Information processing system and program

Abstract

We provide an information processing system and program that can generate minutes with high accuracy using an AI system that generates minutes from the full text of a meeting. [Solution] The information processing system 10 comprises a meeting data acquisition unit 30 that acquires meeting content data, a minutes generation unit 34 that causes a generation AI system 15 to generate minutes, and an information output unit 35 that presents the minutes to the user. The minutes generation unit 34 sends first data including the meeting content data and a first minutes generation prompt including instructions to generate minutes to the generation AI system 15, and upon receiving first draft minutes data from the generation AI system 15, sends to the generation AI system 14 second data including the first draft minutes data and the first data, and a second minutes generation prompt including instructions to generate minutes again by comparing the first draft minutes data with the first data, and converts the second draft minutes data received from the generation AI system 14 into minutes.

Description

[Technical Field]

[0001] The present invention relates to an information processing system and program that uses a generation AI system to generate minutes from the full text of a meeting. [Background technology]

[0002] In recent years, services have been provided that generate sentences interactively with users using generative AI systems that use large-scale language models (LLMs). Many services using such generative AI systems have been provided. For example, Patent Document 1 discloses a system that generates information (prompts) to be input to a generative AI system without user intervention, transmits the prompts to the generative AI system, retrieves the generated information, and provides it to the user. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7416390 Summary of the Invention [Problem to be solved by the invention]

[0004] One possible use of generative AI systems is to automatically generate minutes from the full text of a meeting. While generative AI systems using LLM can efficiently process natural language, the accuracy of the output is not guaranteed, as they can cause hallucinations. As a result, there is a risk of inconsistencies between what was discussed in the meeting and the contents of the minutes, and the accuracy of the minutes may not be fully ensured.

[0005] The present invention has been made in consideration of the above-mentioned problems, and aims to provide an information processing system and program that can generate minutes with high accuracy from the full text of a meeting using a generation AI system. [Means for solving the problem]

[0006] In order to solve the above problem, the information processing system of the present invention is an information processing system that generates minutes from meeting content data obtained by transcribing the audio of a meeting, and includes a control means and a storage means, wherein the control means includes a conference data acquisition unit that acquires the meeting content data and stores it in the storage means, a minutes generation unit that transmits data including the meeting content data stored in the storage means and a minutes generation prompt including an instruction to generate minutes from the meeting content data to a generation AI system, acquires the minutes from the generation AI system, and stores them in the storage means, and an information output unit that presents the minutes stored in the storage means to a user. The minutes generation unit transmits to the generation AI system first data including the meeting content data and a first minutes generation prompt including an instruction to generate minutes from the meeting content data, and upon receiving first draft minutes data from the generation AI system, transmits to the generation AI system second data including the first draft minutes data and the first data and a second minutes generation prompt including an instruction to regenerate minutes after comparing the first draft minutes data with the first data to confirm consistency, receives second draft minutes data from the generation AI system, and stores the second draft minutes data in the storage means as the minutes.

[0007] In addition, in the information processing system of the present invention, the storage means pre-stores minutes format information including format data for minutes and dictionary data associating at least spellings with pronunciations, and the minutes generation unit includes the dictionary data and the minutes format information in addition to the meeting content data in the first data, and includes in the first minutes generation prompt an instruction to refer to the minutes format information and the dictionary data when generating the minutes.

[0008] In addition, in the information processing system of the present invention, the minutes generation unit includes in the second minutes generation prompt an instruction to refer to the minutes format information and the dictionary data when generating the minutes.

[0009] In addition, the information processing system according to the present invention is the information processing system, wherein the minutes generation unit receives from the generation AI system. It is text data The first draft data of the minutes and the second draft data of the minutes are stored in the storage means, respectively, and the first draft data of the minutes is stored in the storage means. manuscript Data and the second draft of the minutes Text data The vectorized data is stored in the storage means.

[0010] In order to solve the above-mentioned problems, the program of the present invention is a program for generating minutes from meeting content data obtained by transcribing the audio of a meeting, and causes a computer to execute a meeting data acquisition process for acquiring the meeting content data and storing it in a storage means, a minutes generation process for transmitting data including the meeting content data stored in the storage means and a minutes generation prompt including an instruction to generate minutes from the meeting content data to a generation AI system, and acquiring the minutes from the generation AI system and storing them in the storage means, and an information output process for presenting the minutes stored in the storage means to a user, thereby generating the minutes. In the processing, first data including the meeting content data and a first minutes generation prompt including instructions to generate minutes from the meeting content data are sent to the generation AI system, and upon receiving first draft minutes data from the generation AI system, second data including the first draft minutes data and the first data and a second minutes generation prompt including instructions to compare the first draft minutes data with the first data to confirm consistency and then generate minutes again are sent to the generation AI system, second draft minutes data is received from the generation AI system, and the second draft minutes data is stored in the storage means as the minutes. [Effects of the Invention]

[0011] According to the information processing system and program of the present invention, the generation AI system generates first draft minutes data from first data including meeting content data, and second data combining the first draft minutes data and the first data is sent to the generation AI system, which then confirms the consistency between the first draft minutes data and the first data and generates second draft minutes data.This makes it possible to suppress inconsistencies and hallucinations in the text that the generation AI system may generate at the time of the first draft minutes data, and to generate highly accurate minutes from the meeting content data. [Brief explanation of the drawings]

[0012] [Figure 1] 1 shows a configuration diagram of an information processing system according to an embodiment of the present invention. [Figure 2] 1 shows a detailed configuration diagram of a control means. [Figure 3] 1 is a flowchart in an information processing system. [Figure 4] 10 is a flowchart of the first half of the process of generating minutes, in which voice data is acquired, a meeting entity is identified, and unknown words are completed. [Figure 5] 10 is a flowchart of the latter half of the process of generating minutes of a meeting from the meeting content data in which unknown words have been complemented, and presenting the minutes to the user. DETAILED DESCRIPTION OF THE INVENTION

[0013] An embodiment of the present invention will be described in detail with reference to the drawings. An information processing system 10 of this embodiment is a system that generates minutes from the full text of a meeting using a generation AI system 15.

[0014] The information processing system 10 uses a generative AI system 15 equipped with an LLM to estimate the meeting body, which indicates the type of meeting, and generate meeting minutes. An LLM (Large-Scale Language Model) is a language model constructed using large amounts of text data and deep learning technology. By using a Transformer model and training it with a huge data set, the LLM performs natural language processing and is capable of human-like natural dialogue and sentence generation. The generative AI system 15 can receive prompts written in natural language from the information processing system 10 and transmit corresponding AI output information. In addition to text data, the generative AI system 15 can also receive and analyze data such as image data and audio data.

[0015] The information processing system 10 of this embodiment generates minutes from meeting content data obtained by transcribing the audio of a meeting. The meeting content data is text data. The information processing system 10 identifies which meeting entity the meeting content data belongs to. A meeting entity refers to a group of people consisting of specific participants for a specific purpose. Specific examples of meeting entities include, for example, the following: ·Progress meeting for XX implementation project for XX General Affairs Department monthly regular meeting ·△△Full-stack development research presentation Third District Disaster Prevention Community Meeting · Marketing and development department cross-department meeting Brainstorming sessions for new AI services

[0016] 1, an information processing system 10 can be connected to a user terminal 12 via a communication network 16. The user terminal 12 has a display means 13 such as a monitor. The user terminal 12 is a computer such as a PC, or a mobile information terminal such as a smartphone. However, the user terminal 12 is not limited to these.

[0017] The information processing system 10 can be connected to a generative AI system 15 via a communication network 16. In FIG. 1, the information processing system 10 is connected to one generative AI system 15, but the information processing system 10 may use multiple generative AI systems 15 as needed. The information processing system 10 itself may include a generative AI system 15. In this embodiment, information is exchanged between the information processing system 10 and the generative AI system 15 by sending and receiving information, but if the information processing system 10 includes a generative AI system 15, information is communicated within the system.

[0018] In Figure 1, the information processing system 10 is connected to multiple user terminals 12 and the generation AI system 15 via a single communication network 16, but the information processing system 10 may also be connected to the user terminals 12 and the generation AI system 15 via separate communication networks. For example, the information processing system 10 can be connected to the user terminals 12 via an in-house LAN and to the generation AI system 15 via the Internet.

[0019] The information processing system 10 can be configured with a computer such as a server. The information processing system 10 includes at least a control means 20, a storage means 21, and a communication means 22. The control means 20 can be configured with a processor such as a CPU. The storage means 21 can be configured using, for example, RAM, DRAM, HDD, SSD, or an external cloud or distributed storage. The communication means 22 can be configured with, for example, a NIC or a wireless LAN connection device.

[0020] The storage means 21 of the information processing system 10 pre-stores, for each conference entity, conference entity attribute data that associates basic information about the conference entity with past minutes data of the conference entity, and dictionary data provided for each conference entity and that associates at least spellings and pronunciations. The basic information in the conference entity attribute data includes information such as the conference name, participants, contact information such as participants' email addresses, and the type of conference. The past minutes data included in the conference entity attribute data may be the text data of previously generated minutes, or may be text data of a meeting summary generated when the conference entity generation AI system 15 (described below) estimates the conference entity. The dictionary data associates not only the spellings and pronunciations of words, but also the type of word (personal name, organization name, product name, technical term, etc.). The dictionary data may also associate supplemental information, such as titles corresponding to people's names. Words registered in the dictionary data are primarily proper nouns and technical terms, which are unique to each conference entity, but other words may also be registered.

[0021] The storage means 21 also stores data on past minutes of meetings. The minutes data is stored as text data and vectorized data of the text data. The vectorized data can be used to determine the degree of similarity with new minutes or meeting content data.

[0022] The storage means 21 also stores minutes format information, which includes format data for minutes. The minutes that are generated include basic information about the meeting, such as the name, date, time, location, and attendees, information about the meeting content, such as the agenda, summary, agenda, decisions, and actions, and information about the next meeting. The minutes format information specifies how this information is to be arranged in the minutes.

[0023] As shown in Figure 2, the control means 20 includes a conference data acquisition unit 30 that acquires conference content data, a conference data analysis unit 31 that causes the generation AI system 15 to generate a list of candidate conference entities and a list of unknown words in the conference content data based on the conference entity attribute data and the conference content data, a conference entity determination unit 32 that presents the candidate conference entity list to a user and determines the conference entity, a data correction unit 33 that complements unknown words in the conference content data with dictionary data corresponding to the conference entity and creates corrected conference content data, a minutes generation unit 34 that sends the corrected meeting content data and a minutes generation prompt to the generation AI system 15 to generate minutes, and an information output unit 35 that presents the generated minutes to a user.

[0024] Next, a flow in which a user causes the information processing system 10 to generate minutes of a conference will be described. It is assumed that the user has audio data of a recorded conference. As shown in FIG. 3, first, the conference data acquisition unit 30 acquires the audio data of the conference from the user (S1). The conference data acquisition unit 30 can acquire the audio data by the user uploading the audio data from the user terminal 12 to the information processing system 10. At this time, the conference data acquisition unit 30 may also acquire additional information about the conference, such as the date and time of the conference, location, participant information, and the creator of the minutes, by having the user input the information.

[0025] The conference data acquisition unit 30 converts the acquired voice data to an optimal format and size for uploading to the generation AI system 15, and then transmits the converted voice data to the generation AI system 15 (S2). At this time, the conference data acquisition unit 30 transmits to the generation AI system 15 a prompt including an instruction to transcribe the attached voice data.

[0026] The conference data acquisition unit 30 receives conference content data obtained by transcribing the voice data from the generation AI system 15 (S3), and stores the received conference content data in the storage means 21.

[0027] Next, the conference data analysis unit 31 generates a conference data analysis prompt including an instruction to generate a candidate list of conference entities based on the conference entity attribute data and the conference content data, and an instruction to generate a list of unknown words in the conference content data (S4).

[0028] Specifically, the meeting data analysis prompt may include the following instructions: First, the analysis should involve (1) reading the entire meeting content data to identify the type of meeting, (2) identifying the predominant language used in the meeting content data, (3) understanding the roles and relationships of the meeting participants, and (4) identifying the main purpose of the meeting. Based on the results of this analysis, the main purpose of the meeting can be more specifically identified by referring to the meeting entity attribute data. To do this, the prompt instructs participants to consider aspects such as explicit explanations of the purpose in the opening and closing sections, as well as repeated mentions of topics and issues. Next, the meeting data analysis prompt instructs participants to identify the meeting entity from the content of the meeting content data. Guidelines for identifying meeting entities include proactively including clear customer or client names in the meeting entity type, excluding names that appear to be the participant's own company name, and retaining product and project names. Furthermore, participants are instructed to use specific and distinctive language when identifying meeting entities to effectively group similar meetings. The meeting data analysis prompt includes instructions for extracting and listing unrecognized or potentially erroneous words in the meeting content data to generate a list of unknown words in the meeting content data. The meeting data analysis prompt also includes instructions for generating a meeting summary from the meeting content data.

[0029] After generating the conference data analysis prompt, the conference data analysis unit 31 attaches the conference entity attribute data and the conference content data and sends the conference data analysis prompt to the generation AI system 15 (S5). In response to this, the conference data analysis unit 31 receives a list of candidate conference entities and a list of unknown words from the generation AI system 15 (S6). The candidate conference entity list includes one or more names of conference entities that the generation AI system 15 inferred from the conference content data. The conference data analysis unit 31 also receives information on a summary of the conference based on the conference content data from the generation AI system 15. The conference data analysis unit 31 stores the received information on the summary of the conference in the storage means 21.

[0030] The conference data analysis unit 31 reexamines the candidate list of conference entities to improve accuracy and determine priorities (S7). As shown in FIG. 4, the conference data analysis unit 31 extracts from past minutes data those conference entities included in the candidate list of conference entities that have a high degree of similarity (S7-1). As described above, the minutes data stored in the storage unit 21 is vectorized and saved. The conference data analysis unit 31 vectorizes the names of conference entities and extracts multiple minutes data similar to the conference entities from the vectorized minutes data in the storage unit 21. Here, a total of 30 to 50 minutes data items are extracted from the candidate list of conference entities that includes the names of multiple conference entities. Note that the number of similar minutes data items to be extracted is not limited to this range.

[0031] Next, the conference data analysis unit 31 attaches the extracted similar minutes data and conference content data and sends a conference entity reconsideration prompt to the generation AI system 15 (S7-2). The conference entity reconsideration prompt includes instructions to identify a conference entity that can be estimated from the conference content data by referring to the candidate conference entity list and similar minutes data acquired in S6, and instructions to score the similarity between the identified conference entity and the similar minutes. The conference data analysis unit 31 receives from the generation AI system 15 a list in which a score of similarity with past minutes data is assigned for each candidate conference entity (S7-3). As a result, a score of similarity with past minutes data for each conference entity is obtained for the candidate conference entity list that includes the names of multiple conference entities.

[0032] Next, the meeting entity determination unit 32 presents a list of candidate meeting entities to the user (S8). The candidate meeting entities presented to the user are displayed in descending order of similarity score with past minutes data. Furthermore, if there is no meeting entity included in the candidate list that has a similarity score with past minutes data of a certain level or higher, a new meeting entity is added to the candidate list and presented to the user. If the user selects a meeting entity presented as a candidate (S9), the process proceeds to S11. If the user does not select a candidate meeting entity in S9, or if the user creates a name for the meeting entity (S10), the meeting entity determination unit 32 receives the name and proceeds to S11. Furthermore, if the user does not create a name for the meeting entity in S10, the steps from S8 are repeated. That is, the meeting entity determination unit 32 presents the next candidate meeting entity to the user. Through the steps up to this point, the meeting entity corresponding to the meeting content data is determined.

[0033] When the user selects a candidate meeting entity or creates a name for a meeting entity, the meeting entity determination unit 32 determines whether the meeting entity is a new meeting entity (S11). Whether the meeting entity is a new meeting entity can be determined by referring to the meeting entity attribute data stored in the storage means 21. If the meeting entity selected or created by the user is a new meeting entity, the meeting entity determination unit 32 adds the new meeting entity to the meeting entity attribute data (S12) and proceeds to S14.

[0034] If the meeting entity selected or created by the user is an existing meeting entity, the data correction unit 33 complements unknown words in the meeting content data (S13). To this end, the data correction unit 33 reads out dictionary data corresponding to the meeting entity confirmed by the meeting entity confirmation unit 32 from the storage means 21, and complements words corresponding to the unknown word list for the meeting content data received from the generation AI system 15 based on the dictionary data. The data correction unit 33 stores the meeting content data with the complemented unknown words in the storage means 21 as modified meeting content data.

[0035] The data correction unit 33 presents to the user, among the unknown words in the unknown word list, unknown words that cannot be completed using dictionary data corresponding to the meeting entity confirmed by the meeting entity confirmation unit 32, and prompts the user to input a word so that the spelling and pronunciation correspond (S14). At this time, the data correction unit 33 also corrects the remaining unknown words in the meeting content data to the spelling entered by the user. The data correction unit 33 adds the user's input to the dictionary data and stores it in the storage means 21 (S15). In the steps up to S15, the meeting entity is identified for the meeting content data from which minutes are generated, and unknown words in the meeting content data are completed using dictionary data corresponding to the identified meeting entity. This increases the accuracy of the meeting content data, which is the original data from which minutes are generated. High accuracy of the meeting content data also increases the accuracy of the minutes generated by the generation AI system 15. Furthermore, by prompting the user to register words that could not be completed using dictionary data in the dictionary, the meeting content data can be made more accurate and more unknown words can be automatically completed for subsequent meeting content data.

[0036] After performing the steps up to S15, the information processing system 10 proceeds to the step of generating minutes. As shown in FIG. 5, the minutes generation unit 34 generates a first minutes generation prompt for causing the generation AI system 15 to generate minutes (S16). The first minutes generation prompt includes instructions to create minutes from the meeting content data. The first minutes generation prompt also includes instructions to analyze the meeting content data to understand the content of the meeting, to refer to attached dictionary data to make the notation accurate, and to reflect any additional information related to the meeting. The first minutes generation prompt also includes instructions to refer to attached minutes format information and generate minutes while maintaining its structure and item names.

[0037] The minutes generation unit 34 attaches the first data to the first minutes generation prompt and transmits it to the generation AI system 15 (S17). The first data includes meeting content data, dictionary data corresponding to the meeting entity of the meeting content data, and minutes format information.

[0038] The minutes generation unit 34 receives the first draft minutes data as a response from the generation AI system 15 to the first minutes generation prompt (S18). The minutes generation unit 34 stores the received first draft minutes data in the storage means 21. Upon receiving the first minutes data, the minutes generation unit 34 generates a second minutes generation prompt (S19). In addition to the instructions included in the first minutes generation prompt, the second minutes generation prompt includes an instruction to compare the first draft minutes data with the first data to confirm consistency and then generate minutes.

[0039] The minutes generation unit 34 attaches the second data to the second minutes generation prompt and sends it to the generation AI system 15 (S20). The second data includes the first data and the first draft minutes data from S17. The minutes generation unit 34 receives the second draft minutes data as a response from the generation AI system 15 to the second minutes generation prompt (S21). The minutes generation unit 34 stores the received second draft minutes data as text data in the storage means 21. In this way, the minutes generation unit 34 causes the generation AI system 15 to generate the first draft minutes data from the first data including the meeting content data, and sends the second data combining the first draft minutes data and the first data to the generation AI system 15, where it checks the consistency between the first draft minutes data and the first data and generates the second draft minutes data. This allows the information processing system 10 to suppress inconsistencies and hallucinations in the text that the generation AI system 15 may generate at the time of the first draft of the minutes data, and enables highly accurate minutes to be generated from the meeting content data.

[0040] The minutes generation unit 34 also vectorizes the first draft minutes data and the second draft minutes data, and stores the vectorized data in the storage means 21 (S23). The vectorization involves chunking the text data of the first draft minutes data and the second draft minutes data, and embedding the chunked data. The chunking and embedding may be performed by the information processing system 10, or may be performed using an external service.

[0041] By vectorizing the first and second draft minutes data stored as minutes data, the minutes data can be used as RAG (Retrieval Augmented Generation) data grouped by meeting body. When a user uses the generation AI system 15, the RAG data is used as a knowledge base that the generation AI system 15 references. Therefore, when a user asks a question about the content of a specific meeting body, the generation AI system 15 can provide an appropriate answer based on the vectorized data, reducing the user's effort to search for minutes data. Possible questions from a user referencing the RAG data include, for example, asking for details about decisions made at a meeting or whether a specific agenda item has been discussed in the past.

[0042] Next, the information output unit 35 extracts the members of the meeting entity corresponding to the generated minutes from the meeting entity attribute data, and notifies each member that the minutes have been generated by sending an email or the like (S24). Furthermore, the information output unit 35 transmits the data of the generated minutes to the user terminal 12 and presents it to the user (S25). This allows the user to obtain the minutes generated from the meeting content data.

[0043] Also, the program used in the information processing system 10 of this embodiment can be provided independently. In this case, the program causes a computer to execute a conference data acquisition process for acquiring conference content data and storing it in the storage means 21, a minutes generation process for transmitting data including the conference content data stored in the storage means 21 and a minutes generation prompt including an instruction to generate minutes from the conference content data to the generation AI system 15, and acquiring the minutes from the generation AI system 15 and storing them in the storage means 21, and an information output process for presenting the minutes stored in the storage means 21 to a user. In the minutes generation process, first data including the conference content data and A first minutes generation prompt including an instruction to generate minutes from the meeting content data is sent to the generation AI system 15, and upon receiving first draft minutes data from the generation AI system 15, second data including the first draft minutes data and the first data and a second minutes generation prompt including an instruction to generate minutes again after comparing the first draft minutes data with the first data to confirm consistency are sent to the generation AI system 15, and second draft minutes data is received from the generation AI system 15 and the second draft minutes data is stored in the storage means 21 as minutes.

[0044] Although the above describes an embodiment of the present invention, the application of the present invention is not limited to this embodiment and may be variously applied within the scope of its technical concept. In this embodiment, the information processing system 10 causes the generation AI system 15 to generate minutes data twice. However, the generation AI system 15 may also generate minutes data three or more times. In this case, when generating the third or subsequent drafts of minutes data, the information processing system 10 transmits to the generation AI system 15 the minutes data and the first data generated up to that point, including the most recently generated minutes data, along with a minutes generation prompt including an instruction to generate the minutes again after comparing the minutes data with the first data to confirm consistency, and an instruction to reference minutes format information and dictionary data when generating the minutes, and acquires the third or subsequent drafts of minutes data. Furthermore, when generating the third or subsequent drafts of minutes data, the minutes generation unit 34 can store text data and vectorized data for all minutes data in the storage unit 21. [Explanation of symbols]

[0045] 10 Information Processing Systems 12 User terminals 13 Display means 15 Generative AI System 17. Communication Networks 20 Control Means 21 Memory means 22 Means of communication 30 Conference Data Acquisition Unit 31 Conference Data Analysis Department 32 Meeting Body Confirmation Division 33 Data Correction Department 34 Minutes Generation Department 35 Information output section

Claims

1. An information processing system for generating minutes from meeting content data obtained by transcribing the audio of a meeting, comprising control means and storage means; The control means a conference data acquisition unit that acquires the conference content data and stores it in the storage means; a minutes generation unit that transmits data including the meeting content data stored in the storage means and a minutes generation prompt including an instruction to generate minutes from the meeting content data to a generation AI system, and acquires the minutes from the generation AI system and stores them in the storage means; an information output unit that presents the minutes stored in the storage means to a user, The minutes generation unit sends to the generation AI system first data including the meeting content data and a first minutes generation prompt including instructions to generate minutes from the meeting content data, and upon receiving first draft minutes data from the generation AI system, sends to the generation AI system second data including the first draft minutes data and the first data and a second minutes generation prompt including instructions to compare the first draft minutes data with the first data to confirm consistency and then generate minutes again, receives second draft minutes data from the generation AI system, and stores the second draft minutes data in the storage means as the minutes.

2. The storage means stores in advance minutes format information including format data for minutes and dictionary data associating at least notation with pronunciation, The information processing system of claim 1, wherein the minutes generation unit includes the dictionary data and the minutes format information in addition to the meeting content data in the first data, and includes instructions in the first minutes generation prompt to refer to the minutes format information and the dictionary data when generating the minutes.

3. The information processing system according to claim 2 , wherein the minutes generating unit includes in the second minutes generating prompt an instruction to refer to the minutes format information and the dictionary data when generating the minutes.

4. An information processing system according to any one of claims 1 to 3, wherein the minutes generation unit stores the first draft minutes data and the second draft minutes data, which are text data received from the generation AI system, in the storage means, and stores vectorized data of the text data of the first draft minutes data and the second draft minutes data in the storage means.

5. A program for generating minutes from meeting content data obtained by transcribing the audio of a meeting, On the computer, a conference data acquisition process for acquiring the conference content data and storing it in a storage means; a minutes generation process for transmitting data including the meeting content data stored in the storage means and a minutes generation prompt including an instruction to generate minutes from the meeting content data to a generation AI system, and acquiring the minutes from the generation AI system and storing them in the storage means; an information output process for presenting the minutes stored in the storage means to a user; Execute In the minutes generation process, the program sends first data including the meeting content data and a first minutes generation prompt including instructions to generate minutes from the meeting content data to the generation AI system, and upon receiving first draft minutes data from the generation AI system, sends second data including the first draft minutes data and the first data and a second minutes generation prompt including instructions to compare the first draft minutes data with the first data to confirm consistency and then generate minutes again to the generation AI system, receives second draft minutes data from the generation AI system, and stores the second draft minutes data in the storage means as the minutes.

Images