System

Abstract

A system is provided.SOLUTION: A system, comprising: means for obtaining a historical meeting record; means for sending the obtained meeting record to a summary generation model; means for storing a summary received from the summary generation model; and means for providing the stored summary to a user terminal.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The technology of the present disclosure relates to a system. [Background technology]

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] In meetings, differences in understanding occur depending on the timing of participants' participation and their backgrounds, which can slow down the progress of the meeting. It is also difficult for people who join midway through a meeting or for new members of a project to quickly grasp the content of past discussions. This results in a decrease in meeting productivity and makes it difficult to hold effective discussions. [Means for solving the problem]

[0005] This invention provides a system including a means for acquiring conference records, a means for transmitting the acquired conference records to a summary generation model, a means for saving summaries received from the summary generation model, and a means for providing the saved summaries to a user terminal. Furthermore, by including a means for generating answers to specific questions from past conference records and a means for providing the generated answers to the user terminal, the system allows users to quickly understand the content of past discussions. Furthermore, by including a means for predicting the content of the next discussion based on past conference records and summaries and a means for providing the predicted content of the discussion to the user terminal, the system allows efficient preparation for the next conference. In this way, differences in understanding among conference participants can be bridged, the conference can proceed smoothly, and productivity can be improved.

[0006] "Past meeting records" are records of discussions, decisions, and participants' comments from previous meetings, recorded in text format or audio data.

[0007] A "summary generation model" is an algorithm or software that uses a large-scale language model or the like to extract important points from given text data and summarize them in a concise form.

[0008] A "user terminal" is a device used by a user, such as a computer, smartphone, tablet, etc., that has input and display means for interfacing with the system.

[0009] A "question answering model" is an algorithm or software that generates the best answer to a specific question based on relevant data.

[0010] A "means for predicting the next discussion topic" is an algorithm or software that uses past meeting records and summaries to predict the discussions and topics that are likely to be discussed in the next meeting. [Brief explanation of the drawings]

[0011] [Figure 1]1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6] FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11] FIG. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0012] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0013] First, the terms used in the following description will be explained.

[0014] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, a processor may be a single type of arithmetic device or a combination of multiple types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), and an APU (Accelerated Processing Unit).

[0015] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0016] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0017] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0018] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0019] [First embodiment]

[0020] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0021] 1, a data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0022] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0023] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0024] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0025] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0026] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0027] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0028] 2, in the data processing device 12, a specific process is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific process is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0029] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0030] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0031] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0032] This paper describes a system for smoothly progressing a conference by bridging differences in understanding between participants due to their timing and background. The system includes a server, user terminals, and a large-scale language model (LLM).

[0033] Summary generation function

[0034] System configuration and operation

[0035] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted to text. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts key points from the meeting records and generates a summary. The generated summary is stored by the server and provided to the user's device before the next meeting.

[0036] Specific examples

[0037] Before joining the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[0038] Question and Answer Function

[0039] System configuration and operation

[0040] A user uses a terminal to enter a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[0041] Specific examples

[0042] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[0043] Prediction function for next discussion topic

[0044] System configuration and operation

[0045] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the likely discussions and topics that will be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[0046] Specific examples

[0047] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[0048] As described above, this system uses past meeting records to summarize, answer questions, and predict the content of the next discussion, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[0049] The processing flow will be explained below.

[0050] Summary generation function

[0051] Processing flow

[0052] Step 1:

[0053] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[0054] Step 2:

[0055] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[0056] Step 3:

[0057] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[0058] Step 4:

[0059] The server receives the generated summaries from the LLM and stores them in an internal database.

[0060] Step 5:

[0061] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[0062] Step 6:

[0063] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[0064] Question and Answer Function

[0065] Processing flow

[0066] Step 1:

[0067] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[0068] Step 2:

[0069] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[0070] Step 3:

[0071] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[0072] Step 4:

[0073] LLM analyzes the question and relevant meeting notes to generate the best possible answer.

[0074] Step 5:

[0075] The server receives the generated response from the LLM and returns it to the user terminal.

[0076] Step 6:

[0077] The user checks the answers on the terminal and obtains the necessary information.

[0078] Prediction function for next discussion topic

[0079] Processing flow

[0080] Step 1:

[0081] The server retrieves past meeting records and summaries from an internal database.

[0082] Step 2:

[0083] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[0084] Step 3:

[0085] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[0086] Step 4:

[0087] The server receives predicted discussion content from the LLM and stores it in an internal database.

[0088] Step 5:

[0089] When a user prepares for their next meeting, they request the prediction results from their device.

[0090] Step 6:

[0091] The server transmits the saved prediction results to the terminal.

[0092] Step 7:

[0093] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[0094] Example 1

[0095] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0096] Modern companies hold many meetings, but participants' understanding of the content often varies depending on their timing and background. It's also difficult for new participants to quickly understand the content of past meetings, which can slow down the progress of the meeting. Furthermore, searching for necessary information from past meeting records and predicting the content of the next meeting takes time and effort. A system that solves these problems and supports productive meetings is needed.

[0097] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0098] In this invention, the server includes means for acquiring past meeting records, means for transmitting the acquired meeting records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for transmitting to a question-answering model that generates answers to specific questions, means for providing to the user terminal the answers received from the question-answering model, means for transmitting to a prediction model that predicts the content of the next discussion based on the past meeting records and summaries, and means for providing to the user terminal the discussion content received from the prediction model. This enables users to quickly understand the content of past meetings, easily obtain necessary information, and efficiently prepare for the next meeting.

[0099] "Past meeting records" refers to text data or audio data recorded in a previous meeting.

[0100] A "summary generation model" refers to a large-scale language model that can extract important points from long texts and summarize them concisely.

[0101] A "question answering model" refers to a large-scale language model that analyzes relevant information based on a given question and generates an appropriate answer.

[0102] A "predictive model" refers to a large-scale language model that has the ability to predict future events or outcomes based on past data.

[0103] A "database" refers to a system for efficiently storing, managing, and searching data in various formats.

[0104] A "server" refers to a computer system that receives requests from user terminals, performs the necessary processing, and provides the data.

[0105] "User terminal" refers to a device used by a user, such as a computer, smartphone, or tablet.

[0106] A "generative AI model" refers to an artificial intelligence model that uses a large-scale language model to generate output that is tailored to a specific purpose.

[0107] A "prompt sentence" refers to input text that provides specific instructions to a generative AI model.

[0108] The present invention relates to a system for smoothly progressing a meeting by bridging differences in understanding due to the timing and background of each participant. The system includes a server, a user terminal, and a large-scale language model (LLM).

[0109] Summary generation function

[0110] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted into text. The retrieved meeting record data is sent to an LLM (e.g., GPT-4) specialized in generating text summaries. The LLM extracts key points from the provided meeting records and generates a summary, which is then stored by the server. Before the next meeting, the user can request a past meeting summary from the server, and the summary is provided to the user's device.

[0111] Specific examples

[0112] Before joining the next meeting, a user can request a summary of the previous meeting from the server. The server then sends the summary to the terminal, allowing the user to quickly review the important points and efficiently prepare for the meeting.

[0113] Example prompt statement:

[0114] Summarize past meeting notes, including key points and organizing them so users can review them before the next meeting.

[0115] Question and Answer Function

[0116] A user uses a terminal to input a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[0117] Specific examples

[0118] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[0119] Example prompt statement:

[0120] Please tell me about the budget that was decided at the last meeting.

[0121] Prediction function for next discussion topic

[0122] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model predicts the discussions and topics likely to be discussed in the next meeting based on past data. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[0123] Specific examples

[0124] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[0125] Example prompt statement:

[0126] Anticipate topics likely to be discussed at the next meeting. Use past meeting notes and summaries to inform future discussions.

[0127] This system allows users to quickly understand the content of past meetings, easily obtain the necessary information, and efficiently prepare for the next meeting, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[0128] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0129] Summary generation function

[0130] Step 1:

[0131] Get past meeting records.

[0132] The server accesses the company's internal database and retrieves past meeting records using SQL queries, etc. The retrieved input data is text data or audio data converted into text.

[0133] What happens: The server retrieves data from the database using an API call, executing an SQL query like "SELECT FROM meeting_records_table WHERE date > '2022-01-01'".

[0134] Input: Meeting recordings from the company's internal database (text format or speech-to-text transcription)

[0135] Output: Past meeting recording data

[0136] Step 2:

[0137] Send the meeting notes to a summary generation model.

[0138] The server sends the acquired meeting notes to an LLM model (e.g., GPT-4) specialized in generating text summaries.

[0139] Specific behavior: Convert the meeting recording to JSON format and send the data to the model via API. Include the prompt "Please summarize this meeting recording."

[0140] Input: Past meeting recording data

[0141] Output: Data request sent to the summary generation model

[0142] Step 3:

[0143] Generate a summary.

[0144] The LLM model extracts key points from the provided meeting notes and generates a summary.

[0145] How it works: Large-scale language models such as GPT-4 analyze meeting recordings and use natural language processing (NLP) techniques to extract key parts.

[0146] Input: Meeting recording data

[0147] Output: Generated summary

[0148] Step 4:

[0149] Save the summary.

[0150] The server receives the generated summary and stores it in a database.

[0151] Specific operation: Convert the summary data into JSON format and insert it into the "Summary" table in the database. Execute an SQL query such as "INSERT INTO SummaryTable(ConferenceID, Summary) VALUES(..., ...)".

[0152] Input: Generated summary

[0153] Output: Summary stored in database

[0154] Step 5:

[0155] Provided to the user terminal.

[0156] Before the next meeting, the user requests a summary of the past meeting from the terminal, and the server provides the summary to the user terminal.

[0157] Specific operation: The user enters "Request past conference summaries" on the terminal and clicks the send button. The server retrieves the relevant summaries from the database and returns them to the user terminal in JSON format.

[0158] Input: A request from the user

[0159] Output: Summary sent to user terminal

[0160] Question and Answer Function

[0161] Step 1:

[0162] Enter and submit a specific question.

[0163] The user uses a terminal to enter a specific question and send it to the server.

[0164] Specific operation: The user types "What was the budget decided at the last meeting?" into the terminal and clicks the send button.

[0165] Input: User question

[0166] Output: The question data sent to the server

[0167] Step 2:

[0168] Receive questions.

[0169] The server receives a query from a user and prepares to search for relevant meeting records.

[0170] Specific operation: The server analyzes the question and retrieves the relevant meeting records from the database.

[0171] Input: Question data from the user

[0172] Output: Data search plan within the server

[0173] Step 3:

[0174] Send it to a question-answering model.

[0175] The server sends the relevant meeting notes to the question-answering model.

[0176] Specific behavior: Compile relevant meeting notes in JSON format and send them to the LLM with a request to "provide relevant information to answer this question."

[0177] Input: User questions and related meeting notes

[0178] Output: Data requests sent to the question-answering model

[0179] Step 4:

[0180] Generate an answer.

[0181] The LLM model analyzes the question and related records to generate the best possible answer.

[0182] What it does: LLM uses natural language processing techniques to generate accurate answers to questions.

[0183] Input: Questions and meeting notes

[0184] Output: The generated answer

[0185] Step 5:

[0186] The answer is provided to the user terminal.

[0187] The server receives the generated response from the LLM and sends it to the user terminal.

[0188] Specific operation: The server converts the answer into JSON format and sends it to the user's device, where the user can check the answer.

[0189] Input: Generated Answer

[0190] Output: The answer sent to the user's terminal

[0191] Prediction function for next discussion topic

[0192] Step 1:

[0193] Obtain past meeting notes and summaries.

[0194] The server retrieves past meeting records and summaries from a database.

[0195] Specific operation: The server extracts "all past meeting records and summaries" using an SQL query. It executes a join query in the form of "SELECT FROM meeting record table JOIN summary table ON meeting ID = summary meeting ID".

[0196] Input: Meeting notes and summaries from the database

[0197] Output: Captured meeting records and summary data

[0198] Step 2:

[0199] Send it to a predictive model.

[0200] The server sends the acquired data to the next discussion content prediction model.

[0201] Specific behavior: The server converts the data into JSON format and sends it to the LLM along with a prompt: "Please predict the next discussion topic."

[0202] Input: Meeting notes and summary data

[0203] Output: Data requests sent to the predictive model

[0204] Step 3:

[0205] Generate predictions.

[0206] The LLM model analyzes past data and predicts the content of the next discussion.

[0207] How it works: LLM analyzes past meeting notes, finds patterns and trends, and predicts what will be discussed.

[0208] Input: Past meeting notes and summaries

[0209] Output: The generated prediction

[0210] Step 4:

[0211] Save the prediction.

[0212] The server stores the generated predictions in a database.

[0213] Specific operation: The prediction content is saved in the "prediction result table" of the database using an INSERT query. An SQL query such as "INSERT INTO prediction result table (conference ID, prediction content) VALUES (..., ...)" is executed.

[0214] Input: Generated predictions

[0215] Output: Predictions stored in the database

[0216] Step 5:

[0217] The predicted content is provided to the user terminal.

[0218] When a user prepares for the next meeting, the user requests the prediction results from the server, and the server provides the prediction results to the user's terminal.

[0219] Specific operation: The user types "Request prediction results for the next discussion content" on their device and sends the request to the server. The server retrieves the target prediction results from the database and sends them to the user's device in JSON format.

[0220] Input: A request from the user

[0221] Output: Prediction sent to the user's device

[0222] (Application example 1)

[0223] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0224] With conventional conference systems, participants had to spend a lot of time understanding the content of the conference, making it difficult to conduct the conference efficiently. Furthermore, when dealing with customers in virtual stores, it was difficult to provide appropriate responses based on past customer questions and purchase history. This resulted in a decline in customer satisfaction and a decline in business efficiency.

[0225] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0226] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to the summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for acquiring past inquiries and generating and providing summaries using the summary generation model, means for generating answers to specific questions, and means for generating questions predicted for the user's next visit. This allows conference participants to efficiently prepare for the conference, while at the same time enabling customers to be served quickly and appropriately in the virtual store, thereby improving customer satisfaction.

[0227] "Past meeting records" are data that record the contents of meetings that have taken place in the past.

[0228] "Means for acquiring" refers to the method or device for inputting the necessary information into the system.

[0229] A "summary generation model" is an algorithm that extracts important information from vast amounts of text data and summarizes it concisely.

[0230] A "transmission medium" is a method or tool for transferring data from one place to another.

[0231] "Storage means" refers to a method or device for long-term storage of acquired or generated data.

[0232] "User terminal" means a device such as a computer, smartphone, or tablet used by an end user of the system.

[0233] "Means of providing" refers to the method or device for providing the necessary information to the user.

[0234] "Past inquiries" are records of questions or requests that customers or users have previously made to the system.

[0235] A "specific question" is a single, clearly defined question.

[0236] "Answer generation means" refers to methods or tools that create appropriate answers to input questions.

[0237] "Predicted questions" are questions that customers and users are likely to ask in the future, predicted based on past data and trends.

[0238] The present invention provides a system for efficiently handling customers in meetings and virtual stores. The system includes a server, a user terminal, and a large-scale language model (LLM).

[0239] Configuration and operation of the conference system

[0240] The server retrieves past meeting records from a company's internal database, etc. These records include text data or audio data converted into text. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting. This allows users to quickly understand the content of past meetings and efficiently prepare for meetings.

[0241] Configuration and operation of customer support system

[0242] In the virtual store, the server acquires records of past inquiries. These records are also processed as text data. The server then sends the acquired inquiry records to the LLM summary generation model, which summarizes and saves important information. The user (staff member) can then check this summary information on their device before serving the next customer.

[0243] When a user inputs a specific question, the server sends the question along with any related meeting or inquiry records to the LLM's question-answering model. The model generates the optimal answer to the question, and the server returns the generated answer to the user's device, allowing the user to quickly and appropriately respond to customer questions.

[0244] The server also obtains data based on past meeting and inquiry records to predict the next discussion topic and customer questions. This data is sent to the LLM's prediction model, which predicts the next discussion topic and customer questions. The predicted content is saved by the server and provided to the user's device before the next meeting or customer interaction. This allows users to understand the next discussion topic or customer questions in advance, enabling them to respond more effectively.

[0245] Hardware and software used

[0246] OpenAI API: Provides large-scale language models (LLMs) for natural language processing.

[0247] SQLite: Used as a database management system.

[0248] Python: A programming language for writing invention programs.

[0249] Examples and prompts

[0250] 1. Summary Generation Example:

[0251] Generate a summary of past inquiries from customer ID 1, such as "Please tell me the stock status of sale items" and "Please tell me how to return items."

[0252] Prompt statement:

[0253] Please summarize your inquiry below:

[0254] Please let me know the availability of sale items and how to return them.

[0255] 2. Question and Answer Example:

[0256] Customers ask, "When was your last sale period?"

[0257] Prompt statement:

[0258] Past inquiries:

[0259] Please let me know the availability of sale items and how to return them.

[0260] Customer Question: When was your last sale period?

[0261] Generate the answer

[0262] 3. Examples of questions to predict the next session:

[0263] Next predicted questions based on past inquiries from customer ID 1: "Please tell me the stock status of sale items" and "Please tell me how to return items."

[0264] Prompt statement:

[0265] Generate anticipated questions for your next visit based on the following inquiries:

[0266] Please let me know the availability of sale items and how to return them.

[0267] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0268] Step 1:

[0269] The server retrieves past meeting records or inquiry records from the company's internal database. At this time, the retrieved data is text or speech-to-text conversion. The input is the company's internal database, and the output is the text-converted meeting records or inquiry records.

[0270] Step 2:

[0271] The server sends the acquired meeting records or inquiry records to the summary generation model. At this time, it creates and sends a prompt to the summary generation model, instructing it to summarize the data. The input is the text record, and the output is the summarized text data. As a concrete example, the prompt used is "Please summarize the following inquiry: Please tell me the stock status of sale items. Please tell me how to return them."

[0272] Step 3:

[0273] The server saves the summaries received from the summary generation model and provides them to the user terminal before the next meeting or customer service. At this time, the saved summaries are stored in a database and provided upon user request. The input is the summarized text data, and the output is the summary data provided to the user terminal.

[0274] Step 4:

[0275] When a user inputs a specific question, the terminal sends this question to the server. The server retrieves the meeting records and inquiry records related to the question and sends them to the question-answering model. At this time, a prompt sentence is used to ask the question-answering model for an answer to the question. The input is the user's question and the related records, and the output is the generated answer. As a concrete example, the prompt used is "Past inquiries: What is the stock status of sale items? How do I return them? Customer question: When was the last sale period? Please generate an answer."

[0276] Step 5:

[0277] The server sends the answers received from the question-answering model back to the user device, allowing the user to quickly and appropriately respond to customer questions. The input is the generated answer, and the output is the answer displayed on the user device.

[0278] Step 6:

[0279] The server obtains data to predict the next discussion topic and customer questions based on past meeting records and inquiry records, and sends this data to the prediction model. At this time, the prediction model uses prompt text to make a prediction for the next session. The input is the past records, and the output is the predicted discussion topic and questions. As a specific example, the prompt used is "Based on the following inquiry content, please generate predicted questions for the next visit: Please tell me the stock status of sale items. Please tell me how to return them."

[0280] Step 7:

[0281] The server stores the predicted content and provides it to the user's device before the next meeting or customer interaction. The input is the predicted content, and the output is the predicted data provided to the user's device. This allows the user to understand the next discussion topic or customer questions in advance, allowing them to respond more effectively.

[0282] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0283] This paper describes a system for smoothly progressing a meeting by bridging differences in understanding between participants due to their timing and background. This system includes a server, user terminals, a large-scale language model (LLM), and an emotion engine.

[0284] Summary generation function

[0285] System configuration and operation

[0286] The server retrieves past meeting records from a company's internal database, etc. The meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting.

[0287] Specific examples

[0288] Before a user joins the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[0289] Question and Answer Function

[0290] System configuration and operation

[0291] A user uses a device to enter a specific question and send it to the server, which then passes the question through the LLM's question-answering model to generate an appropriate answer, which is then stored by the server and sent back to the user's device.

[0292] Specific examples

[0293] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[0294] Prediction function for next discussion topic

[0295] System configuration and operation

[0296] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics likely to be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[0297] Specific examples

[0298] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[0299] Adding an Emotion Engine

[0300] System configuration and operation

[0301] The system also incorporates an emotion engine that recognizes the user's emotions.

[0302] Using an Emotion Engine in Summarization

[0303] The emotion engine recognizes the user's current emotional state and incorporates this information into the summary generation. For example, if the user is feeling stressed, the summary provided will be more concise and to the point.

[0304] Specific examples

[0305] Before the user joins the next meeting, the device will recognize the user's emotions using an emotion engine, and if the user is feeling stressed, the server will provide a more concise and easy-to-understand summary, allowing the user to understand the information more efficiently.

[0306] Using Emotion Engines in Question Answering

[0307] The emotion engine takes into account the user's emotional state and provides answers with an appropriate tone and content. For example, if the user is confused, the answers will be more detailed and clearer.

[0308] Specific examples

[0309] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device. The user can then review this answer to gain a deeper understanding of the information they need.

[0310] Using an emotion engine to predict the next discussion topic

[0311] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects them in predicting the content of the next discussion, thereby taking participants' emotional reactions into account in the predictions.

[0312] Specific examples

[0313] When predicting what will be discussed in the next meeting, the emotion engine records the emotions of participants in past meetings to identify topics that are likely to lead to smooth discussion and those that are likely to cause confusion or conflict. This allows the server to provide a more prepared prediction to the user's device, allowing the user to make adjustments and take measures in advance.

[0314] As described above, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[0315] The processing flow will be explained below.

[0316] Summary generation function

[0317] Processing flow

[0318] Step 1:

[0319] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[0320] Step 2:

[0321] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[0322] Step 3:

[0323] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[0324] Step 4:

[0325] The server receives the generated summaries from the LLM and stores them in an internal database.

[0326] Step 5:

[0327] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[0328] Step 6:

[0329] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[0330] Question and Answer Function

[0331] Processing flow

[0332] Step 1:

[0333] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[0334] Step 2:

[0335] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[0336] Step 3:

[0337] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[0338] Step 4:

[0339] LLM analyzes the question and related records to generate the best possible answer.

[0340] Step 5:

[0341] The server receives the generated response from the LLM and returns it to the user terminal.

[0342] Step 6:

[0343] The user checks the answers on the terminal and obtains the necessary information.

[0344] Prediction function for next discussion topic

[0345] Processing flow

[0346] Step 1:

[0347] The server retrieves past meeting records and summaries from an internal database.

[0348] Step 2:

[0349] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[0350] Step 3:

[0351] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[0352] Step 4:

[0353] The server receives predicted discussion content from the LLM and stores it in an internal database.

[0354] Step 5:

[0355] When a user prepares for their next meeting, they request the prediction results from their device.

[0356] Step 6:

[0357] The server transmits the saved prediction results to the terminal.

[0358] Step 7:

[0359] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[0360] Adding an Emotion Engine

[0361] Processing flow

[0362] Step 1:

[0363] When a user accesses the terminal and uses the meeting summary or question answering function, the emotion recognition sensor acquires the user's emotion data.

[0364] Step 2:

[0365] The terminal transmits the user's emotional data to the emotion engine, which analyzes the user's current emotional state.

[0366] Step 3:

[0367] The emotion engine sends the results of its analysis to a server, which uses this data to adjust the output of a summary generation model or question-answering model.

[0368] Using an Emotion Engine in Summarization

[0369] Step 4:

[0370] The server receives data from the emotion engine and optimizes the summary content based on the user's emotions. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the main points.

[0371] Using Emotion Engines in Question Answering

[0372] Step 5:

[0373] When the emotion engine recognizes a user's confusion or doubt, the server fine-tunes the output of the question-answering model: for example, if the user is confused, the answers provided will be presented in a more understandable format.

[0374] Using an emotion engine to predict the next discussion topic

[0375] Step 6:

[0376] The server uses data from the emotion engine to predict upcoming discussions based on emotional states recorded in past meeting recordings. This data helps identify topics that will flow smoothly and those that are likely to cause confusion or conflict.

[0377] Step 7:

[0378] When a user prepares for the next meeting, the device receives prediction results along with emotion-linked data, allowing the user to prepare for the meeting efficiently.

[0379] By incorporating an emotion engine, the system can take the user's emotional state into account and provide summaries, question and answer sessions, and predictions of upcoming discussion topics in a more user-friendly format, thereby helping to ensure that meetings proceed effectively.

[0380] Example 2

[0381] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0382] In meetings, differences in understanding due to participants' timing and backgrounds can often be a problem, preventing the meeting from proceeding smoothly. Furthermore, understanding the content of a meeting requires a lot of time and effort, which can make pre-meeting preparations and question-and-answer sessions during the meeting inefficient. It is also difficult to predict the next discussion topic based on meeting records, and providing information without considering the emotions of participants can reduce meeting productivity. A system that can solve these issues is needed.

[0383] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0384] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, and means for adjusting the summaries to suit the users using an emotion engine that recognizes the emotional state of the users. This allows users to efficiently understand past conference records and provides summaries that suit their emotions, making it possible to smoothly proceed with the conference.

[0385] "Meeting records" are data that store the contents of discussions and decisions made during a meeting in text format.

[0386] A "summary generation model" is an algorithm or machine learning model that extracts important points from long text and creates a short summary.

[0387] A "user terminal" is an electronic device such as a computer or smartphone that is directly operated by a user.

[0388] An "emotion engine" is a technology that recognizes the user's emotional state from their facial expressions and voice, and reflects that information in the system's operation.

[0389] A "question answering model" is an algorithm or machine learning model that generates appropriate answers to specific questions.

[0390] A "discussion content predictive model" is a technology that analyzes past data and patterns to predict the topics and agenda items that are likely to be discussed in the next meeting.

[0391] "Acquisition means" refers to the functions and technologies that the system uses to acquire the necessary data from outside.

[0392] "Storage means" refers to the function or technology for storing acquired or generated data in an internal storage device.

[0393] "Providing means" refers to the functions and technologies used to present stored data in a form that is accessible to users.

[0394] "Adjustment means" refers to functions and technologies that optimize the generated summaries and answers depending on the user's emotional state and other conditions.

[0395] This invention is a system for smoothly progressing a meeting by bridging the gap in understanding caused by the timing and background of each participant. This system incorporates a server, a user terminal, a generative AI model (hereafter referred to as LLM), and an emotion engine.

[0396] The server retrieves past meeting records from a company's internal database, etc. These meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts important points from the meeting records and generates a summary. The generated summary is stored on the server and provided to the user's device before the next meeting.

[0397] Before joining the next meeting, the user requests a summary of the previous meeting from the server from their device. The server then sends the summary generated from the past records to the device, allowing the user to quickly check the important points. An example of a specific prompt sentence is "Please tell me the summary of the previous meeting." This allows the user to efficiently prepare for the meeting.

[0398] Furthermore, users can input specific questions into their devices and send them to the server. The server sends the questions to the LLM's question-answering model, which generates an appropriate answer. This answer is stored on the server and sent back to the user's device. An example of a specific prompt sentence is, "What was the budget decided at the last meeting?" The server extracts the relevant information from past meeting records and sends the appropriate answer to the device. This allows users to quickly obtain the information they need.

[0399] The server also retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is stored on the server and provided to user devices before the next meeting. An example of a specific prompt is, "Please tell me the topics that will be discussed in the next meeting." This allows users to know the next discussion topic in advance and effectively prepare for the meeting.

[0400] The system also incorporates an emotion engine that recognizes the user's emotions. The emotion engine recognizes the user's current emotional state and reflects that information in summary generation, question answering, and predictions of the next discussion topic. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the key points. Also, when a user submits a specific question, if the emotion engine recognizes the user's confusion, the response will include more detailed explanations. Furthermore, the emotion engine can take into account the emotional state of participants during past meetings and reflect that in predictions of the next discussion topic.

[0401] In this way, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[0402] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0403] Step 1:

[0404] Input: A user requests a past meeting summary from a terminal to prepare for an upcoming meeting.

[0405] Specific operation: A user makes an API request on a terminal and sends the prompt "Please tell me summaries of past meetings" to the server.

[0406] Output: The request is sent to the device.

[0407] Step 2:

[0408] Input: The server receives a request from a user.

[0409] Specific operation: The server accesses its internal database and executes the SQL query "SELECT FROM summaries WHERE meeting_id=past meeting ID;" to retrieve stored past meeting summaries.

[0410] Output: Past meeting summary data is obtained.

[0411] Step 3:

[0412] Input: The server sends the past conference summary data it has acquired to the LLM summary generation model.

[0413] Specific operation: The server sends the acquired meeting records in text format to the summary generation model using the POST request "curl -X POST 'https: / / api.llm.com / summarize' -d 'meeting record text'".

[0414] Output: The summary data received from the summary generation model is returned.

[0415] Step 4:

[0416] Input: The server sends the summary data received from the LLM to the emotion engine.

[0417] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the summary data. If the user is feeling stressed, the summary will be adjusted to be more concise.

[0418] Output: Optimized summary data is generated.

[0419] Step 5:

[0420] Input: The server sends the optimized summary data to the user terminal.

[0421] Specific operation: The server generates an API response containing summary data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_summary?meeting_id=past meeting ID'".

[0422] Output: The summary data is displayed on the user's terminal.

[0423] Step 6:

[0424] Input: The user inputs a specific question about the next meeting and sends it from the terminal to the server.

[0425] Specific behavior: A user makes an API request on their device and sends the prompt "What was the budget decided at the last meeting?" to the server.

[0426] Output: The request is sent to the device.

[0427] Step 7:

[0428] Input: The server receives a question from the user.

[0429] Specific operation: The server sends a question to the LLM question-answering model, for example, "curl -X POST 'https: / / api.llm.com / answer' -d 'What was the budget decided at the last meeting?'".

[0430] Output: Answer data is returned from the question-answering model.

[0431] Step 8:

[0432] Input: The server sends the answer data received from the LLM to the emotion engine.

[0433] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the answer data. If the user is confused, the answer will be further tailored.

[0434] Output: Optimized answer data is generated.

[0435] Step 9:

[0436] Input: The server sends the optimized answer data to the user's terminal.

[0437] Specific operation: The server generates an API response containing the answer data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_answer?question_id=question ID'".

[0438] Output: The answer data is displayed on the user's terminal.

[0439] Step 10:

[0440] Input: The server retrieves past meeting records and summary data and sends them to the LLM model to predict the content of the next discussion.

[0441] Specific operation: The server retrieves data using the SQL queries "SELECT FROM meeting_records WHERE date BETWEEN 'past date' AND 'current date';" and "SELECT FROM summaries;", and sends the data to the predictive model using "curl -X POST 'https: / / api.llm.com / predict_topics' -d 'past meeting data and summaries'".

[0442] Output: The predictive model returns the next discussion content data.

[0443] Step 11:

[0444] Input: The server sends the discussion content data received from the predictive model to the emotion engine.

[0445] Specific operation: The server uses the emotion engine to analyze the emotional state of participants during past meetings and reflects this in predicting the content of the next discussion.

[0446] Output: Optimized discussion content data is generated.

[0447] Step 12:

[0448] Input: The server sends the optimized discussion content data to the user terminal.

[0449] Specific operation: The server generates an API response containing the discussion content data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_predicted_topics?meeting_id=next meeting ID'".

[0450] Output: Discussion content data is displayed on the user's terminal.

[0451] (Application example 2)

[0452] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0453] In today's business environment, improving the efficiency of meetings is extremely important. However, complex meeting content and the emotional state of participants can make it difficult to understand the content, leading to differences in understanding among participants. Furthermore, if participants are unable to check past meeting records or predict the topics that will be discussed in the next meeting, they may not be able to prepare optimally, which can cause the meeting to stall. A system that solves these problems is needed.

[0454] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for acquiring past records, means for transmitting the acquired records to the generative model, means for saving summary information received from the generative model, means for providing the saved summary information to a user terminal, means for recognizing an emotional state, and means for adjusting the summary information based on the recognized emotional state. This enables users to efficiently understand the contents of past meetings and receive information tailored to their own emotional state.

[0455] "Records" refers to information or data about a meeting or event, stored in the form of documents, audio, video, etc.

[0456] A "generative model" refers to an algorithm or system that uses large-scale language models and artificial intelligence to generate summaries or answers from input data.

[0457] "Summary information" is data that concisely summarizes the important points extracted by the generative model.

[0458] "User terminal" refers to an information processing device used by a user, such as a smartphone, tablet, or PC.

[0459] "Emotional state" refers to the user's current psychological state, including psychological states such as stress, confusion, fatigue, etc.

[0460] "Adjusting" means changing the format of summary information and answers depending on the perceived emotional state, and adapting to make them easier for the user to understand.

[0461] "Forecasting" refers to estimating future events or discussions based on past data and trends. It also refers to the results.

[0462] "Saving" means recording the generated summary information and answers in a database or the like, making them available for future reference or use.

[0463] The present invention relates to a system for improving the efficiency of operation management and maintenance meetings for autonomous vehicles. An embodiment of this system will be described in detail below.

[0464] 1. System Configuration

[0465] This system mainly consists of a server, user devices, a large-scale language model (LLM), and an emotion engine. The server stores meeting records and generated summary information and provides them to user devices. User devices include smartphones, tablets, and PCs, through which users obtain information.

[0466] 2. Summary generation function

[0467] The server retrieves past meeting records and sends them to the LLM summary generation model. The model extracts key points from the input information and generates a summary. This summary is stored on the server and provided to users upon request.

[0468] Examples:

[0469] Before a user joins the next meeting, the user requests a summary of the past meeting from the server, and the server sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points.

[0470] 3. Question and Answer Function

[0471] Users can send specific questions from their devices to the server, which then passes the questions through the LLM's question-answering model to generate appropriate answers, which are then stored on the server and sent back to the user's device.

[0472] Examples:

[0473] The user sends a question from the terminal to the server, such as "What was the maintenance schedule decided at the last meeting?" The server uses LLM to extract the relevant information from the past meeting records, generates an appropriate answer, and sends it to the terminal.

[0474] 4. Prediction of next discussion topic

[0475] The server sends the past meeting records and generated summary information to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is saved by the server and provided to user devices before the next meeting.

[0476] Examples:

[0477] When a user prepares for the next meeting, the user requests a prediction result for the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance.

[0478] 5. Use of Emotion Engine

[0479] The system also incorporates an emotion engine that recognizes the user's current emotional state and reflects that information in summary generation, question answering, and prediction of the next discussion topic.

[0480] Use in summary generation:

[0481] The emotion engine provides a more concise summary if the user is feeling stressed.

[0482] Examples:

[0483] Before the user joins the next meeting, the terminal recognizes the user's emotions using an emotion engine, and if the user is feeling stressed, the server adjusts the summary to make it more concise and easy to understand.

[0484] Use in question answering:

[0485] The sentiment engine makes the answers more detailed and clear if the user is confused.

[0486] Examples:

[0487] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device.

[0488] Use in predicting the next discussion topic:

[0489] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects this in predicting the content of the next discussion.

[0490] Examples:

[0491] When predicting what will be discussed in the next meeting, the emotion engine captures participants' emotions during past meetings to identify topics that will be easy to discuss and those that are likely to cause confusion or conflict.

[0492] 6. Examples of prompt sentences

[0493] For example, the prompt for generating a summary using LLM is as follows:

[0494] Summary: Minutes from the last traffic management meeting...

[0495] An example of a question-and-answer prompt is as follows:

[0496] Question: What was the maintenance schedule decided at the last traffic management meeting? Meeting notes: ...

[0497] As described above, this system uses an emotion engine and LLM to provide information according to the user's emotional state, significantly improving the efficiency of meetings.

[0498] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0499] Step 1:

[0500] (User submits request)

[0501] Before participating in a conference, a user sends a request using a terminal to obtain a summary of past conference records, and this request is sent from the user's terminal to the server.

[0502] Input: User request information (e.g., summary of past meeting records)

[0503] Output: User request data

[0504] What it does: Use a smartphone or tablet to enter and submit a specific request.

[0505] Step 2:

[0506] (Server retrieves the meeting recording)

[0507] The server retrieves past meeting records from the company's internal database.

[0508] Input: User request data

[0509] Output: Past meeting recording data

[0510] Specific operation: The server accesses the database and extracts the records of the specified meeting.

[0511] Step 3:

[0512] (The server sends the meeting recording to the generative model)

[0513] The server sends the acquired meeting records to the LLM summary generation model.

[0514] Input: Past meeting recording data

[0515] Output: Input data to the summary generation model

[0516] Specific operation: The server converts the meeting recording into text format and sends it to the generative model API.

[0517] Step 4:

[0518] (The generative model generates the summary)

[0519] LLM's summary generation model generates summaries from submitted meeting records.

[0520] Input: Input data to the summary generation model

[0521] Output: Generated summary data

[0522] What it does: LLM analyzes the input text and extracts and summarizes the key points.

[0523] Step 5:

[0524] (Server stores and serves summaries)

[0525] The generated summary is stored on a server and provided to the user terminal.

[0526] Input: Generated summary data

[0527] Output: Summary data to user terminal

[0528] Specific operation: The server saves the summary data in a database and sends it to the user's device.

[0529] Step 6:

[0530] (User checks summary)

[0531] The user checks the summary provided on the terminal and prepares for the next meeting based on this summary.

[0532] Input: Summary data to user terminal

[0533] Output: User's understanding after reviewing the summary

[0534] What it does: Read summaries on your smartphone or tablet screen and prepare for your next meeting.

[0535] Step 7:

[0536] (user types question)

[0537] A user sends a specific question about past meeting records from a terminal to the server.

[0538] Input: User question data

[0539] Output: Submitted question data

[0540] Specific behavior: The user enters a specific question in text format and sends it to the server.

[0541] Step 8:

[0542] (The server sends the question to the generative model)

[0543] The server sends the question to the LLM's question-answering model, which generates an appropriate answer.

[0544] Input: User question data

[0545] Output: Input data to the question-answering model

[0546] Specific operation: The server converts the question into text format and sends it to the generative model API.

[0547] Step 9:

[0548] (The generative model generates the answer)

[0549] LLM's question-answering model extracts relevant information from past meeting records and generates appropriate answers.

[0550] Input: Input data to the question-answering model

[0551] Output: Generated response data

[0552] What it does: LLM analyzes the input text and generates an answer.

[0553] Step 10:

[0554] (The server stores and serves the answer)

[0555] The generated answers are stored on the server and provided to the user terminal.

[0556] Input: Generated response data

[0557] Output: Answer data to the user's terminal

[0558] Specific operation: The server saves the answer data in a database and sends it to the user's device.

[0559] Step 11:

[0560] (User checks answer)

[0561] The user can check the answers provided on the terminal, obtain the necessary information, and use it to prepare for the next meeting.

[0562] Input: Answer data to user terminal

[0563] Output: User understanding of the answer

[0564] What it does: Read the answers on your smartphone or tablet screen to help prepare for your next meeting.

[0565] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0566] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0567] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0568] [Second embodiment]

[0569] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0570] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0571] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0572] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0573] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0574] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0575] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0576] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0577] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0578] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0579] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0580] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal."

[0581] This paper describes a system for smoothly progressing a conference by bridging differences in understanding between participants due to their timing and background. The system includes a server, user terminals, and a large-scale language model (LLM).

[0582] Summary generation function

[0583] System configuration and operation

[0584] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted to text. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts key points from the meeting records and generates a summary. The generated summary is stored by the server and provided to the user's device before the next meeting.

[0585] Specific examples

[0586] Before joining the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[0587] Question and Answer Function

[0588] System configuration and operation

[0589] A user uses a terminal to enter a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[0590] Specific examples

[0591] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[0592] Prediction function for next discussion topic

[0593] System configuration and operation

[0594] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the likely discussions and topics that will be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[0595] Specific examples

[0596] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[0597] As described above, this system uses past meeting records to summarize, answer questions, and predict the content of the next discussion, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[0598] The processing flow will be explained below.

[0599] Summary generation function

[0600] Processing flow

[0601] Step 1:

[0602] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[0603] Step 2:

[0604] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[0605] Step 3:

[0606] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[0607] Step 4:

[0608] The server receives the generated summaries from the LLM and stores them in an internal database.

[0609] Step 5:

[0610] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[0611] Step 6:

[0612] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[0613] Question and Answer Function

[0614] Processing flow

[0615] Step 1:

[0616] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[0617] Step 2:

[0618] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[0619] Step 3:

[0620] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[0621] Step 4:

[0622] LLM analyzes the question and relevant meeting notes to generate the best possible answer.

[0623] Step 5:

[0624] The server receives the generated response from the LLM and returns it to the user terminal.

[0625] Step 6:

[0626] The user checks the answers on the terminal and obtains the necessary information.

[0627] Prediction function for next discussion topic

[0628] Processing flow

[0629] Step 1:

[0630] The server retrieves past meeting records and summaries from an internal database.

[0631] Step 2:

[0632] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[0633] Step 3:

[0634] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[0635] Step 4:

[0636] The server receives predicted discussion content from the LLM and stores it in an internal database.

[0637] Step 5:

[0638] When a user prepares for their next meeting, they request the prediction results from their device.

[0639] Step 6:

[0640] The server transmits the saved prediction results to the terminal.

[0641] Step 7:

[0642] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[0643] Example 1

[0644] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0645] Modern companies hold many meetings, but participants' understanding of the content often varies depending on their timing and background. It's also difficult for new participants to quickly understand the content of past meetings, which can slow down the progress of the meeting. Furthermore, searching for necessary information from past meeting records and predicting the content of the next meeting takes time and effort. A system that solves these problems and supports productive meetings is needed.

[0646] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0647] In this invention, the server includes means for acquiring past meeting records, means for transmitting the acquired meeting records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for transmitting to a question-answering model that generates answers to specific questions, means for providing to the user terminal the answers received from the question-answering model, means for transmitting to a prediction model that predicts the content of the next discussion based on the past meeting records and summaries, and means for providing to the user terminal the discussion content received from the prediction model. This enables users to quickly understand the content of past meetings, easily obtain necessary information, and efficiently prepare for the next meeting.

[0648] "Past meeting records" refers to text data or audio data recorded in a previous meeting.

[0649] A "summary generation model" refers to a large-scale language model that can extract important points from long texts and summarize them concisely.

[0650] A "question answering model" refers to a large-scale language model that analyzes relevant information based on a given question and generates an appropriate answer.

[0651] A "predictive model" refers to a large-scale language model that has the ability to predict future events or outcomes based on past data.

[0652] A "database" refers to a system for efficiently storing, managing, and searching data in various formats.

[0653] A "server" refers to a computer system that receives requests from user terminals, performs the necessary processing, and provides the data.

[0654] "User terminal" refers to a device used by a user, such as a computer, smartphone, or tablet.

[0655] A "generative AI model" refers to an artificial intelligence model that uses a large-scale language model to generate output that is tailored to a specific purpose.

[0656] A "prompt sentence" refers to input text that provides specific instructions to a generative AI model.

[0657] The present invention relates to a system for smoothly progressing a meeting by bridging differences in understanding due to the timing and background of each participant. The system includes a server, a user terminal, and a large-scale language model (LLM).

[0658] Summary generation function

[0659] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted into text. The retrieved meeting record data is sent to an LLM (e.g., GPT-4) specialized in generating text summaries. The LLM extracts key points from the provided meeting records and generates a summary, which is then stored by the server. Before the next meeting, the user can request a past meeting summary from the server, and the summary is provided to the user's device.

[0660] Specific examples

[0661] Before joining the next meeting, a user can request a summary of the previous meeting from the server. The server then sends the summary to the terminal, allowing the user to quickly review the important points and efficiently prepare for the meeting.

[0662] Example prompt statement:

[0663] Summarize past meeting notes, including key points and organizing them so users can review them before the next meeting.

[0664] Question and Answer Function

[0665] A user uses a terminal to input a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[0666] Specific examples

[0667] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[0668] Example prompt statement:

[0669] Please tell me about the budget that was decided at the last meeting.

[0670] Prediction function for next discussion topic

[0671] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model predicts the discussions and topics likely to be discussed in the next meeting based on past data. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[0672] Specific examples

[0673] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[0674] Example prompt statement:

[0675] Anticipate topics likely to be discussed at the next meeting. Use past meeting notes and summaries to inform future discussions.

[0676] This system allows users to quickly understand the content of past meetings, easily obtain the necessary information, and efficiently prepare for the next meeting, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[0677] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0678] Summary generation function

[0679] Step 1:

[0680] Get past meeting records.

[0681] The server accesses the company's internal database and retrieves past meeting records using SQL queries, etc. The retrieved input data is text data or audio data converted into text.

[0682] What happens: The server retrieves data from the database using an API call, executing an SQL query like "SELECT FROM meeting_records_table WHERE date > '2022-01-01'".

[0683] Input: Meeting recordings from the company's internal database (text format or speech-to-text transcription)

[0684] Output: Past meeting recording data

[0685] Step 2:

[0686] Send the meeting notes to a summary generation model.

[0687] The server sends the acquired meeting notes to an LLM model (e.g., GPT-4) specialized in generating text summaries.

[0688] Specific behavior: Convert the meeting recording to JSON format and send the data to the model via API. Include the prompt "Please summarize this meeting recording."

[0689] Input: Past meeting recording data

[0690] Output: Data request sent to the summary generation model

[0691] Step 3:

[0692] Generate a summary.

[0693] The LLM model extracts key points from the provided meeting notes and generates a summary.

[0694] How it works: Large-scale language models such as GPT-4 analyze meeting recordings and use natural language processing (NLP) techniques to extract key parts.

[0695] Input: Meeting recording data

[0696] Output: Generated summary

[0697] Step 4:

[0698] Save the summary.

[0699] The server receives the generated summary and stores it in a database.

[0700] Specific operation: Convert the summary data into JSON format and insert it into the "Summary" table in the database. Execute an SQL query such as "INSERT INTO SummaryTable(ConferenceID, Summary) VALUES(..., ...)".

[0701] Input: Generated summary

[0702] Output: Summary stored in database

[0703] Step 5:

[0704] Provided to the user terminal.

[0705] Before the next meeting, the user requests a summary of the past meeting from the terminal, and the server provides the summary to the user terminal.

[0706] Specific operation: The user enters "Request past conference summaries" on the terminal and clicks the send button. The server retrieves the relevant summaries from the database and returns them to the user terminal in JSON format.

[0707] Input: A request from the user

[0708] Output: Summary sent to user terminal

[0709] Question and Answer Function

[0710] Step 1:

[0711] Enter and submit a specific question.

[0712] The user uses a terminal to enter a specific question and send it to the server.

[0713] Specific operation: The user types "What was the budget decided at the last meeting?" into the terminal and clicks the send button.

[0714] Input: User question

[0715] Output: The question data sent to the server

[0716] Step 2:

[0717] Receive questions.

[0718] The server receives a query from a user and prepares to search for relevant meeting records.

[0719] Specific operation: The server analyzes the question and retrieves the relevant meeting records from the database.

[0720] Input: Question data from the user

[0721] Output: Data search plan within the server

[0722] Step 3:

[0723] Send it to a question-answering model.

[0724] The server sends the relevant meeting notes to the question-answering model.

[0725] Specific behavior: Compile relevant meeting notes in JSON format and send them to the LLM with a request to "provide relevant information to answer this question."

[0726] Input: User questions and related meeting notes

[0727] Output: Data requests sent to the question-answering model

[0728] Step 4:

[0729] Generate an answer.

[0730] The LLM model analyzes the question and related records to generate the best possible answer.

[0731] What it does: LLM uses natural language processing techniques to generate accurate answers to questions.

[0732] Input: Questions and meeting notes

[0733] Output: The generated answer

[0734] Step 5:

[0735] The answer is provided to the user terminal.

[0736] The server receives the generated response from the LLM and sends it to the user terminal.

[0737] Specific operation: The server converts the answer into JSON format and sends it to the user's device, where the user can check the answer.

[0738] Input: Generated Answer

[0739] Output: The answer sent to the user's terminal

[0740] Prediction function for next discussion topic

[0741] Step 1:

[0742] Obtain past meeting notes and summaries.

[0743] The server retrieves past meeting records and summaries from a database.

[0744] Specific operation: The server extracts "all past meeting records and summaries" using an SQL query. It executes a join query in the form of "SELECT FROM meeting record table JOIN summary table ON meeting ID = summary meeting ID".

[0745] Input: Meeting notes and summaries from the database

[0746] Output: Captured meeting records and summary data

[0747] Step 2:

[0748] Send it to a predictive model.

[0749] The server sends the acquired data to the next discussion content prediction model.

[0750] Specific behavior: The server converts the data into JSON format and sends it to the LLM along with a prompt: "Please predict the next discussion topic."

[0751] Input: Meeting notes and summary data

[0752] Output: Data requests sent to the predictive model

[0753] Step 3:

[0754] Generate predictions.

[0755] The LLM model analyzes past data and predicts the content of the next discussion.

[0756] How it works: LLM analyzes past meeting notes, finds patterns and trends, and predicts what will be discussed.

[0757] Input: Past meeting notes and summaries

[0758] Output: The generated prediction

[0759] Step 4:

[0760] Save the prediction.

[0761] The server stores the generated predictions in a database.

[0762] Specific operation: The prediction content is saved in the "prediction result table" of the database using an INSERT query. An SQL query such as "INSERT INTO prediction result table (conference ID, prediction content) VALUES (..., ...)" is executed.

[0763] Input: Generated predictions

[0764] Output: Predictions stored in the database

[0765] Step 5:

[0766] The predicted content is provided to the user terminal.

[0767] When a user prepares for the next meeting, the user requests the prediction results from the server, and the server provides the prediction results to the user's terminal.

[0768] Specific operation: The user types "Request prediction results for the next discussion content" on their device and sends the request to the server. The server retrieves the target prediction results from the database and sends them to the user's device in JSON format.

[0769] Input: A request from the user

[0770] Output: Prediction sent to the user's device

[0771] (Application example 1)

[0772] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0773] With conventional conference systems, participants had to spend a lot of time understanding the content of the conference, making it difficult to conduct the conference efficiently. Furthermore, when dealing with customers in virtual stores, it was difficult to provide appropriate responses based on past customer questions and purchase history. This resulted in a decline in customer satisfaction and a decline in business efficiency.

[0774] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0775] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to the summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for acquiring past inquiries and generating and providing summaries using the summary generation model, means for generating answers to specific questions, and means for generating questions predicted for the user's next visit. This allows conference participants to efficiently prepare for the conference, while at the same time enabling customers to be served quickly and appropriately in the virtual store, thereby improving customer satisfaction.

[0776] "Past meeting records" are data that record the contents of meetings that have taken place in the past.

[0777] "Means for acquiring" refers to the method or device for inputting the necessary information into the system.

[0778] A "summary generation model" is an algorithm that extracts important information from vast amounts of text data and summarizes it concisely.

[0779] A "transmission medium" is a method or tool for transferring data from one place to another.

[0780] "Storage means" refers to a method or device for long-term storage of acquired or generated data.

[0781] "User terminal" means a device such as a computer, smartphone, or tablet used by an end user of the system.

[0782] "Means of providing" refers to the method or device for providing the necessary information to the user.

[0783] "Past inquiries" are records of questions or requests that customers or users have previously made to the system.

[0784] A "specific question" is a single, clearly defined question.

[0785] "Answer generation means" refers to methods or tools that create appropriate answers to input questions.

[0786] "Predicted questions" are questions that customers and users are likely to ask in the future, predicted based on past data and trends.

[0787] The present invention provides a system for efficiently handling customers in meetings and virtual stores. The system includes a server, a user terminal, and a large-scale language model (LLM).

[0788] Configuration and operation of the conference system

[0789] The server retrieves past meeting records from a company's internal database, etc. These records include text data or audio data converted into text. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting. This allows users to quickly understand the content of past meetings and efficiently prepare for meetings.

[0790] Configuration and operation of customer support system

[0791] In the virtual store, the server acquires records of past inquiries. These records are also processed as text data. The server then sends the acquired inquiry records to the LLM summary generation model, which summarizes and saves important information. The user (staff member) can then check this summary information on their device before serving the next customer.

[0792] When a user inputs a specific question, the server sends the question along with any related meeting or inquiry records to the LLM's question-answering model. The model generates the optimal answer to the question, and the server returns the generated answer to the user's device, allowing the user to quickly and appropriately respond to customer questions.

[0793] The server also obtains data based on past meeting and inquiry records to predict the next discussion topic and customer questions. This data is sent to the LLM's prediction model, which predicts the next discussion topic and customer questions. The predicted content is saved by the server and provided to the user's device before the next meeting or customer interaction. This allows users to understand the next discussion topic or customer questions in advance, enabling them to respond more effectively.

[0794] Hardware and software used

[0795] OpenAI API: Provides large-scale language models (LLMs) for natural language processing.

[0796] SQLite: Used as a database management system.

[0797] Python: A programming language for writing invention programs.

[0798] Examples and prompts

[0799] 1. Summary Generation Example:

[0800] Generate a summary of past inquiries from customer ID 1, such as "Please tell me the stock status of sale items" and "Please tell me how to return items."

[0801] Prompt statement:

[0802] Please summarize your inquiry below:

[0803] Please let me know the availability of sale items and how to return them.

[0804] 2. Question and Answer Example:

[0805] Customers ask, "When was your last sale period?"

[0806] Prompt statement:

[0807] Past inquiries:

[0808] Please let me know the availability of sale items and how to return them.

[0809] Customer Question: When was your last sale period?

[0810] Generate the answer

[0811] 3. Examples of questions to predict the next session:

[0812] Next predicted questions based on past inquiries from customer ID 1: "Please tell me the stock status of sale items" and "Please tell me how to return items."

[0813] Prompt statement:

[0814] Generate anticipated questions for your next visit based on the following inquiries:

[0815] Please let me know the availability of sale items and how to return them.

[0816] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0817] Step 1:

[0818] The server retrieves past meeting records or inquiry records from the company's internal database. At this time, the retrieved data is text or speech-to-text conversion. The input is the company's internal database, and the output is the text-converted meeting records or inquiry records.

[0819] Step 2:

[0820] The server sends the acquired meeting records or inquiry records to the summary generation model. At this time, it creates and sends a prompt to the summary generation model, instructing it to summarize the data. The input is the text record, and the output is the summarized text data. As a concrete example, the prompt used is "Please summarize the following inquiry: Please tell me the stock status of sale items. Please tell me how to return them."

[0821] Step 3:

[0822] The server saves the summaries received from the summary generation model and provides them to the user terminal before the next meeting or customer service. At this time, the saved summaries are stored in a database and provided upon user request. The input is the summarized text data, and the output is the summary data provided to the user terminal.

[0823] Step 4:

[0824] When a user inputs a specific question, the terminal sends this question to the server. The server retrieves the meeting records and inquiry records related to the question and sends them to the question-answering model. At this time, a prompt sentence is used to ask the question-answering model for an answer to the question. The input is the user's question and the related records, and the output is the generated answer. As a concrete example, the prompt used is "Past inquiries: What is the stock status of sale items? How do I return them? Customer question: When was the last sale period? Please generate an answer."

[0825] Step 5:

[0826] The server sends the answers received from the question-answering model back to the user device, allowing the user to quickly and appropriately respond to customer questions. The input is the generated answer, and the output is the answer displayed on the user device.

[0827] Step 6:

[0828] The server obtains data to predict the next discussion topic and customer questions based on past meeting records and inquiry records, and sends this data to the prediction model. At this time, the prediction model uses prompt text to make a prediction for the next session. The input is the past records, and the output is the predicted discussion topic and questions. As a specific example, the prompt used is "Based on the following inquiry content, please generate predicted questions for the next visit: Please tell me the stock status of sale items. Please tell me how to return them."

[0829] Step 7:

[0830] The server stores the predicted content and provides it to the user's device before the next meeting or customer interaction. The input is the predicted content, and the output is the predicted data provided to the user's device. This allows the user to understand the next discussion topic or customer questions in advance, allowing them to respond more effectively.

[0831] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0832] This paper describes a system for smoothly progressing a meeting by bridging differences in understanding between participants due to their timing and background. This system includes a server, user terminals, a large-scale language model (LLM), and an emotion engine.

[0833] Summary generation function

[0834] System configuration and operation

[0835] The server retrieves past meeting records from a company's internal database, etc. The meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting.

[0836] Specific examples

[0837] Before a user joins the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[0838] Question and Answer Function

[0839] System configuration and operation

[0840] A user uses a device to enter a specific question and send it to the server, which then passes the question through the LLM's question-answering model to generate an appropriate answer, which is then stored by the server and sent back to the user's device.

[0841] Specific examples

[0842] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[0843] Prediction function for next discussion topic

[0844] System configuration and operation

[0845] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics likely to be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[0846] Specific examples

[0847] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[0848] Adding an Emotion Engine

[0849] System configuration and operation

[0850] The system also incorporates an emotion engine that recognizes the user's emotions.

[0851] Using an Emotion Engine in Summarization

[0852] The emotion engine recognizes the user's current emotional state and incorporates this information into the summary generation. For example, if the user is feeling stressed, the summary provided will be more concise and to the point.

[0853] Specific examples

[0854] Before the user joins the next meeting, the device will recognize the user's emotions using an emotion engine, and if the user is feeling stressed, the server will provide a more concise and easy-to-understand summary, allowing the user to understand the information more efficiently.

[0855] Using Emotion Engines in Question Answering

[0856] The emotion engine takes into account the user's emotional state and provides answers with an appropriate tone and content. For example, if the user is confused, the answers will be more detailed and clearer.

[0857] Specific examples

[0858] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device. The user can then review this answer to gain a deeper understanding of the information they need.

[0859] Using an emotion engine to predict the next discussion topic

[0860] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects them in predicting the content of the next discussion, thereby taking participants' emotional reactions into account in the predictions.

[0861] Specific examples

[0862] When predicting what will be discussed in the next meeting, the emotion engine records the emotions of participants in past meetings to identify topics that are likely to lead to smooth discussion and those that are likely to cause confusion or conflict. This allows the server to provide a more prepared prediction to the user's device, allowing the user to make adjustments and take measures in advance.

[0863] As described above, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[0864] The processing flow will be explained below.

[0865] Summary generation function

[0866] Processing flow

[0867] Step 1:

[0868] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[0869] Step 2:

[0870] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[0871] Step 3:

[0872] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[0873] Step 4:

[0874] The server receives the generated summaries from the LLM and stores them in an internal database.

[0875] Step 5:

[0876] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[0877] Step 6:

[0878] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[0879] Question and Answer Function

[0880] Processing flow

[0881] Step 1:

[0882] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[0883] Step 2:

[0884] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[0885] Step 3:

[0886] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[0887] Step 4:

[0888] LLM analyzes the question and related records to generate the best possible answer.

[0889] Step 5:

[0890] The server receives the generated response from the LLM and returns it to the user terminal.

[0891] Step 6:

[0892] The user checks the answers on the terminal and obtains the necessary information.

[0893] Prediction function for next discussion topic

[0894] Processing flow

[0895] Step 1:

[0896] The server retrieves past meeting records and summaries from an internal database.

[0897] Step 2:

[0898] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[0899] Step 3:

[0900] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[0901] Step 4:

[0902] The server receives predicted discussion content from the LLM and stores it in an internal database.

[0903] Step 5:

[0904] When a user prepares for their next meeting, they request the prediction results from their device.

[0905] Step 6:

[0906] The server transmits the saved prediction results to the terminal.

[0907] Step 7:

[0908] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[0909] Adding an Emotion Engine

[0910] Processing flow

[0911] Step 1:

[0912] When a user accesses the terminal and uses the meeting summary or question answering function, the emotion recognition sensor acquires the user's emotion data.

[0913] Step 2:

[0914] The terminal transmits the user's emotional data to the emotion engine, which analyzes the user's current emotional state.

[0915] Step 3:

[0916] The emotion engine sends the results of its analysis to a server, which uses this data to adjust the output of a summary generation model or question-answering model.

[0917] Using an Emotion Engine in Summarization

[0918] Step 4:

[0919] The server receives data from the emotion engine and optimizes the summary content based on the user's emotions. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the main points.

[0920] Using Emotion Engines in Question Answering

[0921] Step 5:

[0922] When the emotion engine recognizes a user's confusion or doubt, the server fine-tunes the output of the question-answering model: for example, if the user is confused, the answers provided will be presented in a more understandable format.

[0923] Using an emotion engine to predict the next discussion topic

[0924] Step 6:

[0925] The server uses data from the emotion engine to predict upcoming discussions based on emotional states recorded in past meeting recordings. This data helps identify topics that will flow smoothly and those that are likely to cause confusion or conflict.

[0926] Step 7:

[0927] When a user prepares for the next meeting, the device receives prediction results along with emotion-linked data, allowing the user to prepare for the meeting efficiently.

[0928] By incorporating an emotion engine, the system can take the user's emotional state into account and provide summaries, question and answer sessions, and predictions of upcoming discussion topics in a more user-friendly format, thereby helping to ensure that meetings proceed effectively.

[0929] Example 2

[0930] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0931] In meetings, differences in understanding due to participants' timing and backgrounds can often be a problem, preventing the meeting from proceeding smoothly. Furthermore, understanding the content of a meeting requires a lot of time and effort, which can make pre-meeting preparations and question-and-answer sessions during the meeting inefficient. It is also difficult to predict the next discussion topic based on meeting records, and providing information without considering the emotions of participants can reduce meeting productivity. A system that can solve these issues is needed.

[0932] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0933] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, and means for adjusting the summaries to suit the users using an emotion engine that recognizes the emotional state of the users. This allows users to efficiently understand past conference records and provides summaries that suit their emotions, making it possible to smoothly proceed with the conference.

[0934] "Meeting records" are data that store the contents of discussions and decisions made during a meeting in text format.

[0935] A "summary generation model" is an algorithm or machine learning model that extracts important points from long text and creates a short summary.

[0936] A "user terminal" is an electronic device such as a computer or smartphone that is directly operated by a user.

[0937] An "emotion engine" is a technology that recognizes the user's emotional state from their facial expressions and voice, and reflects that information in the system's operation.

[0938] A "question answering model" is an algorithm or machine learning model that generates appropriate answers to specific questions.

[0939] A "discussion content predictive model" is a technology that analyzes past data and patterns to predict the topics and agenda items that are likely to be discussed in the next meeting.

[0940] "Acquisition means" refers to the functions and technologies that the system uses to acquire the necessary data from outside.

[0941] "Storage means" refers to the function or technology for storing acquired or generated data in an internal storage device.

[0942] "Providing means" refers to the functions and technologies used to present stored data in a form that is accessible to users.

[0943] "Adjustment means" refers to functions and technologies that optimize the generated summaries and answers depending on the user's emotional state and other conditions.

[0944] This invention is a system for smoothly progressing a meeting by bridging the gap in understanding caused by the timing and background of each participant. This system incorporates a server, a user terminal, a generative AI model (hereafter referred to as LLM), and an emotion engine.

[0945] The server retrieves past meeting records from a company's internal database, etc. These meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts important points from the meeting records and generates a summary. The generated summary is stored on the server and provided to the user's device before the next meeting.

[0946] Before joining the next meeting, the user requests a summary of the previous meeting from the server from their device. The server then sends the summary generated from the past records to the device, allowing the user to quickly check the important points. An example of a specific prompt sentence is "Please tell me the summary of the previous meeting." This allows the user to efficiently prepare for the meeting.

[0947] Furthermore, users can input specific questions into their devices and send them to the server. The server sends the questions to the LLM's question-answering model, which generates an appropriate answer. This answer is stored on the server and sent back to the user's device. An example of a specific prompt sentence is, "What was the budget decided at the last meeting?" The server extracts the relevant information from past meeting records and sends the appropriate answer to the device. This allows users to quickly obtain the information they need.

[0948] The server also retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is stored on the server and provided to user devices before the next meeting. An example of a specific prompt is, "Please tell me the topics that will be discussed in the next meeting." This allows users to know the next discussion topic in advance and effectively prepare for the meeting.

[0949] The system also incorporates an emotion engine that recognizes the user's emotions. The emotion engine recognizes the user's current emotional state and reflects that information in summary generation, question answering, and predictions of the next discussion topic. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the key points. Also, when a user submits a specific question, if the emotion engine recognizes the user's confusion, the response will include more detailed explanations. Furthermore, the emotion engine can take into account the emotional state of participants during past meetings and reflect that in predictions of the next discussion topic.

[0950] In this way, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[0951] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0952] Step 1:

[0953] Input: A user requests a past meeting summary from a terminal to prepare for an upcoming meeting.

[0954] Specific operation: A user makes an API request on a terminal and sends the prompt "Please tell me summaries of past meetings" to the server.

[0955] Output: The request is sent to the device.

[0956] Step 2:

[0957] Input: The server receives a request from a user.

[0958] Specific operation: The server accesses its internal database and executes the SQL query "SELECT FROM summaries WHERE meeting_id=past meeting ID;" to retrieve stored past meeting summaries.

[0959] Output: Past meeting summary data is obtained.

[0960] Step 3:

[0961] Input: The server sends the past conference summary data it has acquired to the LLM summary generation model.

[0962] Specific operation: The server sends the acquired meeting records in text format to the summary generation model using the POST request "curl -X POST 'https: / / api.llm.com / summarize' -d 'meeting record text'".

[0963] Output: The summary data received from the summary generation model is returned.

[0964] Step 4:

[0965] Input: The server sends the summary data received from the LLM to the emotion engine.

[0966] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the summary data. If the user is feeling stressed, the summary will be adjusted to be more concise.

[0967] Output: Optimized summary data is generated.

[0968] Step 5:

[0969] Input: The server sends the optimized summary data to the user terminal.

[0970] Specific operation: The server generates an API response containing summary data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_summary?meeting_id=past meeting ID'".

[0971] Output: The summary data is displayed on the user's terminal.

[0972] Step 6:

[0973] Input: The user inputs a specific question about the next meeting and sends it from the terminal to the server.

[0974] Specific behavior: A user makes an API request on their device and sends the prompt "What was the budget decided at the last meeting?" to the server.

[0975] Output: The request is sent to the device.

[0976] Step 7:

[0977] Input: The server receives a question from the user.

[0978] Specific operation: The server sends a question to the LLM question-answering model, for example, "curl -X POST 'https: / / api.llm.com / answer' -d 'What was the budget decided at the last meeting?'".

[0979] Output: Answer data is returned from the question-answering model.

[0980] Step 8:

[0981] Input: The server sends the answer data received from the LLM to the emotion engine.

[0982] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the answer data. If the user is confused, the answer will be further tailored.

[0983] Output: Optimized answer data is generated.

[0984] Step 9:

[0985] Input: The server sends the optimized answer data to the user's terminal.

[0986] Specific operation: The server generates an API response containing the answer data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_answer?question_id=question ID'".

[0987] Output: The answer data is displayed on the user's terminal.

[0988] Step 10:

[0989] Input: The server retrieves past meeting records and summary data and sends them to the LLM model to predict the content of the next discussion.

[0990] Specific operation: The server retrieves data using the SQL queries "SELECT FROM meeting_records WHERE date BETWEEN 'past date' AND 'current date';" and "SELECT FROM summaries;", and sends the data to the predictive model using "curl -X POST 'https: / / api.llm.com / predict_topics' -d 'past meeting data and summaries'".

[0991] Output: The predictive model returns the next discussion content data.

[0992] Step 11:

[0993] Input: The server sends the discussion content data received from the predictive model to the emotion engine.

[0994] Specific operation: The server uses the emotion engine to analyze the emotional state of participants during past meetings and reflects this in predicting the content of the next discussion.

[0995] Output: Optimized discussion content data is generated.

[0996] Step 12:

[0997] Input: The server sends the optimized discussion content data to the user terminal.

[0998] Specific operation: The server generates an API response containing the discussion content data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_predicted_topics?meeting_id=next meeting ID'".

[0999] Output: Discussion content data is displayed on the user's terminal.

[1000] (Application example 2)

[1001] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1002] In today's business environment, improving the efficiency of meetings is extremely important. However, complex meeting content and the emotional state of participants can make it difficult to understand the content, leading to differences in understanding among participants. Furthermore, if participants are unable to check past meeting records or predict the topics that will be discussed in the next meeting, they may not be able to prepare optimally, which can cause the meeting to stall. A system that solves these problems is needed.

[1003] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for acquiring past records, means for transmitting the acquired records to the generative model, means for saving summary information received from the generative model, means for providing the saved summary information to a user terminal, means for recognizing an emotional state, and means for adjusting the summary information based on the recognized emotional state. This enables users to efficiently understand the contents of past meetings and receive information tailored to their own emotional state.

[1004] "Records" refers to information or data about a meeting or event, stored in the form of documents, audio, video, etc.

[1005] A "generative model" refers to an algorithm or system that uses large-scale language models and artificial intelligence to generate summaries or answers from input data.

[1006] "Summary information" is data that concisely summarizes the important points extracted by the generative model.

[1007] "User terminal" refers to an information processing device used by a user, such as a smartphone, tablet, or PC.

[1008] "Emotional state" refers to the user's current psychological state, including psychological states such as stress, confusion, fatigue, etc.

[1009] "Adjusting" means changing the format of summary information and answers depending on the perceived emotional state, and adapting to make them easier for the user to understand.

[1010] "Forecasting" refers to estimating future events or discussions based on past data and trends. It also refers to the results.

[1011] "Saving" means recording the generated summary information and answers in a database or the like, making them available for future reference or use.

[1012] The present invention relates to a system for improving the efficiency of operation management and maintenance meetings for autonomous vehicles. An embodiment of this system will be described in detail below.

[1013] 1. System Configuration

[1014] This system mainly consists of a server, user devices, a large-scale language model (LLM), and an emotion engine. The server stores meeting records and generated summary information and provides them to user devices. User devices include smartphones, tablets, and PCs, through which users obtain information.

[1015] 2. Summary generation function

[1016] The server retrieves past meeting records and sends them to the LLM summary generation model. The model extracts key points from the input information and generates a summary. This summary is stored on the server and provided to users upon request.

[1017] Examples:

[1018] Before a user joins the next meeting, the user requests a summary of the past meeting from the server, and the server sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points.

[1019] 3. Question and Answer Function

[1020] Users can send specific questions from their devices to the server, which then passes the questions through the LLM's question-answering model to generate appropriate answers, which are then stored on the server and sent back to the user's device.

[1021] Examples:

[1022] The user sends a question from the terminal to the server, such as "What was the maintenance schedule decided at the last meeting?" The server uses LLM to extract the relevant information from the past meeting records, generates an appropriate answer, and sends it to the terminal.

[1023] 4. Prediction of next discussion topic

[1024] The server sends the past meeting records and generated summary information to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is saved by the server and provided to user devices before the next meeting.

[1025] Examples:

[1026] When a user prepares for the next meeting, the user requests a prediction result for the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance.

[1027] 5. Use of Emotion Engine

[1028] The system also incorporates an emotion engine that recognizes the user's current emotional state and reflects that information in summary generation, question answering, and prediction of the next discussion topic.

[1029] Use in summary generation:

[1030] The emotion engine provides a more concise summary if the user is feeling stressed.

[1031] Examples:

[1032] Before the user joins the next meeting, the terminal recognizes the user's emotions using an emotion engine, and if the user is feeling stressed, the server adjusts the summary to make it more concise and easy to understand.

[1033] Use in question answering:

[1034] The sentiment engine makes the answers more detailed and clear if the user is confused.

[1035] Examples:

[1036] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device.

[1037] Use in predicting the next discussion topic:

[1038] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects this in predicting the content of the next discussion.

[1039] Examples:

[1040] When predicting what will be discussed in the next meeting, the emotion engine captures participants' emotions during past meetings to identify topics that will be easy to discuss and those that are likely to cause confusion or conflict.

[1041] 6. Examples of prompt sentences

[1042] For example, the prompt for generating a summary using LLM is as follows:

[1043] Summary: Minutes from the last traffic management meeting...

[1044] An example of a question-and-answer prompt is as follows:

[1045] Question: What was the maintenance schedule decided at the last traffic management meeting? Meeting notes: ...

[1046] As described above, this system uses an emotion engine and LLM to provide information according to the user's emotional state, significantly improving the efficiency of meetings.

[1047] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1048] Step 1:

[1049] (User submits request)

[1050] Before participating in a conference, a user sends a request using a terminal to obtain a summary of past conference records, and this request is sent from the user's terminal to the server.

[1051] Input: User request information (e.g., summary of past meeting records)

[1052] Output: User request data

[1053] What it does: Use a smartphone or tablet to enter and submit a specific request.

[1054] Step 2:

[1055] (Server retrieves the meeting recording)

[1056] The server retrieves past meeting records from the company's internal database.

[1057] Input: User request data

[1058] Output: Past meeting recording data

[1059] Specific operation: The server accesses the database and extracts the records of the specified meeting.

[1060] Step 3:

[1061] (The server sends the meeting recording to the generative model)

[1062] The server sends the acquired meeting records to the LLM summary generation model.

[1063] Input: Past meeting recording data

[1064] Output: Input data to the summary generation model

[1065] Specific operation: The server converts the meeting recording into text format and sends it to the generative model API.

[1066] Step 4:

[1067] (The generative model generates the summary)

[1068] LLM's summary generation model generates summaries from submitted meeting records.

[1069] Input: Input data to the summary generation model

[1070] Output: Generated summary data

[1071] What it does: LLM analyzes the input text and extracts and summarizes the key points.

[1072] Step 5:

[1073] (Server stores and serves summaries)

[1074] The generated summary is stored on a server and provided to the user terminal.

[1075] Input: Generated summary data

[1076] Output: Summary data to user terminal

[1077] Specific operation: The server saves the summary data in a database and sends it to the user's device.

[1078] Step 6:

[1079] (User checks summary)

[1080] The user checks the summary provided on the terminal and prepares for the next meeting based on this summary.

[1081] Input: Summary data to user terminal

[1082] Output: User's understanding after reviewing the summary

[1083] What it does: Read summaries on your smartphone or tablet screen and prepare for your next meeting.

[1084] Step 7:

[1085] (user types question)

[1086] A user sends a specific question about past meeting records from a terminal to the server.

[1087] Input: User question data

[1088] Output: Submitted question data

[1089] Specific behavior: The user enters a specific question in text format and sends it to the server.

[1090] Step 8:

[1091] (The server sends the question to the generative model)

[1092] The server sends the question to the LLM's question-answering model, which generates an appropriate answer.

[1093] Input: User question data

[1094] Output: Input data to the question-answering model

[1095] Specific operation: The server converts the question into text format and sends it to the generative model API.

[1096] Step 9:

[1097] (The generative model generates the answer)

[1098] LLM's question-answering model extracts relevant information from past meeting records and generates appropriate answers.

[1099] Input: Input data to the question-answering model

[1100] Output: Generated response data

[1101] What it does: LLM analyzes the input text and generates an answer.

[1102] Step 10:

[1103] (The server stores and serves the answer)

[1104] The generated answers are stored on the server and provided to the user terminal.

[1105] Input: Generated response data

[1106] Output: Answer data to the user's terminal

[1107] Specific operation: The server saves the answer data in a database and sends it to the user's device.

[1108] Step 11:

[1109] (User checks answer)

[1110] The user can check the answers provided on the terminal, obtain the necessary information, and use it to prepare for the next meeting.

[1111] Input: Answer data to user terminal

[1112] Output: User understanding of the answer

[1113] What it does: Read the answers on your smartphone or tablet screen to help prepare for your next meeting.

[1114] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1115] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1116] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[1117] [Third embodiment]

[1118] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[1119] 5, the data processing system 310 includes the data processing device 12 and a headset terminal 314. An example of the data processing device 12 is a server.

[1120] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1121] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[1122] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1123] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1124] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1125] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1126] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1127] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1128] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1129] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the headset type terminal 314 will be referred to as the "terminal."

[1130] This paper describes a system for smoothly progressing a conference by bridging differences in understanding between participants due to their timing and background. The system includes a server, user terminals, and a large-scale language model (LLM).

[1131] Summary generation function

[1132] System configuration and operation

[1133] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted to text. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts key points from the meeting records and generates a summary. The generated summary is stored by the server and provided to the user's device before the next meeting.

[1134] Specific examples

[1135] Before joining the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[1136] Question and Answer Function

[1137] System configuration and operation

[1138] A user uses a terminal to enter a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[1139] Specific examples

[1140] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[1141] Prediction function for next discussion topic

[1142] System configuration and operation

[1143] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the likely discussions and topics that will be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[1144] Specific examples

[1145] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[1146] As described above, this system uses past meeting records to summarize, answer questions, and predict the content of the next discussion, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[1147] The processing flow will be explained below.

[1148] Summary generation function

[1149] Processing flow

[1150] Step 1:

[1151] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[1152] Step 2:

[1153] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[1154] Step 3:

[1155] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[1156] Step 4:

[1157] The server receives the generated summaries from the LLM and stores them in an internal database.

[1158] Step 5:

[1159] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[1160] Step 6:

[1161] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[1162] Question and Answer Function

[1163] Processing flow

[1164] Step 1:

[1165] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[1166] Step 2:

[1167] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[1168] Step 3:

[1169] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[1170] Step 4:

[1171] LLM analyzes the question and relevant meeting notes to generate the best possible answer.

[1172] Step 5:

[1173] The server receives the generated response from the LLM and returns it to the user terminal.

[1174] Step 6:

[1175] The user checks the answers on the terminal and obtains the necessary information.

[1176] Prediction function for next discussion topic

[1177] Processing flow

[1178] Step 1:

[1179] The server retrieves past meeting records and summaries from an internal database.

[1180] Step 2:

[1181] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[1182] Step 3:

[1183] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[1184] Step 4:

[1185] The server receives predicted discussion content from the LLM and stores it in an internal database.

[1186] Step 5:

[1187] When a user prepares for their next meeting, they request the prediction results from their device.

[1188] Step 6:

[1189] The server transmits the saved prediction results to the terminal.

[1190] Step 7:

[1191] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[1192] Example 1

[1193] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1194] Modern companies hold many meetings, but participants' understanding of the content often varies depending on their timing and background. It's also difficult for new participants to quickly understand the content of past meetings, which can slow down the progress of the meeting. Furthermore, searching for necessary information from past meeting records and predicting the content of the next meeting takes time and effort. A system that solves these problems and supports productive meetings is needed.

[1195] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1196] In this invention, the server includes means for acquiring past meeting records, means for transmitting the acquired meeting records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for transmitting to a question-answering model that generates answers to specific questions, means for providing to the user terminal the answers received from the question-answering model, means for transmitting to a prediction model that predicts the content of the next discussion based on the past meeting records and summaries, and means for providing to the user terminal the discussion content received from the prediction model. This enables users to quickly understand the content of past meetings, easily obtain necessary information, and efficiently prepare for the next meeting.

[1197] "Past meeting records" refers to text data or audio data recorded in a previous meeting.

[1198] A "summary generation model" refers to a large-scale language model that can extract important points from long texts and summarize them concisely.

[1199] A "question answering model" refers to a large-scale language model that analyzes relevant information based on a given question and generates an appropriate answer.

[1200] A "predictive model" refers to a large-scale language model that has the ability to predict future events or outcomes based on past data.

[1201] A "database" refers to a system for efficiently storing, managing, and searching data in various formats.

[1202] A "server" refers to a computer system that receives requests from user terminals, performs the necessary processing, and provides the data.

[1203] "User terminal" refers to a device used by a user, such as a computer, smartphone, or tablet.

[1204] A "generative AI model" refers to an artificial intelligence model that uses a large-scale language model to generate output that is tailored to a specific purpose.

[1205] A "prompt sentence" refers to input text that provides specific instructions to a generative AI model.

[1206] The present invention relates to a system for smoothly progressing a meeting by bridging differences in understanding due to the timing and background of each participant. The system includes a server, a user terminal, and a large-scale language model (LLM).

[1207] Summary generation function

[1208] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted into text. The retrieved meeting record data is sent to an LLM (e.g., GPT-4) specialized in generating text summaries. The LLM extracts key points from the provided meeting records and generates a summary, which is then stored by the server. Before the next meeting, the user can request a past meeting summary from the server, and the summary is provided to the user's device.

[1209] Specific examples

[1210] Before joining the next meeting, a user can request a summary of the previous meeting from the server. The server then sends the summary to the terminal, allowing the user to quickly review the important points and efficiently prepare for the meeting.

[1211] Example prompt statement:

[1212] Summarize past meeting notes, including key points and organizing them so users can review them before the next meeting.

[1213] Question and Answer Function

[1214] A user uses a terminal to input a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[1215] Specific examples

[1216] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[1217] Example prompt statement:

[1218] Please tell me about the budget that was decided at the last meeting.

[1219] Prediction function for next discussion topic

[1220] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model predicts the discussions and topics likely to be discussed in the next meeting based on past data. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[1221] Specific examples

[1222] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[1223] Example prompt statement:

[1224] Anticipate topics likely to be discussed at the next meeting. Use past meeting notes and summaries to inform future discussions.

[1225] This system allows users to quickly understand the content of past meetings, easily obtain the necessary information, and efficiently prepare for the next meeting, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[1226] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1227] Summary generation function

[1228] Step 1:

[1229] Get past meeting records.

[1230] The server accesses the company's internal database and retrieves past meeting records using SQL queries, etc. The retrieved input data is text data or audio data converted into text.

[1231] What happens: The server retrieves data from the database using an API call, executing an SQL query like "SELECT FROM meeting_records_table WHERE date > '2022-01-01'".

[1232] Input: Meeting recordings from the company's internal database (text format or speech-to-text transcription)

[1233] Output: Past meeting recording data

[1234] Step 2:

[1235] Send the meeting notes to a summary generation model.

[1236] The server sends the acquired meeting notes to an LLM model (e.g., GPT-4) specialized in generating text summaries.

[1237] Specific behavior: Convert the meeting recording to JSON format and send the data to the model via API. Include the prompt "Please summarize this meeting recording."

[1238] Input: Past meeting recording data

[1239] Output: Data request sent to the summary generation model

[1240] Step 3:

[1241] Generate a summary.

[1242] The LLM model extracts key points from the provided meeting notes and generates a summary.

[1243] How it works: Large-scale language models such as GPT-4 analyze meeting recordings and use natural language processing (NLP) techniques to extract key parts.

[1244] Input: Meeting recording data

[1245] Output: Generated summary

[1246] Step 4:

[1247] Save the summary.

[1248] The server receives the generated summary and stores it in a database.

[1249] Specific operation: Convert the summary data into JSON format and insert it into the "Summary" table in the database. Execute an SQL query such as "INSERT INTO SummaryTable(ConferenceID, Summary) VALUES(..., ...)".

[1250] Input: Generated summary

[1251] Output: Summary stored in database

[1252] Step 5:

[1253] Provided to the user terminal.

[1254] Before the next meeting, the user requests a summary of the past meeting from the terminal, and the server provides the summary to the user terminal.

[1255] Specific operation: The user enters "Request past conference summaries" on the terminal and clicks the send button. The server retrieves the relevant summaries from the database and returns them to the user terminal in JSON format.

[1256] Input: A request from the user

[1257] Output: Summary sent to user terminal

[1258] Question and Answer Function

[1259] Step 1:

[1260] Enter and submit a specific question.

[1261] The user uses a terminal to enter a specific question and send it to the server.

[1262] Specific operation: The user types "What was the budget decided at the last meeting?" into the terminal and clicks the send button.

[1263] Input: User question

[1264] Output: The question data sent to the server

[1265] Step 2:

[1266] Receive questions.

[1267] The server receives a query from a user and prepares to search for relevant meeting records.

[1268] Specific operation: The server analyzes the question and retrieves the relevant meeting records from the database.

[1269] Input: Question data from the user

[1270] Output: Data search plan within the server

[1271] Step 3:

[1272] Send it to a question-answering model.

[1273] The server sends the relevant meeting notes to the question-answering model.

[1274] Specific behavior: Compile relevant meeting notes in JSON format and send them to the LLM with a request to "provide relevant information to answer this question."

[1275] Input: User questions and related meeting notes

[1276] Output: Data requests sent to the question-answering model

[1277] Step 4:

[1278] Generate an answer.

[1279] The LLM model analyzes the question and related records to generate the best possible answer.

[1280] What it does: LLM uses natural language processing techniques to generate accurate answers to questions.

[1281] Input: Questions and meeting notes

[1282] Output: The generated answer

[1283] Step 5:

[1284] The answer is provided to the user terminal.

[1285] The server receives the generated response from the LLM and sends it to the user terminal.

[1286] Specific operation: The server converts the answer into JSON format and sends it to the user's device, where the user can check the answer.

[1287] Input: Generated Answer

[1288] Output: The answer sent to the user's terminal

[1289] Prediction function for next discussion topic

[1290] Step 1:

[1291] Obtain past meeting notes and summaries.

[1292] The server retrieves past meeting records and summaries from a database.

[1293] Specific operation: The server extracts "all past meeting records and summaries" using an SQL query. It executes a join query in the form of "SELECT FROM meeting record table JOIN summary table ON meeting ID = summary meeting ID".

[1294] Input: Meeting notes and summaries from the database

[1295] Output: Captured meeting records and summary data

[1296] Step 2:

[1297] Send it to a predictive model.

[1298] The server sends the acquired data to the next discussion content prediction model.

[1299] Specific behavior: The server converts the data into JSON format and sends it to the LLM along with a prompt: "Please predict the next discussion topic."

[1300] Input: Meeting notes and summary data

[1301] Output: Data requests sent to the predictive model

[1302] Step 3:

[1303] Generate predictions.

[1304] The LLM model analyzes past data and predicts the content of the next discussion.

[1305] How it works: LLM analyzes past meeting notes, finds patterns and trends, and predicts what will be discussed.

[1306] Input: Past meeting notes and summaries

[1307] Output: The generated prediction

[1308] Step 4:

[1309] Save the prediction.

[1310] The server stores the generated predictions in a database.

[1311] Specific operation: The prediction content is saved in the "prediction result table" of the database using an INSERT query. An SQL query such as "INSERT INTO prediction result table (conference ID, prediction content) VALUES (..., ...)" is executed.

[1312] Input: Generated predictions

[1313] Output: Predictions stored in the database

[1314] Step 5:

[1315] The predicted content is provided to the user terminal.

[1316] When a user prepares for the next meeting, the user requests the prediction results from the server, and the server provides the prediction results to the user's terminal.

[1317] Specific operation: The user types "Request prediction results for the next discussion content" on their device and sends the request to the server. The server retrieves the target prediction results from the database and sends them to the user's device in JSON format.

[1318] Input: A request from the user

[1319] Output: Prediction sent to the user's device

[1320] (Application example 1)

[1321] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1322] With conventional conference systems, participants had to spend a lot of time understanding the content of the conference, making it difficult to conduct the conference efficiently. Furthermore, when dealing with customers in virtual stores, it was difficult to provide appropriate responses based on past customer questions and purchase history. This resulted in a decline in customer satisfaction and a decline in business efficiency.

[1323] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1324] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to the summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for acquiring past inquiries and generating and providing summaries using the summary generation model, means for generating answers to specific questions, and means for generating questions predicted for the user's next visit. This allows conference participants to efficiently prepare for the conference, while at the same time enabling customers to be served quickly and appropriately in the virtual store, thereby improving customer satisfaction.

[1325] "Past meeting records" are data that record the contents of meetings that have taken place in the past.

[1326] "Means for acquiring" refers to the method or device for inputting the necessary information into the system.

[1327] A "summary generation model" is an algorithm that extracts important information from vast amounts of text data and summarizes it concisely.

[1328] A "transmission medium" is a method or tool for transferring data from one place to another.

[1329] "Storage means" refers to a method or device for long-term storage of acquired or generated data.

[1330] "User terminal" means a device such as a computer, smartphone, or tablet used by an end user of the system.

[1331] "Means of providing" refers to the method or device for providing the necessary information to the user.

[1332] "Past inquiries" are records of questions or requests that customers or users have previously made to the system.

[1333] A "specific question" is a single, clearly defined question.

[1334] "Answer generation means" refers to methods or tools that create appropriate answers to input questions.

[1335] "Predicted questions" are questions that customers and users are likely to ask in the future, predicted based on past data and trends.

[1336] The present invention provides a system for efficiently handling customers in meetings and virtual stores. The system includes a server, a user terminal, and a large-scale language model (LLM).

[1337] Configuration and operation of the conference system

[1338] The server retrieves past meeting records from a company's internal database, etc. These records include text data or audio data converted into text. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting. This allows users to quickly understand the content of past meetings and efficiently prepare for meetings.

[1339] Configuration and operation of customer support system

[1340] In the virtual store, the server acquires records of past inquiries. These records are also processed as text data. The server then sends the acquired inquiry records to the LLM summary generation model, which summarizes and saves important information. The user (staff member) can then check this summary information on their device before serving the next customer.

[1341] When a user inputs a specific question, the server sends the question along with any related meeting or inquiry records to the LLM's question-answering model. The model generates the optimal answer to the question, and the server returns the generated answer to the user's device, allowing the user to quickly and appropriately respond to customer questions.

[1342] The server also obtains data based on past meeting and inquiry records to predict the next discussion topic and customer questions. This data is sent to the LLM's prediction model, which predicts the next discussion topic and customer questions. The predicted content is saved by the server and provided to the user's device before the next meeting or customer interaction. This allows users to understand the next discussion topic or customer questions in advance, enabling them to respond more effectively.

[1343] Hardware and software used

[1344] OpenAI API: Provides large-scale language models (LLMs) for natural language processing.

[1345] SQLite: Used as a database management system.

[1346] Python: A programming language for writing invention programs.

[1347] Examples and prompts

[1348] 1. Summary Generation Example:

[1349] Generate a summary of past inquiries from customer ID 1, such as "Please tell me the stock status of sale items" and "Please tell me how to return items."

[1350] Prompt statement:

[1351] Please summarize your inquiry below:

[1352] Please let me know the availability of sale items and how to return them.

[1353] 2. Question and Answer Example:

[1354] Customers ask, "When was your last sale period?"

[1355] Prompt statement:

[1356] Past inquiries:

[1357] Please let me know the availability of sale items and how to return them.

[1358] Customer Question: When was your last sale period?

[1359] Generate the answer

[1360] 3. Examples of questions to predict the next session:

[1361] Next predicted questions based on past inquiries from customer ID 1: "Please tell me the stock status of sale items" and "Please tell me how to return items."

[1362] Prompt statement:

[1363] Generate anticipated questions for your next visit based on the following inquiries:

[1364] Please let me know the availability of sale items and how to return them.

[1365] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1366] Step 1:

[1367] The server retrieves past meeting records or inquiry records from the company's internal database. At this time, the retrieved data is text or speech-to-text conversion. The input is the company's internal database, and the output is the text-converted meeting records or inquiry records.

[1368] Step 2:

[1369] The server sends the acquired meeting records or inquiry records to the summary generation model. At this time, it creates and sends a prompt to the summary generation model, instructing it to summarize the data. The input is the text record, and the output is the summarized text data. As a concrete example, the prompt used is "Please summarize the following inquiry: Please tell me the stock status of sale items. Please tell me how to return them."

[1370] Step 3:

[1371] The server saves the summaries received from the summary generation model and provides them to the user terminal before the next meeting or customer service. At this time, the saved summaries are stored in a database and provided upon user request. The input is the summarized text data, and the output is the summary data provided to the user terminal.

[1372] Step 4:

[1373] When a user inputs a specific question, the terminal sends this question to the server. The server retrieves the meeting records and inquiry records related to the question and sends them to the question-answering model. At this time, a prompt sentence is used to ask the question-answering model for an answer to the question. The input is the user's question and the related records, and the output is the generated answer. As a concrete example, the prompt used is "Past inquiries: What is the stock status of sale items? How do I return them? Customer question: When was the last sale period? Please generate an answer."

[1374] Step 5:

[1375] The server sends the answers received from the question-answering model back to the user device, allowing the user to quickly and appropriately respond to customer questions. The input is the generated answer, and the output is the answer displayed on the user device.

[1376] Step 6:

[1377] The server obtains data to predict the next discussion topic and customer questions based on past meeting records and inquiry records, and sends this data to the prediction model. At this time, the prediction model uses prompt text to make a prediction for the next session. The input is the past records, and the output is the predicted discussion topic and questions. As a specific example, the prompt used is "Based on the following inquiry content, please generate predicted questions for the next visit: Please tell me the stock status of sale items. Please tell me how to return them."

[1378] Step 7:

[1379] The server stores the predicted content and provides it to the user's device before the next meeting or customer interaction. The input is the predicted content, and the output is the predicted data provided to the user's device. This allows the user to understand the next discussion topic or customer questions in advance, allowing them to respond more effectively.

[1380] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1381] This paper describes a system for smoothly progressing a meeting by bridging differences in understanding between participants due to their timing and background. This system includes a server, user terminals, a large-scale language model (LLM), and an emotion engine.

[1382] Summary generation function

[1383] System configuration and operation

[1384] The server retrieves past meeting records from a company's internal database, etc. The meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting.

[1385] Specific examples

[1386] Before a user joins the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[1387] Question and Answer Function

[1388] System configuration and operation

[1389] A user uses a device to enter a specific question and send it to the server, which then passes the question through the LLM's question-answering model to generate an appropriate answer, which is then stored by the server and sent back to the user's device.

[1390] Specific examples

[1391] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[1392] Prediction function for next discussion topic

[1393] System configuration and operation

[1394] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics likely to be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[1395] Specific examples

[1396] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[1397] Adding an Emotion Engine

[1398] System configuration and operation

[1399] The system also incorporates an emotion engine that recognizes the user's emotions.

[1400] Using an Emotion Engine in Summarization

[1401] The emotion engine recognizes the user's current emotional state and incorporates this information into the summary generation. For example, if the user is feeling stressed, the summary provided will be more concise and to the point.

[1402] Specific examples

[1403] Before the user joins the next meeting, the device will recognize the user's emotions using an emotion engine, and if the user is feeling stressed, the server will provide a more concise and easy-to-understand summary, allowing the user to understand the information more efficiently.

[1404] Using Emotion Engines in Question Answering

[1405] The emotion engine takes into account the user's emotional state and provides answers with an appropriate tone and content. For example, if the user is confused, the answers will be more detailed and clearer.

[1406] Specific examples

[1407] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device. The user can then review this answer to gain a deeper understanding of the information they need.

[1408] Using an emotion engine to predict the next discussion topic

[1409] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects them in predicting the content of the next discussion, thereby taking participants' emotional reactions into account in the predictions.

[1410] Specific examples

[1411] When predicting what will be discussed in the next meeting, the emotion engine records the emotions of participants in past meetings to identify topics that are likely to lead to smooth discussion and those that are likely to cause confusion or conflict. This allows the server to provide a more prepared prediction to the user's device, allowing the user to make adjustments and take measures in advance.

[1412] As described above, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[1413] The processing flow will be explained below.

[1414] Summary generation function

[1415] Processing flow

[1416] Step 1:

[1417] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[1418] Step 2:

[1419] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[1420] Step 3:

[1421] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[1422] Step 4:

[1423] The server receives the generated summaries from the LLM and stores them in an internal database.

[1424] Step 5:

[1425] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[1426] Step 6:

[1427] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[1428] Question and Answer Function

[1429] Processing flow

[1430] Step 1:

[1431] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[1432] Step 2:

[1433] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[1434] Step 3:

[1435] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[1436] Step 4:

[1437] LLM analyzes the question and related records to generate the best possible answer.

[1438] Step 5:

[1439] The server receives the generated response from the LLM and returns it to the user terminal.

[1440] Step 6:

[1441] The user checks the answers on the terminal and obtains the necessary information.

[1442] Prediction function for next discussion topic

[1443] Processing flow

[1444] Step 1:

[1445] The server retrieves past meeting records and summaries from an internal database.

[1446] Step 2:

[1447] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[1448] Step 3:

[1449] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[1450] Step 4:

[1451] The server receives predicted discussion content from the LLM and stores it in an internal database.

[1452] Step 5:

[1453] When a user prepares for their next meeting, they request the prediction results from their device.

[1454] Step 6:

[1455] The server transmits the saved prediction results to the terminal.

[1456] Step 7:

[1457] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[1458] Adding an Emotion Engine

[1459] Processing flow

[1460] Step 1:

[1461] When a user accesses the terminal and uses the meeting summary or question answering function, the emotion recognition sensor acquires the user's emotion data.

[1462] Step 2:

[1463] The terminal transmits the user's emotional data to the emotion engine, which analyzes the user's current emotional state.

[1464] Step 3:

[1465] The emotion engine sends the results of its analysis to a server, which uses this data to adjust the output of a summary generation model or question-answering model.

[1466] Using an Emotion Engine in Summarization

[1467] Step 4:

[1468] The server receives data from the emotion engine and optimizes the summary content based on the user's emotions. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the main points.

[1469] Using Emotion Engines in Question Answering

[1470] Step 5:

[1471] When the emotion engine recognizes a user's confusion or doubt, the server fine-tunes the output of the question-answering model: for example, if the user is confused, the answers provided will be presented in a more understandable format.

[1472] Using an emotion engine to predict the next discussion topic

[1473] Step 6:

[1474] The server uses data from the emotion engine to predict upcoming discussions based on emotional states recorded in past meeting recordings. This data helps identify topics that will flow smoothly and those that are likely to cause confusion or conflict.

[1475] Step 7:

[1476] When a user prepares for the next meeting, the device receives prediction results along with emotion-linked data, allowing the user to prepare for the meeting efficiently.

[1477] By incorporating an emotion engine, the system can take the user's emotional state into account and provide summaries, question and answer sessions, and predictions of upcoming discussion topics in a more user-friendly format, thereby helping to ensure that meetings proceed effectively.

[1478] Example 2

[1479] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1480] In meetings, differences in understanding due to participants' timing and backgrounds can often be a problem, preventing the meeting from proceeding smoothly. Furthermore, understanding the content of a meeting requires a lot of time and effort, which can make pre-meeting preparations and question-and-answer sessions during the meeting inefficient. It is also difficult to predict the next discussion topic based on meeting records, and providing information without considering the emotions of participants can reduce meeting productivity. A system that can solve these issues is needed.

[1481] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1482] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, and means for adjusting the summaries to suit the users using an emotion engine that recognizes the emotional state of the users. This allows users to efficiently understand past conference records and provides summaries that suit their emotions, making it possible to smoothly proceed with the conference.

[1483] "Meeting records" are data that store the contents of discussions and decisions made during a meeting in text format.

[1484] A "summary generation model" is an algorithm or machine learning model that extracts important points from long text and creates a short summary.

[1485] A "user terminal" is an electronic device such as a computer or smartphone that is directly operated by a user.

[1486] An "emotion engine" is a technology that recognizes the user's emotional state from their facial expressions and voice, and reflects that information in the system's operation.

[1487] A "question answering model" is an algorithm or machine learning model that generates appropriate answers to specific questions.

[1488] A "discussion content predictive model" is a technology that analyzes past data and patterns to predict the topics and agenda items that are likely to be discussed in the next meeting.

[1489] "Acquisition means" refers to the functions and technologies that the system uses to acquire the necessary data from outside.

[1490] "Storage means" refers to the function or technology for storing acquired or generated data in an internal storage device.

[1491] "Providing means" refers to the functions and technologies used to present stored data in a form that is accessible to users.

[1492] "Adjustment means" refers to functions and technologies that optimize the generated summaries and answers depending on the user's emotional state and other conditions.

[1493] This invention is a system for smoothly progressing a meeting by bridging the gap in understanding caused by the timing and background of each participant. This system incorporates a server, a user terminal, a generative AI model (hereafter referred to as LLM), and an emotion engine.

[1494] The server retrieves past meeting records from a company's internal database, etc. These meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts important points from the meeting records and generates a summary. The generated summary is stored on the server and provided to the user's device before the next meeting.

[1495] Before joining the next meeting, the user requests a summary of the previous meeting from the server from their device. The server then sends the summary generated from the past records to the device, allowing the user to quickly check the important points. An example of a specific prompt sentence is "Please tell me the summary of the previous meeting." This allows the user to efficiently prepare for the meeting.

[1496] Furthermore, users can input specific questions into their devices and send them to the server. The server sends the questions to the LLM's question-answering model, which generates an appropriate answer. This answer is stored on the server and sent back to the user's device. An example of a specific prompt sentence is, "What was the budget decided at the last meeting?" The server extracts the relevant information from past meeting records and sends the appropriate answer to the device. This allows users to quickly obtain the information they need.

[1497] The server also retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is stored on the server and provided to user devices before the next meeting. An example of a specific prompt is, "Please tell me the topics that will be discussed in the next meeting." This allows users to know the next discussion topic in advance and effectively prepare for the meeting.

[1498] The system also incorporates an emotion engine that recognizes the user's emotions. The emotion engine recognizes the user's current emotional state and reflects that information in summary generation, question answering, and predictions of the next discussion topic. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the key points. Also, when a user submits a specific question, if the emotion engine recognizes the user's confusion, the response will include more detailed explanations. Furthermore, the emotion engine can take into account the emotional state of participants during past meetings and reflect that in predictions of the next discussion topic.

[1499] In this way, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[1500] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1501] Step 1:

[1502] Input: A user requests a past meeting summary from a terminal to prepare for an upcoming meeting.

[1503] Specific operation: A user makes an API request on a terminal and sends the prompt "Please tell me summaries of past meetings" to the server.

[1504] Output: The request is sent to the device.

[1505] Step 2:

[1506] Input: The server receives a request from a user.

[1507] Specific operation: The server accesses its internal database and executes the SQL query "SELECT FROM summaries WHERE meeting_id=past meeting ID;" to retrieve stored past meeting summaries.

[1508] Output: Past meeting summary data is obtained.

[1509] Step 3:

[1510] Input: The server sends the past conference summary data it has acquired to the LLM summary generation model.

[1511] Specific operation: The server sends the acquired meeting records in text format to the summary generation model using the POST request "curl -X POST 'https: / / api.llm.com / summarize' -d 'meeting record text'".

[1512] Output: The summary data received from the summary generation model is returned.

[1513] Step 4:

[1514] Input: The server sends the summary data received from the LLM to the emotion engine.

[1515] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the summary data. If the user is feeling stressed, the summary will be adjusted to be more concise.

[1516] Output: Optimized summary data is generated.

[1517] Step 5:

[1518] Input: The server sends the optimized summary data to the user terminal.

[1519] Specific operation: The server generates an API response containing summary data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_summary?meeting_id=past meeting ID'".

[1520] Output: The summary data is displayed on the user's terminal.

[1521] Step 6:

[1522] Input: The user inputs a specific question about the next meeting and sends it from the terminal to the server.

[1523] Specific behavior: A user makes an API request on their device and sends the prompt "What was the budget decided at the last meeting?" to the server.

[1524] Output: The request is sent to the device.

[1525] Step 7:

[1526] Input: The server receives a question from the user.

[1527] Specific operation: The server sends a question to the LLM question-answering model, for example, "curl -X POST 'https: / / api.llm.com / answer' -d 'What was the budget decided at the last meeting?'".

[1528] Output: Answer data is returned from the question-answering model.

[1529] Step 8:

[1530] Input: The server sends the answer data received from the LLM to the emotion engine.

[1531] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the answer data. If the user is confused, the answer will be further tailored.

[1532] Output: Optimized answer data is generated.

[1533] Step 9:

[1534] Input: The server sends the optimized answer data to the user's terminal.

[1535] Specific operation: The server generates an API response containing the answer data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_answer?question_id=question ID'".

[1536] Output: The answer data is displayed on the user's terminal.

[1537] Step 10:

[1538] Input: The server retrieves past meeting records and summary data and sends them to the LLM model to predict the content of the next discussion.

[1539] Specific operation: The server retrieves data using the SQL queries "SELECT FROM meeting_records WHERE date BETWEEN 'past date' AND 'current date';" and "SELECT FROM summaries;", and sends the data to the predictive model using "curl -X POST 'https: / / api.llm.com / predict_topics' -d 'past meeting data and summaries'".

[1540] Output: The predictive model returns the next discussion content data.

[1541] Step 11:

[1542] Input: The server sends the discussion content data received from the predictive model to the emotion engine.

[1543] Specific operation: The server uses the emotion engine to analyze the emotional state of participants during past meetings and reflects this in predicting the content of the next discussion.

[1544] Output: Optimized discussion content data is generated.

[1545] Step 12:

[1546] Input: The server sends the optimized discussion content data to the user terminal.

[1547] Specific operation: The server generates an API response containing the discussion content data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_predicted_topics?meeting_id=next meeting ID'".

[1548] Output: Discussion content data is displayed on the user's terminal.

[1549] (Application example 2)

[1550] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1551] In today's business environment, improving the efficiency of meetings is extremely important. However, complex meeting content and the emotional state of participants can make it difficult to understand the content, leading to differences in understanding among participants. Furthermore, if participants are unable to check past meeting records or predict the topics that will be discussed in the next meeting, they may not be able to prepare optimally, which can cause the meeting to stall. A system that solves these problems is needed.

[1552] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for acquiring past records, means for transmitting the acquired records to the generative model, means for saving summary information received from the generative model, means for providing the saved summary information to a user terminal, means for recognizing an emotional state, and means for adjusting the summary information based on the recognized emotional state. This enables users to efficiently understand the contents of past meetings and receive information tailored to their own emotional state.

[1553] "Records" refers to information or data about a meeting or event, stored in the form of documents, audio, video, etc.

[1554] A "generative model" refers to an algorithm or system that uses large-scale language models and artificial intelligence to generate summaries or answers from input data.

[1555] "Summary information" is data that concisely summarizes the important points extracted by the generative model.

[1556] "User terminal" refers to an information processing device used by a user, such as a smartphone, tablet, or PC.

[1557] "Emotional state" refers to the user's current psychological state, including psychological states such as stress, confusion, fatigue, etc.

[1558] "Adjusting" means changing the format of summary information and answers depending on the perceived emotional state, and adapting to make them easier for the user to understand.

[1559] "Forecasting" refers to estimating future events or discussions based on past data and trends. It also refers to the results.

[1560] "Saving" means recording the generated summary information and answers in a database or the like, making them available for future reference or use.

[1561] The present invention relates to a system for improving the efficiency of operation management and maintenance meetings for autonomous vehicles. An embodiment of this system will be described in detail below.

[1562] 1. System Configuration

[1563] This system mainly consists of a server, user devices, a large-scale language model (LLM), and an emotion engine. The server stores meeting records and generated summary information and provides them to user devices. User devices include smartphones, tablets, and PCs, through which users obtain information.

[1564] 2. Summary generation function

[1565] The server retrieves past meeting records and sends them to the LLM summary generation model. The model extracts key points from the input information and generates a summary. This summary is stored on the server and provided to users upon request.

[1566] Examples:

[1567] Before a user joins the next meeting, the user requests a summary of the past meeting from the server, and the server sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points.

[1568] 3. Question and Answer Function

[1569] Users can send specific questions from their devices to the server, which then passes the questions through the LLM's question-answering model to generate appropriate answers, which are then stored on the server and sent back to the user's device.

[1570] Examples:

[1571] The user sends a question from the terminal to the server, such as "What was the maintenance schedule decided at the last meeting?" The server uses LLM to extract the relevant information from the past meeting records, generates an appropriate answer, and sends it to the terminal.

[1572] 4. Prediction of next discussion topic

[1573] The server sends the past meeting records and generated summary information to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is saved by the server and provided to user devices before the next meeting.

[1574] Examples:

[1575] When a user prepares for the next meeting, the user requests a prediction result for the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance.

[1576] 5. Use of Emotion Engine

[1577] The system also incorporates an emotion engine that recognizes the user's current emotional state and reflects that information in summary generation, question answering, and prediction of the next discussion topic.

[1578] Use in summary generation:

[1579] The emotion engine provides a more concise summary if the user is feeling stressed.

[1580] Examples:

[1581] Before the user joins the next meeting, the terminal recognizes the user's emotions using an emotion engine, and if the user is feeling stressed, the server adjusts the summary to make it more concise and easy to understand.

[1582] Use in question answering:

[1583] The sentiment engine makes the answers more detailed and clear if the user is confused.

[1584] Examples:

[1585] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device.

[1586] Use in predicting the next discussion topic:

[1587] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects this in predicting the content of the next discussion.

[1588] Examples:

[1589] When predicting what will be discussed in the next meeting, the emotion engine captures participants' emotions during past meetings to identify topics that will be easy to discuss and those that are likely to cause confusion or conflict.

[1590] 6. Examples of prompt sentences

[1591] For example, the prompt for generating a summary using LLM is as follows:

[1592] Summary: Minutes from the last traffic management meeting...

[1593] An example of a question-and-answer prompt is as follows:

[1594] Question: What was the maintenance schedule decided at the last traffic management meeting? Meeting notes: ...

[1595] As described above, this system uses an emotion engine and LLM to provide information according to the user's emotional state, significantly improving the efficiency of meetings.

[1596] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1597] Step 1:

[1598] (User submits request)

[1599] Before participating in a conference, a user sends a request using a terminal to obtain a summary of past conference records, and this request is sent from the user's terminal to the server.

[1600] Input: User request information (e.g., summary of past meeting records)

[1601] Output: User request data

[1602] What it does: Use a smartphone or tablet to enter and submit a specific request.

[1603] Step 2:

[1604] (Server retrieves the meeting recording)

[1605] The server retrieves past meeting records from the company's internal database.

[1606] Input: User request data

[1607] Output: Past meeting recording data

[1608] Specific operation: The server accesses the database and extracts the records of the specified meeting.

[1609] Step 3:

[1610] (The server sends the meeting recording to the generative model)

[1611] The server sends the acquired meeting records to the LLM summary generation model.

[1612] Input: Past meeting recording data

[1613] Output: Input data to the summary generation model

[1614] Specific operation: The server converts the meeting recording into text format and sends it to the generative model API.

[1615] Step 4:

[1616] (The generative model generates the summary)

[1617] LLM's summary generation model generates summaries from submitted meeting records.

[1618] Input: Input data to the summary generation model

[1619] Output: Generated summary data

[1620] What it does: LLM analyzes the input text and extracts and summarizes the key points.

[1621] Step 5:

[1622] (Server stores and serves summaries)

[1623] The generated summary is stored on a server and provided to the user terminal.

[1624] Input: Generated summary data

[1625] Output: Summary data to user terminal

[1626] Specific operation: The server saves the summary data in a database and sends it to the user's device.

[1627] Step 6:

[1628] (User checks summary)

[1629] The user checks the summary provided on the terminal and prepares for the next meeting based on this summary.

[1630] Input: Summary data to user terminal

[1631] Output: User's understanding after reviewing the summary

[1632] What it does: Read summaries on your smartphone or tablet screen and prepare for your next meeting.

[1633] Step 7:

[1634] (user types question)

[1635] A user sends a specific question about past meeting records from a terminal to the server.

[1636] Input: User question data

[1637] Output: Submitted question data

[1638] Specific behavior: The user enters a specific question in text format and sends it to the server.

[1639] Step 8:

[1640] (The server sends the question to the generative model)

[1641] The server sends the question to the LLM's question-answering model, which generates an appropriate answer.

[1642] Input: User question data

[1643] Output: Input data to the question-answering model

[1644] Specific operation: The server converts the question into text format and sends it to the generative model API.

[1645] Step 9:

[1646] (The generative model generates the answer)

[1647] LLM's question-answering model extracts relevant information from past meeting records and generates appropriate answers.

[1648] Input: Input data to the question-answering model

[1649] Output: Generated response data

[1650] What it does: LLM analyzes the input text and generates an answer.

[1651] Step 10:

[1652] (The server stores and serves the answer)

[1653] The generated answers are stored on the server and provided to the user terminal.

[1654] Input: Generated response data

[1655] Output: Answer data to the user's terminal

[1656] Specific operation: The server saves the answer data in a database and sends it to the user's device.

[1657] Step 11:

[1658] (User checks answer)

[1659] The user can check the answers provided on the terminal, obtain the necessary information, and use it to prepare for the next meeting.

[1660] Input: Answer data to user terminal

[1661] Output: User understanding of the answer

[1662] What it does: Read the answers on your smartphone or tablet screen to help prepare for your next meeting.

[1663] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1664] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1665] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1666] [Fourth embodiment]

[1667] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1668] 7, a data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1669] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1670] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the control target 443 are also connected to the bus 52.

[1671] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1672] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1673] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1674] The control object 443 includes a display device, LEDs in the eyes, and motors for driving the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1675] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1676] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1677] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1678] In the robot 414, the processor 46 performs the reception output process. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1679] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1680] This paper describes a system for smoothly progressing a conference by bridging differences in understanding between participants due to their timing and background. The system includes a server, user terminals, and a large-scale language model (LLM).

[1681] Summary generation function

[1682] System configuration and operation

[1683] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted to text. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts key points from the meeting records and generates a summary. The generated summary is stored by the server and provided to the user's device before the next meeting.

[1684] Specific examples

[1685] Before joining the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[1686] Question and Answer Function

[1687] System configuration and operation

[1688] A user uses a terminal to enter a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[1689] Specific examples

[1690] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[1691] Prediction function for next discussion topic

[1692] System configuration and operation

[1693] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the likely discussions and topics that will be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[1694] Specific examples

[1695] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[1696] As described above, this system uses past meeting records to summarize, answer questions, and predict the content of the next discussion, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[1697] The processing flow will be explained below.

[1698] Summary generation function

[1699] Processing flow

[1700] Step 1:

[1701] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[1702] Step 2:

[1703] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[1704] Step 3:

[1705] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[1706] Step 4:

[1707] The server receives the generated summaries from the LLM and stores them in an internal database.

[1708] Step 5:

[1709] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[1710] Step 6:

[1711] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[1712] Question and Answer Function

[1713] Processing flow

[1714] Step 1:

[1715] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[1716] Step 2:

[1717] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[1718] Step 3:

[1719] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[1720] Step 4:

[1721] LLM analyzes the question and relevant meeting notes to generate the best possible answer.

[1722] Step 5:

[1723] The server receives the generated response from the LLM and returns it to the user terminal.

[1724] Step 6:

[1725] The user checks the answers on the terminal and obtains the necessary information.

[1726] Prediction function for next discussion topic

[1727] Processing flow

[1728] Step 1:

[1729] The server retrieves past meeting records and summaries from an internal database.

[1730] Step 2:

[1731] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[1732] Step 3:

[1733] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[1734] Step 4:

[1735] The server receives predicted discussion content from the LLM and stores it in an internal database.

[1736] Step 5:

[1737] When a user prepares for their next meeting, they request the prediction results from their device.

[1738] Step 6:

[1739] The server transmits the saved prediction results to the terminal.

[1740] Step 7:

[1741] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[1742] Example 1

[1743] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1744] Modern companies hold many meetings, but participants' understanding of the content often varies depending on their timing and background. It's also difficult for new participants to quickly understand the content of past meetings, which can slow down the progress of the meeting. Furthermore, searching for necessary information from past meeting records and predicting the content of the next meeting takes time and effort. A system that solves these problems and supports productive meetings is needed.

[1745] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1746] In this invention, the server includes means for acquiring past meeting records, means for transmitting the acquired meeting records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for transmitting to a question-answering model that generates answers to specific questions, means for providing to the user terminal the answers received from the question-answering model, means for transmitting to a prediction model that predicts the content of the next discussion based on the past meeting records and summaries, and means for providing to the user terminal the discussion content received from the prediction model. This enables users to quickly understand the content of past meetings, easily obtain necessary information, and efficiently prepare for the next meeting.

[1747] "Past meeting records" refers to text data or audio data recorded in a previous meeting.

[1748] A "summary generation model" refers to a large-scale language model that can extract important points from long texts and summarize them concisely.

[1749] A "question answering model" refers to a large-scale language model that analyzes relevant information based on a given question and generates an appropriate answer.

[1750] A "predictive model" refers to a large-scale language model that has the ability to predict future events or outcomes based on past data.

[1751] A "database" refers to a system for efficiently storing, managing, and searching data in various formats.

[1752] A "server" refers to a computer system that receives requests from user terminals, performs the necessary processing, and provides the data.

[1753] "User terminal" refers to a device used by a user, such as a computer, smartphone, or tablet.

[1754] A "generative AI model" refers to an artificial intelligence model that uses a large-scale language model to generate output that is tailored to a specific purpose.

[1755] A "prompt sentence" refers to input text that provides specific instructions to a generative AI model.

[1756] The present invention relates to a system for smoothly progressing a meeting by bridging differences in understanding due to the timing and background of each participant. The system includes a server, a user terminal, and a large-scale language model (LLM).

[1757] Summary generation function

[1758] The server retrieves past meeting records from a company's internal database or other source. These records include text data or audio data converted into text. The retrieved meeting record data is sent to an LLM (e.g., GPT-4) specialized in generating text summaries. The LLM extracts key points from the provided meeting records and generates a summary, which is then stored by the server. Before the next meeting, the user can request a past meeting summary from the server, and the summary is provided to the user's device.

[1759] Specific examples

[1760] Before joining the next meeting, a user can request a summary of the previous meeting from the server. The server then sends the summary to the terminal, allowing the user to quickly review the important points and efficiently prepare for the meeting.

[1761] Example prompt statement:

[1762] Summarize past meeting notes, including key points and organizing them so users can review them before the next meeting.

[1763] Question and Answer Function

[1764] A user uses a terminal to input a specific question and send it to the server. The server receives the question and sends the relevant meeting notes to the LLM's question-answering model. This model analyzes the question and the relevant notes and generates an optimal answer. The server receives the generated answer and sends it back to the user's terminal.

[1765] Specific examples

[1766] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[1767] Example prompt statement:

[1768] Please tell me about the budget that was decided at the last meeting.

[1769] Prediction function for next discussion topic

[1770] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model predicts the discussions and topics likely to be discussed in the next meeting based on past data. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[1771] Specific examples

[1772] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[1773] Example prompt statement:

[1774] Anticipate topics likely to be discussed at the next meeting. Use past meeting notes and summaries to inform future discussions.

[1775] This system allows users to quickly understand the content of past meetings, easily obtain the necessary information, and efficiently prepare for the next meeting, thereby bridging the gap in understanding between meeting participants and supporting productive meetings.

[1776] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1777] Summary generation function

[1778] Step 1:

[1779] Get past meeting records.

[1780] The server accesses the company's internal database and retrieves past meeting records using SQL queries, etc. The retrieved input data is text data or audio data converted into text.

[1781] What happens: The server retrieves data from the database using an API call, executing an SQL query like "SELECT FROM meeting_records_table WHERE date > '2022-01-01'".

[1782] Input: Meeting recordings from the company's internal database (text format or speech-to-text transcription)

[1783] Output: Past meeting recording data

[1784] Step 2:

[1785] Send the meeting notes to a summary generation model.

[1786] The server sends the acquired meeting notes to an LLM model (e.g., GPT-4) specialized in generating text summaries.

[1787] Specific behavior: Convert the meeting recording to JSON format and send the data to the model via API. Include the prompt "Please summarize this meeting recording."

[1788] Input: Past meeting recording data

[1789] Output: Data request sent to the summary generation model

[1790] Step 3:

[1791] Generate a summary.

[1792] The LLM model extracts key points from the provided meeting notes and generates a summary.

[1793] How it works: Large-scale language models such as GPT-4 analyze meeting recordings and use natural language processing (NLP) techniques to extract key parts.

[1794] Input: Meeting recording data

[1795] Output: Generated summary

[1796] Step 4:

[1797] Save the summary.

[1798] The server receives the generated summary and stores it in a database.

[1799] Specific operation: Convert the summary data into JSON format and insert it into the "Summary" table in the database. Execute an SQL query such as "INSERT INTO SummaryTable(ConferenceID, Summary) VALUES(..., ...)".

[1800] Input: Generated summary

[1801] Output: Summary stored in database

[1802] Step 5:

[1803] Provided to the user terminal.

[1804] Before the next meeting, the user requests a summary of the past meeting from the terminal, and the server provides the summary to the user terminal.

[1805] Specific operation: The user enters "Request past conference summaries" on the terminal and clicks the send button. The server retrieves the relevant summaries from the database and returns them to the user terminal in JSON format.

[1806] Input: A request from the user

[1807] Output: Summary sent to user terminal

[1808] Question and Answer Function

[1809] Step 1:

[1810] Enter and submit a specific question.

[1811] The user uses a terminal to enter a specific question and send it to the server.

[1812] Specific operation: The user types "What was the budget decided at the last meeting?" into the terminal and clicks the send button.

[1813] Input: User question

[1814] Output: The question data sent to the server

[1815] Step 2:

[1816] Receive questions.

[1817] The server receives a query from a user and prepares to search for relevant meeting records.

[1818] Specific operation: The server analyzes the question and retrieves the relevant meeting records from the database.

[1819] Input: Question data from the user

[1820] Output: Data search plan within the server

[1821] Step 3:

[1822] Send it to a question-answering model.

[1823] The server sends the relevant meeting notes to the question-answering model.

[1824] Specific behavior: Compile relevant meeting notes in JSON format and send them to the LLM with a request to "provide relevant information to answer this question."

[1825] Input: User questions and related meeting notes

[1826] Output: Data requests sent to the question-answering model

[1827] Step 4:

[1828] Generate an answer.

[1829] The LLM model analyzes the question and related records to generate the best possible answer.

[1830] What it does: LLM uses natural language processing techniques to generate accurate answers to questions.

[1831] Input: Questions and meeting notes

[1832] Output: The generated answer

[1833] Step 5:

[1834] The answer is provided to the user terminal.

[1835] The server receives the generated response from the LLM and sends it to the user terminal.

[1836] Specific operation: The server converts the answer into JSON format and sends it to the user's device, where the user can check the answer.

[1837] Input: Generated Answer

[1838] Output: The answer sent to the user's terminal

[1839] Prediction function for next discussion topic

[1840] Step 1:

[1841] Obtain past meeting notes and summaries.

[1842] The server retrieves past meeting records and summaries from a database.

[1843] Specific operation: The server extracts "all past meeting records and summaries" using an SQL query. It executes a join query in the form of "SELECT FROM meeting record table JOIN summary table ON meeting ID = summary meeting ID".

[1844] Input: Meeting notes and summaries from the database

[1845] Output: Captured meeting records and summary data

[1846] Step 2:

[1847] Send it to a predictive model.

[1848] The server sends the acquired data to the next discussion content prediction model.

[1849] Specific behavior: The server converts the data into JSON format and sends it to the LLM along with a prompt: "Please predict the next discussion topic."

[1850] Input: Meeting notes and summary data

[1851] Output: Data requests sent to the predictive model

[1852] Step 3:

[1853] Generate predictions.

[1854] The LLM model analyzes past data and predicts the content of the next discussion.

[1855] How it works: LLM analyzes past meeting notes, finds patterns and trends, and predicts what will be discussed.

[1856] Input: Past meeting notes and summaries

[1857] Output: The generated prediction

[1858] Step 4:

[1859] Save the prediction.

[1860] The server stores the generated predictions in a database.

[1861] Specific operation: The prediction content is saved in the "prediction result table" of the database using an INSERT query. An SQL query such as "INSERT INTO prediction result table (conference ID, prediction content) VALUES (..., ...)" is executed.

[1862] Input: Generated predictions

[1863] Output: Predictions stored in the database

[1864] Step 5:

[1865] The predicted content is provided to the user terminal.

[1866] When a user prepares for the next meeting, the user requests the prediction results from the server, and the server provides the prediction results to the user's terminal.

[1867] Specific operation: The user types "Request prediction results for the next discussion content" on their device and sends the request to the server. The server retrieves the target prediction results from the database and sends them to the user's device in JSON format.

[1868] Input: A request from the user

[1869] Output: Prediction sent to the user's device

[1870] (Application example 1)

[1871] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1872] With conventional conference systems, participants had to spend a lot of time understanding the content of the conference, making it difficult to conduct the conference efficiently. Furthermore, when dealing with customers in virtual stores, it was difficult to provide appropriate responses based on past customer questions and purchase history. This resulted in a decline in customer satisfaction and a decline in business efficiency.

[1873] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1874] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to the summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, means for acquiring past inquiries and generating and providing summaries using the summary generation model, means for generating answers to specific questions, and means for generating questions predicted for the user's next visit. This allows conference participants to efficiently prepare for the conference, while at the same time enabling customers to be served quickly and appropriately in the virtual store, thereby improving customer satisfaction.

[1875] "Past meeting records" are data that record the contents of meetings that have taken place in the past.

[1876] "Means for acquiring" refers to the method or device for inputting the necessary information into the system.

[1877] A "summary generation model" is an algorithm that extracts important information from vast amounts of text data and summarizes it concisely.

[1878] A "transmission medium" is a method or tool for transferring data from one place to another.

[1879] "Storage means" refers to a method or device for long-term storage of acquired or generated data.

[1880] "User terminal" means a device such as a computer, smartphone, or tablet used by an end user of the system.

[1881] "Means of providing" refers to the method or device for providing the necessary information to the user.

[1882] "Past inquiries" are records of questions or requests that customers or users have previously made to the system.

[1883] A "specific question" is a single, clearly defined question.

[1884] "Answer generation means" refers to methods or tools that create appropriate answers to input questions.

[1885] "Predicted questions" are questions that customers and users are likely to ask in the future, predicted based on past data and trends.

[1886] The present invention provides a system for efficiently handling customers in meetings and virtual stores. The system includes a server, a user terminal, and a large-scale language model (LLM).

[1887] Configuration and operation of the conference system

[1888] The server retrieves past meeting records from a company's internal database, etc. These records include text data or audio data converted into text. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting. This allows users to quickly understand the content of past meetings and efficiently prepare for meetings.

[1889] Configuration and operation of customer support system

[1890] In the virtual store, the server acquires records of past inquiries. These records are also processed as text data. The server then sends the acquired inquiry records to the LLM summary generation model, which summarizes and saves important information. The user (staff member) can then check this summary information on their device before serving the next customer.

[1891] When a user inputs a specific question, the server sends the question along with any related meeting or inquiry records to the LLM's question-answering model. The model generates the optimal answer to the question, and the server returns the generated answer to the user's device, allowing the user to quickly and appropriately respond to customer questions.

[1892] The server also obtains data based on past meeting and inquiry records to predict the next discussion topic and customer questions. This data is sent to the LLM's prediction model, which predicts the next discussion topic and customer questions. The predicted content is saved by the server and provided to the user's device before the next meeting or customer interaction. This allows users to understand the next discussion topic or customer questions in advance, enabling them to respond more effectively.

[1893] Hardware and software used

[1894] OpenAI API: Provides large-scale language models (LLMs) for natural language processing.

[1895] SQLite: Used as a database management system.

[1896] Python: A programming language for writing invention programs.

[1897] Examples and prompts

[1898] 1. Summary Generation Example:

[1899] Generate a summary of past inquiries from customer ID 1, such as "Please tell me the stock status of sale items" and "Please tell me how to return items."

[1900] Prompt statement:

[1901] Please summarize your inquiry below:

[1902] Please let me know the availability of sale items and how to return them.

[1903] 2. Question and Answer Example:

[1904] Customers ask, "When was your last sale period?"

[1905] Prompt statement:

[1906] Past inquiries:

[1907] Please let me know the availability of sale items and how to return them.

[1908] Customer Question: When was your last sale period?

[1909] Generate the answer

[1910] 3. Examples of questions to predict the next session:

[1911] Next predicted questions based on past inquiries from customer ID 1: "Please tell me the stock status of sale items" and "Please tell me how to return items."

[1912] Prompt statement:

[1913] Generate anticipated questions for your next visit based on the following inquiries:

[1914] Please let me know the availability of sale items and how to return them.

[1915] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1916] Step 1:

[1917] The server retrieves past meeting records or inquiry records from the company's internal database. At this time, the retrieved data is text or speech-to-text conversion. The input is the company's internal database, and the output is the text-converted meeting records or inquiry records.

[1918] Step 2:

[1919] The server sends the acquired meeting records or inquiry records to the summary generation model. At this time, it creates and sends a prompt to the summary generation model, instructing it to summarize the data. The input is the text record, and the output is the summarized text data. As a concrete example, the prompt used is "Please summarize the following inquiry: Please tell me the stock status of sale items. Please tell me how to return them."

[1920] Step 3:

[1921] The server saves the summaries received from the summary generation model and provides them to the user terminal before the next meeting or customer service. At this time, the saved summaries are stored in a database and provided upon user request. The input is the summarized text data, and the output is the summary data provided to the user terminal.

[1922] Step 4:

[1923] When a user inputs a specific question, the terminal sends this question to the server. The server retrieves the meeting records and inquiry records related to the question and sends them to the question-answering model. At this time, a prompt sentence is used to ask the question-answering model for an answer to the question. The input is the user's question and the related records, and the output is the generated answer. As a concrete example, the prompt used is "Past inquiries: What is the stock status of sale items? How do I return them? Customer question: When was the last sale period? Please generate an answer."

[1924] Step 5:

[1925] The server sends the answers received from the question-answering model back to the user device, allowing the user to quickly and appropriately respond to customer questions. The input is the generated answer, and the output is the answer displayed on the user device.

[1926] Step 6:

[1927] The server obtains data to predict the next discussion topic and customer questions based on past meeting records and inquiry records, and sends this data to the prediction model. At this time, the prediction model uses prompt text to make a prediction for the next session. The input is the past records, and the output is the predicted discussion topic and questions. As a specific example, the prompt used is "Based on the following inquiry content, please generate predicted questions for the next visit: Please tell me the stock status of sale items. Please tell me how to return them."

[1928] Step 7:

[1929] The server stores the predicted content and provides it to the user's device before the next meeting or customer interaction. The input is the predicted content, and the output is the predicted data provided to the user's device. This allows the user to understand the next discussion topic or customer questions in advance, allowing them to respond more effectively.

[1930] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1931] This paper describes a system for smoothly progressing a meeting by bridging differences in understanding between participants due to their timing and background. This system includes a server, user terminals, a large-scale language model (LLM), and an emotion engine.

[1932] Summary generation function

[1933] System configuration and operation

[1934] The server retrieves past meeting records from a company's internal database, etc. The meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model, which extracts important points from the meeting records and generates a summary. The generated summary is saved by the server and provided to the user's device before the next meeting.

[1935] Specific examples

[1936] Before a user joins the next meeting, the user requests a summary of the previous meeting from the server. The server then sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points. This allows the user to efficiently prepare for the meeting.

[1937] Question and Answer Function

[1938] System configuration and operation

[1939] A user uses a device to enter a specific question and send it to the server, which then passes the question through the LLM's question-answering model to generate an appropriate answer, which is then stored by the server and sent back to the user's device.

[1940] Specific examples

[1941] The user sends a question from their device to the server, such as "What was the budget decided at the last meeting?" The server uses LLM to extract the relevant information from past meeting records, generates an appropriate answer, and sends it to the device. The user can then check the answer on their device and quickly obtain the information they need.

[1942] Prediction function for next discussion topic

[1943] System configuration and operation

[1944] The server retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics likely to be discussed in the next meeting. The predicted discussion content is stored by the server and provided to user devices before the next meeting.

[1945] Specific examples

[1946] When a user prepares for the next meeting, the user requests a prediction of the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance. This allows the user to effectively prepare for the meeting.

[1947] Adding an Emotion Engine

[1948] System configuration and operation

[1949] The system also incorporates an emotion engine that recognizes the user's emotions.

[1950] Using an Emotion Engine in Summarization

[1951] The emotion engine recognizes the user's current emotional state and incorporates this information into the summary generation. For example, if the user is feeling stressed, the summary provided will be more concise and to the point.

[1952] Specific examples

[1953] Before the user joins the next meeting, the device will recognize the user's emotions using an emotion engine, and if the user is feeling stressed, the server will provide a more concise and easy-to-understand summary, allowing the user to understand the information more efficiently.

[1954] Using Emotion Engines in Question Answering

[1955] The emotion engine takes into account the user's emotional state and provides answers with an appropriate tone and content. For example, if the user is confused, the answers will be more detailed and clearer.

[1956] Specific examples

[1957] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device. The user can then review this answer to gain a deeper understanding of the information they need.

[1958] Using an emotion engine to predict the next discussion topic

[1959] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects them in predicting the content of the next discussion, thereby taking participants' emotional reactions into account in the predictions.

[1960] Specific examples

[1961] When predicting what will be discussed in the next meeting, the emotion engine records the emotions of participants in past meetings to identify topics that are likely to lead to smooth discussion and those that are likely to cause confusion or conflict. This allows the server to provide a more prepared prediction to the user's device, allowing the user to make adjustments and take measures in advance.

[1962] As described above, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[1963] The processing flow will be explained below.

[1964] Summary generation function

[1965] Processing flow

[1966] Step 1:

[1967] The server queries and retrieves past meeting records from an internal database, which are stored in text format.

[1968] Step 2:

[1969] The server sends the acquired meeting recordings to the large-scale language model (LLM) summary generation API, where the recordings are formatted in a format that the model can process.

[1970] Step 3:

[1971] LLM analyzes submitted meeting notes, extracts key points and generates a concise summary.

[1972] Step 4:

[1973] The server receives the generated summaries from the LLM and stores them in an internal database.

[1974] Step 5:

[1975] Before a meeting, a user requests a summary from their terminal, and the server returns the stored summary to the terminal.

[1976] Step 6:

[1977] Users can view summaries through their devices to quickly understand the contents of recent discussions.

[1978] Question and Answer Function

[1979] Processing flow

[1980] Step 1:

[1981] The user uses the terminal to enter a specific question and clicks a send button to send the question to the server.

[1982] Step 2:

[1983] The server analyzes the question received from the user and sends the question to the large-scale language model (LLM) question-answering API.

[1984] Step 3:

[1985] The server provides the LLM with past meeting notes relevant to the question, which are retrieved from the server's internal database.

[1986] Step 4:

[1987] LLM analyzes the question and related records to generate the best possible answer.

[1988] Step 5:

[1989] The server receives the generated response from the LLM and returns it to the user terminal.

[1990] Step 6:

[1991] The user checks the answers on the terminal and obtains the necessary information.

[1992] Prediction function for next discussion topic

[1993] Processing flow

[1994] Step 1:

[1995] The server retrieves past meeting records and summaries from an internal database.

[1996] Step 2:

[1997] The server sends the acquired data to the prediction API of the large-scale language model (LLM), which uses the data as input to predict the next discussion topic.

[1998] Step 3:

[1999] LLM analyzes past data to predict the topics and agendas that are likely to be discussed at upcoming meetings.

[2000] Step 4:

[2001] The server receives predicted discussion content from the LLM and stores it in an internal database.

[2002] Step 5:

[2003] When a user prepares for their next meeting, they request the prediction results from their device.

[2004] Step 6:

[2005] The server transmits the saved prediction results to the terminal.

[2006] Step 7:

[2007] Users can check the prediction results through their devices, effectively understand the content of the next discussion, and prepare for the meeting.

[2008] Adding an Emotion Engine

[2009] Processing flow

[2010] Step 1:

[2011] When a user accesses the terminal and uses the meeting summary or question answering function, the emotion recognition sensor acquires the user's emotion data.

[2012] Step 2:

[2013] The terminal transmits the user's emotional data to the emotion engine, which analyzes the user's current emotional state.

[2014] Step 3:

[2015] The emotion engine sends the results of its analysis to a server, which uses this data to adjust the output of a summary generation model or question-answering model.

[2016] Using an Emotion Engine in Summarization

[2017] Step 4:

[2018] The server receives data from the emotion engine and optimizes the summary content based on the user's emotions. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the main points.

[2019] Using Emotion Engines in Question Answering

[2020] Step 5:

[2021] When the emotion engine recognizes a user's confusion or doubt, the server fine-tunes the output of the question-answering model: for example, if the user is confused, the answers provided will be presented in a more understandable format.

[2022] Using an emotion engine to predict the next discussion topic

[2023] Step 6:

[2024] The server uses data from the emotion engine to predict upcoming discussions based on emotional states recorded in past meeting recordings. This data helps identify topics that will flow smoothly and those that are likely to cause confusion or conflict.

[2025] Step 7:

[2026] When a user prepares for the next meeting, the device receives prediction results along with emotion-linked data, allowing the user to prepare for the meeting efficiently.

[2027] By incorporating an emotion engine, the system can take the user's emotional state into account and provide summaries, question and answer sessions, and predictions of upcoming discussion topics in a more user-friendly format, thereby helping to ensure that meetings proceed effectively.

[2028] Example 2

[2029] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[2030] In meetings, differences in understanding due to participants' timing and backgrounds can often be a problem, preventing the meeting from proceeding smoothly. Furthermore, understanding the content of a meeting requires a lot of time and effort, which can make pre-meeting preparations and question-and-answer sessions during the meeting inefficient. It is also difficult to predict the next discussion topic based on meeting records, and providing information without considering the emotions of participants can reduce meeting productivity. A system that can solve these issues is needed.

[2031] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[2032] In this invention, the server includes means for acquiring past conference records, means for transmitting the acquired conference records to a summary generation model, means for saving summaries received from the summary generation model, means for providing the saved summaries to a user terminal, and means for adjusting the summaries to suit the users using an emotion engine that recognizes the emotional state of the users. This allows users to efficiently understand past conference records and provides summaries that suit their emotions, making it possible to smoothly proceed with the conference.

[2033] "Meeting records" are data that store the contents of discussions and decisions made during a meeting in text format.

[2034] A "summary generation model" is an algorithm or machine learning model that extracts important points from long text and creates a short summary.

[2035] A "user terminal" is an electronic device such as a computer or smartphone that is directly operated by a user.

[2036] An "emotion engine" is a technology that recognizes the user's emotional state from their facial expressions and voice, and reflects that information in the system's operation.

[2037] A "question answering model" is an algorithm or machine learning model that generates appropriate answers to specific questions.

[2038] A "discussion content predictive model" is a technology that analyzes past data and patterns to predict the topics and agenda items that are likely to be discussed in the next meeting.

[2039] "Acquisition means" refers to the functions and technologies that the system uses to acquire the necessary data from outside.

[2040] "Storage means" refers to the function or technology for storing acquired or generated data in an internal storage device.

[2041] "Providing means" refers to the functions and technologies used to present stored data in a form that is accessible to users.

[2042] "Adjustment means" refers to functions and technologies that optimize the generated summaries and answers depending on the user's emotional state and other conditions.

[2043] This invention is a system for smoothly progressing a meeting by bridging the gap in understanding caused by the timing and background of each participant. This system incorporates a server, a user terminal, a generative AI model (hereafter referred to as LLM), and an emotion engine.

[2044] The server retrieves past meeting records from a company's internal database, etc. These meeting records are stored in text format. The server then sends the retrieved meeting records to the LLM summary generation model. This model extracts important points from the meeting records and generates a summary. The generated summary is stored on the server and provided to the user's device before the next meeting.

[2045] Before joining the next meeting, the user requests a summary of the previous meeting from the server from their device. The server then sends the summary generated from the past records to the device, allowing the user to quickly check the important points. An example of a specific prompt sentence is "Please tell me the summary of the previous meeting." This allows the user to efficiently prepare for the meeting.

[2046] Furthermore, users can input specific questions into their devices and send them to the server. The server sends the questions to the LLM's question-answering model, which generates an appropriate answer. This answer is stored on the server and sent back to the user's device. An example of a specific prompt sentence is, "What was the budget decided at the last meeting?" The server extracts the relevant information from past meeting records and sends the appropriate answer to the device. This allows users to quickly obtain the information they need.

[2047] The server also retrieves past meeting records and summaries and sends them to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is stored on the server and provided to user devices before the next meeting. An example of a specific prompt is, "Please tell me the topics that will be discussed in the next meeting." This allows users to know the next discussion topic in advance and effectively prepare for the meeting.

[2048] The system also incorporates an emotion engine that recognizes the user's emotions. The emotion engine recognizes the user's current emotional state and reflects that information in summary generation, question answering, and predictions of the next discussion topic. For example, if the user is feeling stressed, the summary provided will be more concise and emphasize the key points. Also, when a user submits a specific question, if the emotion engine recognizes the user's confusion, the response will include more detailed explanations. Furthermore, the emotion engine can take into account the emotional state of participants during past meetings and reflect that in predictions of the next discussion topic.

[2049] In this way, by combining the emotion engine, the system can take into account the user's emotional state and provide summaries, Q&A, and predictions of the next discussion topic in a more user-friendly format. This makes it possible to more effectively bridge the gap in understanding among meeting participants and support the progress of productive meetings.

[2050] The flow of the identification process in the second embodiment will be described with reference to FIG.

[2051] Step 1:

[2052] Input: A user requests a past meeting summary from a terminal to prepare for an upcoming meeting.

[2053] Specific operation: A user makes an API request on a terminal and sends the prompt "Please tell me summaries of past meetings" to the server.

[2054] Output: The request is sent to the device.

[2055] Step 2:

[2056] Input: The server receives a request from a user.

[2057] Specific operation: The server accesses its internal database and executes the SQL query "SELECT FROM summaries WHERE meeting_id=past meeting ID;" to retrieve stored past meeting summaries.

[2058] Output: Past meeting summary data is obtained.

[2059] Step 3:

[2060] Input: The server sends the past conference summary data it has acquired to the LLM summary generation model.

[2061] Specific operation: The server sends the acquired meeting records in text format to the summary generation model using the POST request "curl -X POST 'https: / / api.llm.com / summarize' -d 'meeting record text'".

[2062] Output: The summary data received from the summary generation model is returned.

[2063] Step 4:

[2064] Input: The server sends the summary data received from the LLM to the emotion engine.

[2065] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the summary data. If the user is feeling stressed, the summary will be adjusted to be more concise.

[2066] Output: Optimized summary data is generated.

[2067] Step 5:

[2068] Input: The server sends the optimized summary data to the user terminal.

[2069] Specific operation: The server generates an API response containing summary data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_summary?meeting_id=past meeting ID'".

[2070] Output: The summary data is displayed on the user's terminal.

[2071] Step 6:

[2072] Input: The user inputs a specific question about the next meeting and sends it from the terminal to the server.

[2073] Specific behavior: A user makes an API request on their device and sends the prompt "What was the budget decided at the last meeting?" to the server.

[2074] Output: The request is sent to the device.

[2075] Step 7:

[2076] Input: The server receives a question from the user.

[2077] Specific operation: The server sends a question to the LLM question-answering model, for example, "curl -X POST 'https: / / api.llm.com / answer' -d 'What was the budget decided at the last meeting?'".

[2078] Output: Answer data is returned from the question-answering model.

[2079] Step 8:

[2080] Input: The server sends the answer data received from the LLM to the emotion engine.

[2081] Specific operation: The server uses the emotion engine to recognize the user's emotional state and further optimize the answer data. If the user is confused, the answer will be further tailored.

[2082] Output: Optimized answer data is generated.

[2083] Step 9:

[2084] Input: The server sends the optimized answer data to the user's terminal.

[2085] Specific operation: The server generates an API response containing the answer data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_answer?question_id=question ID'".

[2086] Output: The answer data is displayed on the user's terminal.

[2087] Step 10:

[2088] Input: The server retrieves past meeting records and summary data and sends them to the LLM model to predict the content of the next discussion.

[2089] Specific operation: The server retrieves data using the SQL queries "SELECT FROM meeting_records WHERE date BETWEEN 'past date' AND 'current date';" and "SELECT FROM summaries;", and sends the data to the predictive model using "curl -X POST 'https: / / api.llm.com / predict_topics' -d 'past meeting data and summaries'".

[2090] Output: The predictive model returns the next discussion content data.

[2091] Step 11:

[2092] Input: The server sends the discussion content data received from the predictive model to the emotion engine.

[2093] Specific operation: The server uses the emotion engine to analyze the emotional state of participants during past meetings and reflects this in predicting the content of the next discussion.

[2094] Output: Optimized discussion content data is generated.

[2095] Step 12:

[2096] Input: The server sends the optimized discussion content data to the user terminal.

[2097] Specific operation: The server generates an API response containing the discussion content data and sends it to the terminal in the form of "curl -X GET 'https: / / api.server.com / get_predicted_topics?meeting_id=next meeting ID'".

[2098] Output: Discussion content data is displayed on the user's terminal.

[2099] (Application example 2)

[2100] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[2101] In today's business environment, improving the efficiency of meetings is extremely important. However, complex meeting content and the emotional state of participants can make it difficult to understand the content, leading to differences in understanding among participants. Furthermore, if participants are unable to check past meeting records or predict the topics that will be discussed in the next meeting, they may not be able to prepare optimally, which can cause the meeting to stall. A system that solves these problems is needed.

[2102] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for acquiring past records, means for transmitting the acquired records to the generative model, means for saving summary information received from the generative model, means for providing the saved summary information to a user terminal, means for recognizing an emotional state, and means for adjusting the summary information based on the recognized emotional state. This enables users to efficiently understand the contents of past meetings and receive information tailored to their own emotional state.

[2103] "Records" refers to information or data about a meeting or event, stored in the form of documents, audio, video, etc.

[2104] A "generative model" refers to an algorithm or system that uses large-scale language models and artificial intelligence to generate summaries or answers from input data.

[2105] "Summary information" is data that concisely summarizes the important points extracted by the generative model.

[2106] "User terminal" refers to an information processing device used by a user, such as a smartphone, tablet, or PC.

[2107] "Emotional state" refers to the user's current psychological state, including psychological states such as stress, confusion, fatigue, etc.

[2108] "Adjusting" means changing the format of summary information and answers depending on the perceived emotional state, and adapting to make them easier for the user to understand.

[2109] "Forecasting" refers to estimating future events or discussions based on past data and trends. It also refers to the results.

[2110] "Saving" means recording the generated summary information and answers in a database or the like, making them available for future reference or use.

[2111] The present invention relates to a system for improving the efficiency of operation management and maintenance meetings for autonomous vehicles. An embodiment of this system will be described in detail below.

[2112] 1. System Configuration

[2113] This system mainly consists of a server, user devices, a large-scale language model (LLM), and an emotion engine. The server stores meeting records and generated summary information and provides them to user devices. User devices include smartphones, tablets, and PCs, through which users obtain information.

[2114] 2. Summary generation function

[2115] The server retrieves past meeting records and sends them to the LLM summary generation model. The model extracts key points from the input information and generates a summary. This summary is stored on the server and provided to users upon request.

[2116] Examples:

[2117] Before a user joins the next meeting, the user requests a summary of the past meeting from the server, and the server sends the summary generated from the past recording to the terminal, allowing the user to quickly check the important points.

[2118] 3. Question and Answer Function

[2119] Users can send specific questions from their devices to the server, which then passes the questions through the LLM's question-answering model to generate appropriate answers, which are then stored on the server and sent back to the user's device.

[2120] Examples:

[2121] The user sends a question from the terminal to the server, such as "What was the maintenance schedule decided at the last meeting?" The server uses LLM to extract the relevant information from the past meeting records, generates an appropriate answer, and sends it to the terminal.

[2122] 4. Prediction of next discussion topic

[2123] The server sends the past meeting records and generated summary information to an LLM model to predict the next meeting's discussion content. This model uses past data to predict the agenda and topics that are likely to be discussed in the next meeting. The predicted discussion content is saved by the server and provided to user devices before the next meeting.

[2124] Examples:

[2125] When a user prepares for the next meeting, the user requests a prediction result for the next discussion topic from the device to the server. The server then sends the prediction result to the device, allowing the user to know the next discussion topic in advance.

[2126] 5. Use of Emotion Engine

[2127] The system also incorporates an emotion engine that recognizes the user's current emotional state and reflects that information in summary generation, question answering, and prediction of the next discussion topic.

[2128] Use in summary generation:

[2129] The emotion engine provides a more concise summary if the user is feeling stressed.

[2130] Examples:

[2131] Before the user joins the next meeting, the terminal recognizes the user's emotions using an emotion engine, and if the user is feeling stressed, the server adjusts the summary to make it more concise and easy to understand.

[2132] Use in question answering:

[2133] The sentiment engine makes the answers more detailed and clear if the user is confused.

[2134] Examples:

[2135] When a user submits a specific question, if the emotion engine recognizes the user's confusion, the server uses LLM to generate an answer with more detailed explanation and send it to the device.

[2136] Use in predicting the next discussion topic:

[2137] The emotion engine takes into account the emotional states of participants recorded in past meeting records and reflects this in predicting the content of the next discussion.

[2138] Examples:

[2139] When predicting what will be discussed in the next meeting, the emotion engine captures participants' emotions during past meetings to identify topics that will be easy to discuss and those that are likely to cause confusion or conflict.

[2140] 6. Examples of prompt sentences

[2141] For example, the prompt for generating a summary using LLM is as follows:

[2142] Summary: Minutes from the last traffic management meeting...

[2143] An example of a question-and-answer prompt is as follows:

[2144] Question: What was the maintenance schedule decided at the last traffic management meeting? Meeting notes: ...

[2145] As described above, this system uses an emotion engine and LLM to provide information according to the user's emotional state, significantly improving the efficiency of meetings.

[2146] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[2147] Step 1:

[2148] (User submits request)

[2149] Before participating in a conference, a user sends a request using a terminal to obtain a summary of past conference records, and this request is sent from the user's terminal to the server.

[2150] Input: User request information (e.g., summary of past meeting records)

[2151] Output: User request data

[2152] What it does: Use a smartphone or tablet to enter and submit a specific request.

[2153] Step 2:

[2154] (Server retrieves the meeting recording)

[2155] The server retrieves past meeting records from the company's internal database.

[2156] Input: User request data

[2157] Output: Past meeting recording data

[2158] Specific operation: The server accesses the database and extracts the records of the specified meeting.

[2159] Step 3:

[2160] (The server sends the meeting recording to the generative model)

[2161] The server sends the acquired meeting records to the LLM summary generation model.

[2162] Input: Past meeting recording data

[2163] Output: Input data to the summary generation model

[2164] Specific operation: The server converts the meeting recording into text format and sends it to the generative model API.

[2165] Step 4:

[2166] (The generative model generates the summary)

[2167] LLM's summary generation model generates summaries from submitted meeting records.

[2168] Input: Input data to the summary generation model

[2169] Output: Generated summary data

[2170] What it does: LLM analyzes the input text and extracts and summarizes the key points.

[2171] Step 5:

[2172] (Server stores and serves summaries)

[2173] The generated summary is stored on a server and provided to the user terminal.

[2174] Input: Generated summary data

[2175] Output: Summary data to user terminal

[2176] Specific operation: The server saves the summary data in a database and sends it to the user's device.

[2177] Step 6:

[2178] (User checks summary)

[2179] The user checks the summary provided on the terminal and prepares for the next meeting based on this summary.

[2180] Input: Summary data to user terminal

[2181] Output: User's understanding after reviewing the summary

[2182] What it does: Read summaries on your smartphone or tablet screen and prepare for your next meeting.

[2183] Step 7:

[2184] (user types question)

[2185] A user sends a specific question about past meeting records from a terminal to the server.

[2186] Input: User question data

[2187] Output: Submitted question data

[2188] Specific behavior: The user enters a specific question in text format and sends it to the server.

[2189] Step 8:

[2190] (The server sends the question to the generative model)

[2191] The server sends the question to the LLM's question-answering model, which generates an appropriate answer.

[2192] Input: User question data

[2193] Output: Input data to the question-answering model

[2194] Specific operation: The server converts the question into text format and sends it to the generative model API.

[2195] Step 9:

[2196] (The generative model generates the answer)

[2197] LLM's question-answering model extracts relevant information from past meeting records and generates appropriate answers.

[2198] Input: Input data to the question-answering model

[2199] Output: Generated response data

[2200] What it does: LLM analyzes the input text and generates an answer.

[2201] Step 10:

[2202] (The server stores and serves the answer)

[2203] The generated answers are stored on the server and provided to the user terminal.

[2204] Input: Generated response data

[2205] Output: Answer data to the user's terminal

[2206] Specific operation: The server saves the answer data in a database and sends it to the user's device.

[2207] Step 11:

[2208] (User checks answer)

[2209] The user can check the answers provided on the terminal, obtain the necessary information, and use it to prepare for the next meeting.

[2210] Input: Answer data to user terminal

[2211] Output: User understanding of the answer

[2212] What it does: Read the answers on your smartphone or tablet screen to help prepare for your next meeting.

[2213] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[2214] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[2215] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[2216] The emotion identification model 59 as an emotion engine may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[2217] FIG. 9 is a diagram illustrating an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and actions arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[2218] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[2219] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[2220] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[2221] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[2222] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[2223] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[2224] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[2225] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[2226] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via ...

Claims

1. a means for obtaining past meeting records; a means for transmitting the acquired meeting records to a summary generation model; means for storing summaries received from the summary generation model; means for providing the stored summary to a user terminal; A system including:

2. a means of generating answers to specific questions from past meeting records; means for providing the generated answer to a user terminal; The system of claim 1 further comprising:

3. A means of predicting the content of upcoming discussions based on past meeting records and summaries; A means for providing the predicted discussion content to a user terminal; The system of claim 1 further comprising:

Images