generation method, and a program capable of generating minutes in a question-and-answer format regardless of a question format of a conference.SOLUTION: The processor is configured to execute a process of inputting, to a generation model, processing target data including at least a question and an answer and a prompt including an instruction to cause the generation model to generate minutes data in a question-and-answer format based on the processing target data, and a process of acquiring the minutes data generated by the generation model.SELECTED DRAWING: Figure 11

Description

[Technical Field]

[0001] The present disclosure relates to a minutes generation device, a minutes generation method, and a program for generating minutes data based on data to be processed. [Background technology]

[0002] Generally, minutes are generated by organizing the content of speeches made by each speaker at a meeting or the like and summarizing it as necessary. For this reason, creating minutes of a meeting or the like requires a great deal of effort. To reduce the effort required for creating minutes, techniques for supporting the creation of minutes have been developed. For example, Patent Document 1 discloses a minutes-creation support device that can extract important parts of a meeting to create minutes. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2013-105374 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, in meetings such as those held in parliament, questions and answers are often repeated. Examples of question formats in meetings include a question-and-answer format, in which one question is asked and then answered immediately after it, a general question-and-answer format, in which multiple questions are asked at once and then answered all at once, and a general question-and-answer format, in which multiple questions are asked at once and then answered separately for each question.

[0005] Minutes of meetings using a bulk question and answer format or a bulk question and divided answer format may be generated in which the main points of multiple questions asked at once and the main points of multiple answers given at once or dividedly are written in different positions. Minutes generated in this way can make it very difficult for readers to understand which of the multiple questions asked at once corresponds to which of the multiple answers. On the other hand, minutes written in a question and answer format place questions and their answers close to each other, making them easier for readers to read and understand.

[0006] The present disclosure has been made in consideration of the above circumstances, and aims to provide a minutes generation device, a minutes generation method, and a program that can generate minutes in a question-and-answer format regardless of the question format of the meeting. [Means for solving the problem]

[0007] A minutes generation device according to one aspect of the present disclosure is a minutes generation device including a processor, which executes the following processes: inputting processing target data including at least questions and answers, and a prompt including instructions for causing the generating model to generate minutes data in a question-and-answer format based on the processing target data, into the generating model; and acquiring the minutes data generated by the generating model.

[0008] A minutes generation method according to one aspect of the present disclosure is a minutes generation method executed by a computer, in which the computer executes the following processes: inputting processing target data including at least questions and answers, and a prompt including instructions for causing the generating model to generate minutes data in a question-and-answer format based on the processing target data, into the generating model; and acquiring the minutes data generated by the generating model.

[0009] A program according to one aspect of the present disclosure causes a computer to execute the following processes: inputting processing target data including at least questions and answers, and a prompt including instructions for causing the generating model to generate minutes data in a question-and-answer format based on the processing target data, into the generating model; and acquiring the minutes data generated by the generating model. [Effects of the Invention]

[0010] According to the present disclosure, minutes can be generated in a question-and-answer format regardless of the question format of the meeting. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a diagram illustrating a configuration example of a minutes-of-meetings generation system including a minutes-of-meetings generation device according to an embodiment of the present disclosure. [Figure 2] A block diagram showing an example of the hardware configuration of a minutes generating device. [Figure 3] FIG. 1 is a diagram illustrating an example of a functional configuration of a minutes generating device. [Figure 4] FIG. 10 is a diagram showing an example of data to be processed. [Figure 5] Figure showing the first example prompt [Figure 6] Figure showing a second example of a prompt [Figure 7] Figure showing a second example of a prompt [Figure 8] Figure showing the third example prompt [Figure 9] Figure showing the fourth example prompt [Figure 10] Figure showing the fifth example prompt [Figure 11] Flowchart for explaining an example of operation of the minutes generating device [Figure 12] FIG. 10 is a diagram showing a minutes display screen as a display example 1 of a screen displaying minutes data. [Figure 13] A diagram for explaining the change in the display screen when the user selects the question and answer in the first column in Display Example 1. [Figure 14]A diagram showing a minutes display screen as display example 2 of a screen displaying minutes data. [Figure 15] A diagram showing two different highlighting displays in the second column in display example 2. [Figure 16] A diagram to explain the screen transition of the minutes display screen in Display Example 2 when the Summary Results Summary button is pressed by the user. [Figure 17] A diagram illustrating the screen transitions of the minutes display screen in Display Example 2 when multiple summary selection switches are selected by the user. [Figure 18] A diagram for explaining the screen transition of the minutes display screen of Display Example 2 when the user operates the summary start button and the summarized results are displayed in the summary display field. [Figure 19] A diagram to explain the screen transition of the minutes display screen in Display Example 2 when the Write to Output Document button is operated by the user DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. However, the present disclosure is not limited to the examples described in the embodiments. In the following description, components having the same functions and configurations are designated by the same reference numerals, and their description will be omitted.

[0013] <Summary> The present disclosure relates to a minutes generation device that automatically generates minutes data in a question-and-answer format based on processing target data that records the contents of speeches made by each participant in a meeting or the like.

[0014] In this disclosure, the data to be processed is data including a verbatim transcript of a meeting from which minutes are generated. A verbatim transcript is a verbatim recording of the speech of each participant in the meeting. In this disclosure, the data to be processed includes at least questions posed by participants in the meeting and answers given to those questions by other participants.

[0015] In addition, in the present disclosure, minutes data refers to data that includes summaries of questions included in the data to be processed and summaries of answers corresponding to those questions, in a one-to-one relationship with each other. In other words, the minutes data of the present disclosure is minutes data presented in a question-and-answer format. In the present disclosure, minutes data presented in a question-and-answer format may be referred to as minutes format in question-and-answer format.

[0016] The question summary or answer summary in the minutes data of the present disclosure refers to an extracted summary of the main points of the question or answer. Note that the question or answer contained in the data to be processed and the question summary or answer summary contained in the minutes data do not necessarily need to be different; if the question or answer contained in the data to be processed is originally concise, they may be the same.

[0017] In the present disclosure, the format of questions in a meeting for which the minutes creation device generates minutes may be any of a question-and-answer format, a batch question-and-answer format, a batch question-and-divided answer format, or any other format. In other words, the minutes creation device of the present disclosure can generate minutes data in a question-and-answer format regardless of the format in which questions are asked in the meeting for which minutes are created. Question-and-answer format minutes are easy for readers to read and understand because questions and answers are written in association with each other. Therefore, the minutes creation device of the present disclosure can generate minutes data that are easy for readers to read and understand, regardless of the format of questions in the meeting.

[0018] <System configuration> FIG. 1 is a diagram illustrating an example of the configuration of a minutes-of-meetings generation system including a minutes-of-meetings generation device according to an embodiment of the present disclosure.

[0019] In the example shown in FIG. 1, the minutes generation system 100 includes a minutes generation device 10, a generative model 20, and a network 30.

[0020] The minutes generating device 10 is a device that generates minutes data based on data to be processed. Details of the minutes generating device 10 will be described later.

[0021] The generative model 20 is a deep learning model that includes a huge number of parameters. An example of the generative model 20 is a large language model that is generated by deep learning based on a huge amount of text data and is useful in natural language processing. A large language model can interpret the content of input text and generate an appropriate response depending on the content. An example of a large language model is the Generative Pre-trained Transformer (GPT) developed by OpenAI.

[0022] In the present disclosure, the generative model 20 is not limited to a large-scale language model. Furthermore, the data input to the generative model 20 is not limited to text data. The generative model 20 may be a model that receives data other than text data as input and outputs data other than text data as a response. Examples of data other than text data include image data, audio data, and a format that combines text, image, and audio data.

[0023] The network 30 may be any network that can connect the minutes generation device 10 and the generative model 20. The network 30 is, for example, the Internet.

[0024] 1, the minutes generation system 100 inputs data to be processed and a prepared prompt into the generative model 20, thereby causing the generative model 20 to generate minutes data and acquiring the generated minutes data. This allows the minutes generation device 10 to generate minutes data with relatively little effort.

[0025] <Hardware Configuration of the Minutes Generation Device 10> 2 is a block diagram showing an example of the hardware configuration of the minutes generation device 10. The minutes generation device 10 is a computer including a processor 11, an input receiving unit 12, a display unit 13, a communication unit 14, and a storage unit 15.

[0026] The processor 11 is, for example, a CPU (Central Processing Unit). The processor 11 reads and executes programs stored in the storage unit 15, thereby realizing various functions of the minutes generating device 10.

[0027] The storage unit 15 is configured by, for example, one or a combination of a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), an SSD (Solid State Drive), etc. The storage unit 15 stores programs that are read and executed by the processor 11, as well as various data used in the execution of the programs.

[0028] The input receiving unit 12 is a user interface that receives various operations and inputs of various data by the user of the minutes generating device 10. The input receiving unit 12 is configured with operation devices such as a keyboard, a mouse, a touchpad, and a trackball. Alternatively, the input receiving unit 12 may be an input device that receives data input from outside the minutes generating device 10.

[0029] The display unit 13 displays the minutes generated by the minutes generating device 10. The display unit 13 is a display device such as a liquid crystal display or an organic EL (Electro Luminescence) display.

[0030] The display unit 13 is an example of a display device of the present disclosure. Note that, although FIG. 2 shows an example in which the minutes generation device 10 has the display unit 13 as a display device of the present disclosure, the present disclosure is not limited to this. For example, the minutes generation device of the present disclosure does not need to include a display unit (display device) as a component. In this case, for example, the minutes generation device may transmit data of display content to an external display device and display the desired display content.

[0031] The communication unit 14 communicatively connects the minutes generation device 10 to at least the generative model 20 via the network 30. For example, if the minutes generation device does not have a display device, the communication unit may communicatively connect the minutes generation device to the generative model and the display device.

[0032] <Functional configuration of the minutes generation device 10> Fig. 3 is a diagram illustrating an example of the functional configuration of the minutes generation device 10. In the example shown in Fig. 3, the minutes generation device 10 includes a processing target data acquisition unit 41, a prompt selection unit 42, an input processing unit 43, a minutes data acquisition unit 44, and an output processing unit 45. As described above, these functional configurations are realized by the processor 11 shown in Fig. 2 reading and executing a program stored in the storage unit 15.

[0033] The processing target data acquisition unit 41 executes a process of acquiring processing target data. As described above, the processing target data is verbatim data of a meeting or the like for which minutes are to be generated.

[0034] In the present disclosure, the data format of the data to be processed is, for example, text data, audio data, image data, or a combination of at least any of these. When the data format of the data to be processed is image data, it may be still image data or moving image data. When the data to be processed is text data, the text data may be, for example, audio data recorded during a meeting that has been automatically converted into text using known voice recognition technology.

[0035] As described above, the question format in the meeting for which minutes are to be generated may be any of a question-and-answer format, a batch question-and-batch answer format, a batch question-and-divide answer format, or other formats. The processing target data acquisition unit 41 may automatically determine the format of the questions in the meeting indicated by the acquired processing target data based on the acquired processing target data, past samples, etc.

[0036] The processing target data acquisition unit 41 acquires processing target data input by the user using, for example, the input acceptance unit 12 shown in FIG.

[0037] The prompt selection unit 42 executes a process in which an input processing unit 43, which will be described later, selects a prompt to be input to the generative model 20 together with the data to be processed. A prompt is a command statement for the generative model 20. In the present disclosure, a prompt includes at least an instruction to generate minutes data based on the data to be processed.

[0038] The prompt selection unit 42 selects a prompt selected by the user via the input reception unit 12 from, for example, a plurality of types of prompts prepared in advance. The plurality of types of prompts prepared in advance may be stored, for example, in the storage unit 15 shown in Fig. 2. Each of the plurality of types of prompts may include an instruction to cause the generative model 20 to generate minutes data in a question-and-answer format based on the data to be processed.

[0039] The input processing unit 43 executes a process of inputting the target data for generating minutes and a prompt selected by the prompt selection unit 42, which includes an instruction to cause the generative model 20 to generate minutes data in a question and answer format based on the target data, to the generative model 20. Details of the prompt including an instruction to cause the generative model 20 to generate minutes data in a question and answer format based on the target data will be described later.

[0040] The minutes data acquisition unit 44 executes a process of acquiring the minutes data generated by the generative model 20 from the generative model 20 based on the processing target data and prompt input by the input processing unit 43.

[0041] The output processing unit 45 executes a process of outputting the minutes data acquired by the minutes data acquisition unit 44. The output processing unit 45, for example, displays the minutes data on the display unit 13 (an example of a display device of the present disclosure) shown in FIG. 2. The process executed by the output processing unit 45 is not limited to the process of displaying the minutes data on the display unit 13, but may also be, for example, a process of reading the minutes data aloud from a speaker. In this case, the output processing unit 45 may generate audio data to be output from the speaker using a known reading technology.

[0042] <Example> A specific example in which the minutes generating device 10 generates minutes data based on processing target data will be described below.

[0043] (Specific examples of data to be processed) In the following example, the data to be processed is text data written in JSON format.

[0044] FIG. 4 is a diagram showing an example of data to be processed. In the example shown in FIG. 4, the data to be processed Dobject includes a speaker name "speaker", an utterance number "utt_n", and utterance content "content". The speaker name indicates the name of the person who spoke in the meeting. The speaker name may include not only the name but also the speaker's title (such as member of parliament or chairman). The utterance number is a serial number assigned to all utterances in the meeting. The utterance content indicates the content spoken by the speaker. The utterance number is an example of an identification number in the present disclosure.

[0045] In the processing target data Dobject illustrated in Fig. 4, one utterance content "content" assigned one utterance number "utt_n" is composed of one sentence or multiple sentences. In the example illustrated in Fig. 4, utterance content including one topic by one speaker is treated as one utterance content, and each utterance content is assigned an utterance number.

[0046] In the processing target data Dobject illustrated in Fig. 4, each utterance contains various contents. Some utterances contain questions or answers to questions, while others do not. Utterances that do not contain questions or answers include, for example, introducing oneself, greetings, and providing information.

[0047] In addition, in the processing target data Dobject exemplified in Fig. 4, one speaker continuously speaks about a plurality of speech contents on a plurality of topics. As a result, the processing target data Dobject exemplified in Fig. 4 includes a plurality of questions by one speaker or a plurality of answers by one speaker. In other words, the processing target data Dobject exemplified in Fig. 4 is verbatim data of a meeting held in a batch question and batch answer format.

[0048] (Example of a prompt) Next, a specific example of a prompt input to the generative model 20 together with the processing target data Dobject shown in FIG. 4 will be described.

[0049] (First example) Figure 5 shows a first example of a prompt.

[0050] 5, the first example of a prompt presents the format and data structure of the input data to be processed in the [system] field, and presents the user's request in the [user] field to the generative model 20. The request asks for a question and answer to be compiled in a question-and-answer format.

[0051] With such prompts, it is expected that the generative model 20 will extract questions and answers from the data to be processed, and generate minutes data compiled in a question-and-answer format.

[0052] (Second example) 5, a request is made to the generative model 20 to simply extract questions and answers contained in the data to be processed and compile them in a question-and-answer format. However, in the first example of the prompt, if the data to be processed is input in a batch question-and-answer format or a divided answer format, the questions and answers may not correspond well in the minutes data generated by the generative model 20, especially if there are many batch questions and batch answers on various topics.

[0053] In a second example of the prompt, the generative model 20 is requested to classify the questions and answers contained in the data to be processed by topic, and then to compile the questions and answers for each topic in a question-and-answer format. The second example of the prompt is shown in Figures 6 and 7.

[0054] The second example of the prompt shown in Figures 6 and 7 is similar to the first example in that the format and data structure of the data to be input to the generative model 20 are presented in the [system] column, and the user's request is presented in the [user] column.

[0055] In the [user] column, the second example prompt requests the following: The numbers in parentheses in the following description correspond to the numbered areas shown in Figures 6 and 7. (1) Review the topic list in which the speech content of the data to be processed is pre-classified by topic. (2) Check whether all utterances contained in the data to be processed are included in the topic.

[0056] (3) For each topic, the system extracts the speech content that corresponds to the answer and creates a summary of it. Then, it extracts questions that match the answer summary in content and generates a summary of the questions. (4) Based on the generated answer summaries and question summaries, a question-and-answer format of minutes data is generated.

[0057] In the second example of the prompt, by requesting the generative model 20 to proceed with processing in stages based on the input data to be processed, the generative model 20 can be expected to generate minutes data with high accuracy.

[0058] In particular, requests (1) and (2) allow the data to be processed to be classified by topic. Therefore, subsequent requests (3) and (4) can extract questions and answers by topic, reducing the possibility of incorrectly associating questions and answers with unrelated content.

[0059] Furthermore, request (3) requests the generative model 20 to first extract utterance content corresponding to the answer and generate a summary, and then extract utterance content corresponding to a question whose content matches the content of the summary. For example, as in the first example of the prompt, simply extracting a question and its answer from the data to be processed and summarizing it may result in important elements of the question or answer being omitted when summarizing. However, by first extracting utterance content corresponding to the answer and generating a summary as in request (3), and then extracting utterance content corresponding to a question that matches the content of the summary, that is, that is appropriate for the summary, it is possible to reduce this possibility.

[0060] Note that the example shown in FIG. 6 shows an example in which no topic list is input, as indicated by "(topic list not provided)." However, in practice, in the second example of the prompt, it is desirable that the topic list included in the request (1) be input to the generative model 20 in addition to the data to be processed. The topic list may be one prepared in advance by the user, or may be one generated by the generative model 20 based on the data to be processed using, for example, the third example of the prompt described below. The topic list is an example of a topic list in the present disclosure.

[0061] Examples of topic lists prepared in advance by users include speech announcements and agendas, which are created by meeting participants and organizers before the meeting. A speech announcement is a document in which participants notify the chairperson or meeting organizer in advance of what they will say in order to obtain permission to speak. An agenda is a document that includes a meeting schedule, indicating the topics to be discussed in the meeting and the order in which they will be discussed. These documents are useful as topic lists because they include an outline of what each participant will say.

[0062] In the second example of the prompt shown in FIGS. 6 and 7, the following points are presented to the generative model 20 as precautions. 1. Each question and answer is accompanied by a list of the utterances that served as the basis for it. 2. Emphasize the background of the statement and specific points, and summarize without missing any important points. 3. The summary of the answer should include specific examples and quote exactly what the respondent said. 4. The summary of the question should be a direct quote of the questioner's statement. 5. Ignore "UNKNOWN" topics.

[0063] According to Note 1, the question summaries and answer summaries included in the minutes data generated by the generative model 20 are assigned the utterance numbers in the processing target data from which they were generated. This allows each question summary and answer summary included in the minutes data to be associated with an utterance number (an example of an identification number in the present disclosure) indicating the question or answer in the processing target data from which the summary was generated. This makes it easy to identify which content in the processing target data corresponds to which content in the minutes data. In particular, when the processing target data is in a batch question / batch answer format, the positions of the questions and answers in the minutes data may not match their positions in the processing target data at all. However, according to Note 1, such a situation can be easily handled.

[0064] In addition, precautions 2 to 5 can prevent situations in which the specificity of questions or answers is diluted by summarization, resulting in unclear minutes data.

[0065] (Third example) The third example of the prompt is a prompt for pre-generating a list of topics to be covered in the second example of the prompt. Figure 8 shows the third example of the prompt.

[0066] When generating a topic list in advance, the user first selects a prompt for generating a topic list, as illustrated in FIG. 8, to have the generative model 20 generate the topic list, and then has the generative model 20 generate minutes data in a question-and-answer format using the prompts illustrated in FIGS. 6 and 7.

[0067] (Examples 4 and 5) The fourth and fifth examples of prompts are prompts for causing the generative model 20 to extract words (hereinafter, "important words") that are particularly important for generating minutes data from among the phrases contained in the data to be processed. As described above, minutes data includes question summaries and answer summaries, and important words indicate in detail which parts of each sentence (statement) contained in the data to be processed these summaries refer to. In other words, important words are words that are the source, are closely related, or appear frequently when generating a summary of a specific question or answer. Note that important words are not limited to one word, and may be multiple words. Figure 9 shows a fourth example of a prompt, and Figure 10 shows a fifth example of a prompt.

[0068] Figure 9 is a fourth example of a prompt for extracting key words from phrases included in the data to be processed. Figure 10 is a fifth example of a prompt for extracting key words corresponding to the question summary and answer summary included in the minutes data generated using the first or second example of the prompt. The fourth and fifth example prompts may be sent to the generative model 20, for example, together with the first or second example prompt (the prompt that generates the minutes data based on the data to be processed). This enables the generative model 20 to extract key words from the data to be processed when generating the minutes data based on the data to be processed.

[0069] The fourth and fifth examples of prompts enable accurate extraction of important words from the text contained in the data to be processed that are useful for generating minutes data that include a summary of questions or a summary of answers.

[0070] The first to fifth examples of prompts have been described above. Such various prompts are prepared in advance and stored in the storage unit 15 (see FIG. 2) of the minutes generation device 10. Note that the prompts prepared in advance in the minutes generation device of the present disclosure are not limited to the first to fifth examples described above. The minutes generation device of the present disclosure may prepare in advance any prompts as long as they are prompts for causing a generative model to generate minutes data in a question-and-answer format, prompts for causing a generative model to generate data necessary for generating minutes data in a question-and-answer format, and prompts related thereto.

[0071] (Example of operation) The following describes an example of the operation of the minutes generating device 10. FIG.

[0072] In step S1, the minutes generating device 10 accepts input of processing target data by a user. The input of processing target data by a user may be performed via the input accepting unit 12 shown in FIG.

[0073] In step S2, the minutes generation device 10 accepts a prompt selection by the user. The prompt selection by the user may be performed via the input accepting unit 12 shown in Fig. 2. Here, the minutes generation device 10 may display information about each of the multiple prompts that the user can select (such as the content of each prompt or the properties of each prompt) on the display unit 13, and then accept the user's selection.

[0074] In step S3, the minutes generation device 10 inputs the processing target data received in step S1 and the prompt selected in step S2 to the generative model 20. As a result, the generative model 20 generates minutes data in a question-and-answer format based on the input processing target data and prompt.

[0075] In step S4 , the minutes generation device 10 acquires the generated minutes data from the generative model 20 .

[0076] In step S5, the minutes generation device 10 outputs the acquired minutes data. The output of the minutes data may be realized, for example, by displaying the minutes data on the display unit 13 (an example of a display device) shown in Fig. 2. Alternatively, the output of the minutes data may be realized by reading the minutes data aloud using a speaker or the like.

[0077] By performing such operations, the minutes generation device 10 can cause the generative model 20 to generate minutes data in a question-and-answer format, regardless of the question format of the processing target data, without placing a heavy burden on the user. This makes it possible to generate minutes data that is easy for readers to read and understand with relatively little effort.

[0078] (Examples of display) A specific example of displaying the minutes data in step S5 of the operation example shown in FIG. 11 will be described below.

[0079] (Display example 1) Fig. 12 is a diagram showing a minutes display screen 400 as a display example 1 of a screen displaying minutes data. Note that the dashed lines in Fig. 12 are used to indicate the areas included in the screen, and are not displayed on the actual screen. The same applies to the other figures below used to explain the screen.

[0080] As shown in FIG. 12, the minutes display screen 400 includes a first column 401 in which question summaries and answer summaries included in the minutes data are displayed in association with each other. In the first column 401, the question summaries are annotated with [Question], and the answer summaries are annotated with [Answer]. The question summaries displayed in the first column 401 are, for example, data output by the generative model 20 in response to the command "content":"(summary of the questioner's utterance suitable for summarizing the answerer's utterance)" in the second example of the prompt illustrated in FIG. 7. The answer summaries displayed in the first column 401 are, for example, data output by the generative model 20 in response to the command "content":"(summary of the answerer's utterance)" in the second example of the prompt illustrated in FIG. 7.

[0081] 12, the question summary and answer summary that are related to each other may be displayed surrounded by a frame or the like, allowing the user to understand at a glance which sentences shown in the first column 401 are questions and which sentences are answers.

[0082] 12, the minutes display screen 400 includes, in addition to a first column 401 that displays information about the minutes data, a second column 402 that displays information about the processing target data from which the minutes data was generated. The display in the second column 402 about the processing target data from which the minutes data shown in the first column 401 was generated is based on, for example, a list of utterances that served as the basis for the summary of each question and the summary of each answer included in the minutes data. This allows the user to understand at a glance what processing target data the minutes data displayed in the first column 401 was generated based on.

[0083] In the first column 401 that displays the minutes data, question summaries and answer summaries are displayed in association with each other, and at this time, headings indicating topics corresponding to the question summaries and answer summaries may also be displayed in an associated display manner. In the example shown in Fig. 12, headings related to the topic "Regarding this year's budget" and question summaries and answer summaries corresponding to that topic are displayed in an associated display manner in the first column 401. In the example shown in Fig. 12, headings related to the topic "Regarding proper management of public facilities" and question summaries and answer summaries corresponding to that topic are displayed in an associated display manner in the first column 401. The display of topic headings in the first column 401 is based on data indicating to which topic each question summary and answer summary in the minutes data is classified. The data indicating which topic each question summary and answer summary in the minutes data is classified into is, for example, the data output by the generative model 20 in response to the command "topic":"(title of the Q&A)" in the second example of the prompt illustrated in Figure 7.

[0084] 12, the minutes generation device 10 may display scroll areas 401S and 402S and scroll bars 401B and 402B in the first column 401 and the second column 402, respectively. The scroll bars 401B and 402B indicate the position of the displayed contents of the first column 401 and the second column 402 at that time when the entire scroll areas 401S and 402S correspond to the entire minutes data and the entire data to be processed.

[0085] 12, the minutes display screen 400 further includes a third column 403 that displays a list of the names of speakers of each utterance included in the data to be processed. The display contents of the third column 403 are based on the data of the speaker names included in the data to be processed.

[0086] Here, the minutes generation device 10 may be configured to allow the user to select at least one of the question summary and the answer summary displayed in the first column 401 by an operation via the input receiving unit 12. The operation via the input receiving unit 12 by the user is, for example, an operation in which the user uses a mouse as the input receiving unit 12 to move the pointer to (mouse over) the location that the user wants to select. Figure 13 is a diagram for explaining changes in the display screen when the user selects a question and answer in the first column in Display Example 1.

[0087] As shown in Fig. 13A, the minutes generation device 10 highlights the question summary or answer summary displayed in the first column 401 that the user has selected through an operation via the input receiving unit 12. Fig. 13A shows the question summary selected by the user highlighted in a fluorescent color or the like.

[0088] In response to the user's selection shown in FIG. 13A, the minutes generation device 10 highlights in the second column 402 the question or answer of the processing target data from which the user selected a question summary or an answer summary was generated, as shown in FIG. 13B. It is desirable that the highlighting in the second column 402 and the highlighting in the first column be different. For example, the highlighting in the first column 401 may be yellow highlighting, and the highlighting in the second column 402 may be green highlighting. In FIGS. 13A and 13B, the highlighting is represented by white text on a black background. The highlighting display mode illustrated in FIGS. 13A and 13B is an example of a first mode of the present disclosure.

[0089] The minutes generation device 10 may allow the user to select only one of the associated question summaries and answer summaries in the first column 401, or may allow the user to select both the associated question and answer summaries at the same time. When the data to be processed is in a batch question / batch answer format or a batch question / divided answer format, the questions and answers in the data to be processed may be located far apart. In such cases, it may be difficult to simultaneously display in the second column 402 the questions and answers of the data to be processed that were used to generate the question summaries and answer summaries selected in the first column 401. In such cases, the second column 402 may highlight only either the question or the answer, or the entire data to be processed may be displayed in a reduced size in the second column 402, with the corresponding portions of the question and answer highlighted. Figure 13B shows an example in which only the question displayed in the second column 402 is highlighted.

[0090] The minutes generation device 10 may display, in the scroll area 402S of the second column 402, an indication of the positions in the entire data to be processed of the questions and answers of the data to be processed that were used to generate the question summary and answer summary selected in the first column 401. In the minutes data generated by the second example of the prompt, each of the question summary and answer summary included in the minutes data is associated with an utterance number (an example of an identification number in the present disclosure) that indicates the question or answer of the data to be processed that was used to generate the summary. This allows the minutes generation device 10 to easily identify the positions in the data to be processed of the selected question and answer.

[0091] Figure 13B shows displays 402SE1 and 402SE2 in scroll area 402S of second column 402 that indicate the position of the question and answer of the data to be processed that were used to generate the summary and answer of the question selected in first column 401 when the entire scroll area 402S is compared to the entire data to be processed.

[0092] 13B, the question in the processing target data that was the source for generating the summary of the question selected in first column 401 is highlighted in second column 402. For this reason, in scroll region 402S in FIG. 13B, indicator 402SE1 indicating the position of the question in the processing target data is displayed in a position that overlaps with scroll bar 402B, and indicator 402SE2 indicating the position of the answer is displayed in a position away from scroll bar 402B. This allows the user to easily grasp the position of the question and answer in the processing target data, even if the positions of the question and answer in the processing target data that were the source for generating the summary of the question and the summary of the answer selected in first column 401 are far apart within the processing target data.

[0093] In this way, the questions and answers in the processing target data that were selected in the first column 401 and were used to generate the question summaries and answer summaries in the minutes data are highlighted in the second column 402, allowing the user to easily understand which question or answer in the processing target data was used to generate which question summary or answer summary in the minutes data. This allows the user to evaluate the minutes data generated by the generative model 20 with little effort.

[0094] (Display example 2) 14 is a diagram showing a minutes display screen 500 as Display Example 2 of a screen displaying minutes data. The minutes display screen 500 differs in layout and some of the displayed objects from the minutes display screen 400 of Display Example 1 (see FIGS. 12, 13A, and 13B).

[0095] The minutes display screen 500 illustrated in Fig. 14 has a first column 501, a second column 502, and a third column 503. The first column 501 is a column for displaying information related to the minutes data and is an example of the first column of the present disclosure. The second column 502 is a column for displaying information related to the processing target data from which the minutes data was generated and is an example of the second column of the present disclosure. The third column 503 is a column for displaying a list of the names of speakers of each utterance included in the processing target data.

[0096] In display example 2, the user can select either the question summary or the answer summary from the question summary and answer summary displayed in first column 501. In the example shown in Fig. 14, the fact that the user selected the answer by "Answerer: BBB" is visible to the user because the border of the answer summary display column is displayed thicker than the others.

[0097] In addition, in display example 2, words (hereinafter referred to as important words) that are important to the summary of the question or the summary of the answer selected by the user in first column 501 are highlighted in the question sentence or the answer sentence in the data to be processed displayed in second column 502. In Fig. 14, the important words are highlighted in fluorescent color or the like and are displayed in white letters on a shaded background.

[0098] The minutes generation device 10 displays the minutes display screen 500 of Display Example 2, and when the user selects either the question summary or the answer summary in the first column 501, highlights key words related to the selected question summary or answer summary, as illustrated in Fig. 14. As described above, key words are extracted from the data to be processed when the generative model 20 generates the minutes data. This makes it possible to highlight key words in the second column 502, as illustrated in Fig. 14.

[0099] 14 illustrates a display mode in which only key words related to a selected summary are highlighted, but the present disclosure is not limited to this. For example, a sentence (question sentence or answer sentence) in the processed data that was used to generate a summary of a selected question or answer in the minutes data may be highlighted in a first mode, and particularly important key words in that sentence may be highlighted in a second mode.

[0100] Fig. 15 is a diagram showing two different highlighting styles in the second column 502 in display example 2. Fig. 15 shows the question sentence that was the basis for generating the summary of the selected question highlighted in a first style, and key words in the question sentence highlighted in a second style, on the minutes display screen 500 in the second display example. In Fig. 15, the highlighting style of the question sentence that was the basis for generating the summary of the selected question is shown in white letters on a black background, and the highlighting style of the key words in the question sentence is shown in white letters on a shaded background.

[0101] This display allows the user to understand whether the selected question summary or answer summary is from a sentence (question sentence or answer sentence) included in the data to be processed, and also accurately understands which words in that sentence were used to generate the summary, making it easy to evaluate the appropriateness of the generated summary.

[0102] 14 shows an example in which an answer summary is selected, and FIG. 15 shows an example in which a question summary is selected, but in the present disclosure, for example, a question summary and a corresponding answer summary may be selected at the same time. In this case, similar to FIG. 13B, the question sentence of the processing target data from which the selected question summary was generated and the answer sentence of the processing target data from which the selected answer summary was generated may be highlighted in the first manner in the same second column 502, and key words in the question sentence and the answer sentence may be highlighted in the second manner.

[0103] As shown in FIG. 14, the minutes display screen 500 further includes a summary result summary button 504 and a write in output document button 505.

[0104] The summary result summary button 504 is a button that causes the minutes generation device 10 to create a summary of at least one of the multiple question summaries and answer summaries when the minutes data displayed in the first column 501 includes multiple question summaries and answer summaries.

[0105] When the user presses the summary result summary button 504, the minutes generation device 10 first displays a selection button for each question summary and answer summary of the minutes data displayed in the first column 501, allowing the user to select whether to combine them with other question summaries and answer summaries. Fig. 16 is a diagram for explaining the screen transition of the minutes display screen 500 of display example 2 when the summary result summary button 504 is pressed by the user. Note that Fig. 16 illustrates only the first column 501 extracted from the minutes display screen 500 of Fig. 14, and the other illustrations are omitted.

[0106] In Fig. 16, when the user presses summary result summary button 504, a summary selection switch 506 is newly displayed in first column 501. Summary selection switch 506 is a switch that accepts the user's selection of whether or not to combine the question summary and answer summary with others. Summary selection switch 506 is displayed for each question summary and answer summary displayed in first column 501. Fig. 16 shows a state in which summary selection switch 506 has not yet been pressed by the user (hereinafter referred to as the OFF state).

[0107] When the user operates the summarization / non-summarization selection switch 506, and multiple summarization / non-summarization selection switches 506 are turned on, the minutes generation device 10 transitions the minutes display screen 500 to the screen shown in Fig. 17. Fig. 17 is a diagram for explaining the screen transition of the minutes display screen 500 of display example 2 when multiple summarization / non-summarization selection switches 506 are selected by the user. The third column 503 is not shown in Fig. 17.

[0108] 17, the grouping / non-grouping selection switch 506 is displayed in different colors depending on whether it is in a state selected by the user (hereinafter referred to as the ON state) or an OFF state, so that the user can recognize that it has been operated. However, in the present disclosure, various display modes other than different colors depending on whether it is in the ON state or the OFF state can be adopted as a display that allows the user to recognize that the grouping / non-grouping selection switch has been operated.

[0109] 17, when the user operates multiple summaries selection switches 506, a new summary display field 507 is displayed on the minutes display screen 500. The summary display field 507 is a field that displays a summary of the question summaries and answer summaries that the user has selected to combine with others (the summaries selection switches 506 are turned ON) from the question summaries and answer summaries in the first field 501.

[0110] However, the summary display field 507 includes a start summarizing button 508 and an apply button 509. After the user selects multiple question summaries and answer summaries to be summarized with others, the minutes generation device 10 generates a summary of the selected question summaries and answer summaries and displays it in the summary display field 507 when the user further operates the start summarizing button 508. In the example shown in Fig. 17, the start summarizing button 508 has not yet been operated by the user.

[0111] FIG. 18 is a diagram illustrating the screen transition of the minutes display screen 500 of display example 2 in a state in which the user operates the summary start button 508 and the summary results are displayed in the summary display field 507. The summary display field 507 displays the summary of the question summaries and answer summaries selected by the user from those displayed in the first field 501. As a result, the summary display field 507 always displays the question summary and the answer summary corresponding to the question in a combined state. This makes it possible to prevent, for example, a situation in which the question summary and the answer summary corresponding to the question are separated.

[0112] Furthermore, when the Apply button 509 is operated in this state, the content summarized in the summary display column 507 is displayed in the first column 501, instead of the multiple question summaries and answer summaries that were previously displayed. At this point, the summary selection switch 506 is not displayed in the first column 501. This prevents the question summary and the answer summary corresponding to that question from being separated.

[0113] When the user selects the question summaries and answer summaries to be summarized and operates the summarization start button 508, the minutes generating device 10 may generate a summary result as follows.

[0114] (First summary result generation method) For example, the minutes generation device 10 may generate a summary result by re-inputting the minutes data generated by the generative model 20 and the following summary generation prompt into the generative model 20.

[0115] (Example of a summary generation prompt) Please combine the following questions and answers into one question and answer. ## Output format - Question: {{One question}} - Answer: {{One Answer}} topic:{summary} According to this example of a prompt for generating a summary, a summary can be easily generated by using the generative model 20.

[0116] (Second summary result generation method) As described above, depending on the prompt used to generate the minutes data, the question summaries and answer summaries in the minutes data can be associated with the questions and answers in the data to be processed that served as the source for generating them. For example, the minutes generation device 10 extracts the question summaries and answers from the data to be processed that are used to generate the question summaries and answer summaries selected by the user to be summarized, and inputs the extracted questions and answers and the prompt used to generate the minutes data into the generative model 20. In this case, the generative model 20 ultimately generates question summaries and answer summaries that combine the question summaries and answer summaries selected by the user. This allows the minutes generation device 10 to obtain the desired summary result.

[0117] Returning to Fig. 14, the Write in Output Document button 505 will be described. The Write in Output Document button 505 is a button for writing only the question summary and answer summary selected by the user from the question summaries and answer summaries displayed in the first column 501 in a document file output by the minutes generation device 10. The document file to be output is a file that contains all or part of the minutes data in text format or HTML format.

[0118] When the user presses the Write in Output Document button 505, a description / non-description selection button 510 similar to the summary / non-description selection switch 506 illustrated in Figures 16 and 17 is displayed for each question summary and answer summary in the first column 501. Figure 19 is a diagram for explaining the screen transition of the minutes display screen 500 of display example 2 when the user operates the Write in Output Document button 505. Figure 19 illustrates only the first column 501 extracted from the minutes display screen 500 of Figure 14, with the other parts omitted.

[0119] As shown in Fig. 19, the user can select the question summaries and answer summaries that the user wants to include in the document file from the question summaries and answer summaries displayed in the first column 501 by operating the description selection button 510. When the user's selection is complete, the minutes generation device 10 creates a document file containing the question summaries and answer summaries selected by the user in text format and outputs the document file to an external device. Note that the document file may also include the question summaries and answer summaries that the user did not select in a state that makes it clear that they were not selected (for example, by a strikethrough). [Industrial Applicability]

[0120] The present disclosure is useful for a minutes generation device that automatically generates minutes data. [Explanation of symbols]

[0121] 10. Minutes Generator 11 processors 12 Input reception section 13 Display section 14 Communications Department 15 Storage section 41 Processing target data acquisition unit 42 Prompt Selection 43 Input processing section 44 Minutes Data Acquisition Department 45 Output processing section 100 Minutes Generation System 20 Generative Model 30 Network

Claims

1. A minutes generating device including a processor, The processor: inputting processing target data including at least questions and answers, and a prompt including an instruction to generate minutes data in a question-and-answer format based on the processing target data, into the generative model; A process of acquiring the minutes data generated by the generative model; A minutes generating device that executes the above.

2. The processor further executes a process of outputting the acquired minutes data. The minutes generating device according to claim 1 .

3. In the input process, the processor inputs the prompt, which includes an instruction to extract the questions and the answers from the data to be processed, and to generate the minutes data including a summary of the extracted questions and the answers, to the generative model. The minutes generating device according to claim 2 .

4. In the input process, the processor extracts the answers from the data to be processed, generates a summary of the extracted answers, and then extracts questions from the data to be processed that match the content of the generated answers, and inputs the prompt, which includes instructions to generate a summary of the extracted questions, to the generative model. The minutes generating device according to claim 3 .

5. In the output process, the processor causes a display device to display summaries of questions included in the minutes data and summaries of answers to the questions in association with each other. The minutes generating device according to claim 4.

6. In the output process, the processor causes the display device to display a second column in which the data to be processed is displayed, in addition to a first column in which summaries of the questions and answers included in the minutes data are displayed in association with each other. The minutes generating device according to claim 5 .

7. an identification number is associated with each of the questions and answers included in the processing target data; In the input process, the processor inputs to the generative model the prompt, which further includes an instruction to associate, with each of the summaries of the questions and answers included in the minutes data, the identification number indicating the question or the answer of the processing target data from which the summaries were generated. The minutes generating device according to claim 6.

8. The processor: further performing a process of accepting an operation to select at least one of the question summary or the answer summary from among the question summary and the answer summary displayed in the first column; In the output process, at least one question or answer that was the basis for generating the selected summary is highlighted in a first manner among the sentences constituting the processing target data displayed in the second column based on the identification number associated with at least one of the summaries selected by the operation. The minutes generating device according to claim 7.

9. The processor: In the input process, the prompt including an instruction to extract important words that are important to the summary of the question or the summary of the answer from the data to be processed is input to the generative model; In the output process, based on the key words extracted by the generative model, the key words contained in at least one question or answer that was the basis for generating the selected summary are highlighted in a second manner different from the first manner, among the sentences contained in the processing target data displayed in the second column. The minutes generating device according to claim 8.

10. The processor: In the input process, the prompt including an instruction to extract important words that are important to the summary of the question or the summary of the answer from the data to be processed is input to the generative model; In the output process, the key words are highlighted in the sentences included in the processing target data displayed in the second column based on the key words extracted by the generative model. The minutes generating device according to claim 6.

11. The processor: further performing a process of accepting an operation to select whether or not to combine each of the summaries of the question and the answer displayed in the first column with summaries of other questions and the answers; In the input process, the prompt including an instruction to generate a summary result summarizing the question and the answer selected by the operation is input to the generative model; In the output process, the summary result is displayed on the display device. The minutes generating device according to claim 6.

12. The processor: a process of accepting an operation to select whether or not to include each of the summaries of the question and the answer displayed in the first column in an output document to be output to an external device; creating an output document including a summary of the questions and answers selected by the operations; The minutes generating device according to claim 6 , further comprising:

13. the processor further executes a process of accepting an operation of selecting a prompt to be used in the input process from a plurality of predetermined prompts; In the input process, the prompt selected by the operation is input to the generative model. The minutes generating device according to claim 1 .

14. In the input process, the processor inputs the prompt, which includes an instruction to classify the questions and answers included in the data to be processed by topic, and generate the minutes data for each topic, to the generative model. The minutes generating device according to claim 2 .

15. In the input process, the processor inputs a topic list including a list of topics corresponding to each of the questions and answers included in the data to be processed together with the prompt to the generation model, thereby classifying the questions and answers included in the data to be processed by topic. The minutes generating device according to claim 14.

16. In the output process, the processor causes a display device to display, for each topic, summaries of questions included in the minutes data and summaries of answers to the questions in association with each other. The minutes generating device according to claim 14 or 15.

17. 1. A computer-implemented method for generating minutes of a meeting, comprising: The computer inputting processing target data including at least questions and answers, and a prompt including an instruction to generate minutes data in a question-and-answer format based on the processing target data, into the generative model; A process of acquiring the minutes data generated by the generative model; A method for generating minutes of a meeting.

18. On the computer, inputting processing target data including at least questions and answers, and a prompt including an instruction to generate minutes data in a question-and-answer format based on the processing target data, into the generative model; A process of acquiring the minutes data generated by the generative model; A program that executes.

Images