Support device, support method, support system, and support program

Abstract

Leveraging large-scale language models, we support mutual understanding among multiple users. [Solution] The support device is characterized by comprising: a generation AI connection relay unit that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic; a support unit that instructs the large-scale language model to organize the content of the acquired conversation data using the user's input information and make suggestions; and a communication unit that transmits the suggestions to a predetermined terminal.

Description

Technical Field

[0001] The present invention relates to a support device, a support method, a support system, and a support program.

Background Art

[0002] Conventional LLM (Large Language Model) services can be used by a user using an LLM module corresponding to a user account created for personal use. For example, it is possible to consult an LLM about various topics such as work, home, education, love, etc., or to request store searches, reservation procedures, etc., and the LLM can return various answers such as synchronization, in-depth exploration, advice, proposal of search results, execution results of reservation procedures, etc. according to the content of the user's speech.

[0003] For example, the behavior control system disclosed in Patent Document 1 includes an input unit that receives user input, a processing unit that performs specific processing using a text generation model that generates text according to the input text, and an output unit that controls the behavior of an electronic device so as to output the result of the specific processing. The processing unit determines whether or not a predetermined trigger condition is satisfied, and when the trigger condition is satisfied, uses the output of the text generation model when the information obtained from the user input is used as the input text to obtain the result of the specific processing. The specific processing is characterized in that it is processing that uses the output of the text generation model from a fictional character whose personality has been formed in advance.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] According to the behavior control system described above, a device that engages in natural language dialogue can perform actions according to the requested content and display the results in the form of dialogue. Generally, when multiple people are involved in a discussion, topic, or matter under consideration (topic), a discussion takes place and a conclusion is reached. Therefore, various difficulties can arise before reaching a certain conclusion, such as conflicting opinions or stalled progress due to differing values. Conventional LLM services are limited to one-on-one dialogue between the user and the LLM, and cannot assist in discussions among multiple people or match users from among multiple people.

[0006] The object of this invention has been made in view of the above points, and aims to support mutual understanding among multiple users by utilizing LLM. [Means for solving the problem]

[0007] The present invention includes several means for solving at least some of the above problems, the following being an example. An assistance device according to one aspect of the present invention is characterized by comprising: a generation AI connection relay unit that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic; a support unit that instructs the large-scale language model to organize the content of the stored conversation data using the user's input information and make suggestions; and a communication unit that transmits the suggestions to a predetermined terminal.

[0008] Furthermore, in the above-described support device, the topic in the conversation data may be a topic relating to a predetermined agenda, the user's input information may be an opinion on the agenda, and the proposal may include a conclusion on the agenda.

[0009] Furthermore, in the above-described support device, the topic in the conversation data is a topic relating to the relationships between multiple users, the user's input information is their opinion on the grouping of multiple users, and the proposal may include the presentation of the grouping.

[0010] Furthermore, in the above-described support device, the large-scale language model includes individual modules corresponding to each user and an integrated module common to all users, the conversation data is conversation data between the user and the corresponding individual module, and the support unit issues the instructions to the integrated module, the instructions may include causing the individual module to organize the content using the conversation data with the user.

[0011] Furthermore, in the above-described support device, the large-scale language model includes individual modules corresponding to each user and an integrated module common to all users, the conversation data is conversation data between the user and the corresponding individual module, and the support unit issues the instructions to the integrated module, the instructions of which may include causing the integrated module to organize the content using the conversation data.

[0012] Furthermore, in the above-described support device, the support unit may instruct the large-scale language model to remove privacy information from the conversation data, organize the content, and then make suggestions.

[0013] Furthermore, in the above-described support device, the support unit may instruct the large-scale language model to organize the content and make multiple suggestions.

[0014] Furthermore, in the above-described support device, the support unit may instruct the large-scale language model to organize the content, prioritize it, and make multiple suggestions.

[0015] Furthermore, in the above-described support device, the support unit may instruct the large-scale language model to organize the content in accordance with the majority vote using the input information of multiple users from the acquired conversation data and make a proposal.

[0016] Furthermore, the above-described support device includes a space provision unit that defines a population targeting the users who have expressed their intention to participate, wherein the user's input information is an opinion on the grouping of the multiple users included in the population, and the proposal may include the presentation of the grouping of the multiple users included in the population.

[0017] Furthermore, a support method according to another aspect of the present invention is characterized in that the processor performs a generation AI connection relay process that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic; a support process that instructs the large-scale language model to organize the content and make suggestions using the user's input information from the stored conversation data; and a communication process that transmits the suggestions to a predetermined terminal.

[0018] Furthermore, a support system according to another aspect of the present invention is characterized by performing a generation AI connection relay step of accumulating conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic; a support step of instructing the large-scale language model to organize the content of the accumulated conversation data using the user's input information and make suggestions; and a display step of displaying the suggestions.

[0019] Furthermore, a support program according to another aspect of the present invention is characterized in that the processor performs a generation AI connection relay step of storing conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic; a support step of instructing the large-scale language model to organize the content of the stored conversation data using the user's input information and make a proposal; and a communication step of transmitting the proposal to a predetermined terminal. [Effects of the Invention]

[0020] According to the present invention, mutual understanding among multiple users can be supported by utilizing LLM.

[0021] Problems, configurations, and effects other than those described above will be clarified by the following description of the embodiments.

Brief Description of the Drawings

[0022] [Figure 1] It is an example of a block diagram of an opinion aggregation support system according to an embodiment. [Figure 2] It is a diagram showing an example of the hardware configuration of a user terminal. [Figure 3] It is a diagram showing an example of the hardware configuration of a support device. [Figure 4] It is a diagram showing an example of the flow of opinion aggregation space creation processing. [Figure 5] It is a diagram showing an example of the flow of opinion aggregation processing. [Figure 6] It is a diagram showing an example of an opinion aggregation prompt between users. [Figure 7] It is a diagram showing an example of the flow of opinion aggregation processing (using integrated AI). [Figure 8] It is a diagram showing an example of an opinion aggregation prompt between users (using integrated AI). [Figure 9] It is an example of a block diagram of an ingredient support system according to another embodiment. [Figure 10] It is a diagram showing an example of the flow of population creation processing. [Figure 11] It is a diagram showing an example of the flow of ingredient processing. [Figure 12] It is a diagram showing an example of an ingredient prompt. [Figure 13] It is a diagram showing an example of the flow of ingredient processing (using integrated AI). [Figure 14] It is a diagram showing an example of an ingredient prompt (using integrated AI).

Modes for Carrying Out the Invention

[0023] Below, an opinion aggregation support system 1 and a grouping support system 1' (or both collectively referred to as the support system, or each of them individually referred to as the support system) equipped with support functions according to one embodiment of the present invention will be described with reference to the drawings.

[0024] The opinion aggregation support system 1 according to this embodiment can utilize LLM to support mutual understanding and conclusions in discussions among multiple users. Furthermore, the grouping support system 1' according to this embodiment can utilize LLM to support matching among multiple users.

[0025] Figure 1 is an example of a block diagram of the opinion aggregation support system according to this embodiment. The opinion aggregation support system 1 includes a user terminal 100 (and participating user terminals 100' used by other users), a support device 200, and a generation AI service 300.

[0026] User terminal 100 is a terminal used by users of the opinion aggregation support system 1. User terminal 100 is connected to the generation AI service 300 via network 50 for communication. User terminal 100 is also connected to the support device 200 via network 50 for communication.

[0027] Network 50 is a communication network such as a LAN (Local Area Network), WAN (Wide Area Network), the Internet, or a mobile phone network. Network 50 may also be a VPN (Virtual Private Network) on a wireless communication network such as a mobile phone network.

[0028] The user terminal 100 is a so-called smartphone or personal computer. However, it is not limited to these; the user terminal 100 may also be various wearable devices such as smartwatches, smart glasses, or wireless earphones, or various information processing devices such as PDAs (Personal Data Assistants) or tablet devices.

[0029] The user terminal 100 receives input from the user via input devices such as voice, text, and other screen operations, and communicates appropriately with the generated AI service 300 or support device 200 according to the input.

[0030] In this case, for example, the user terminal 100 engages in chat with the Large Language Model (LLM) provided by the AI ​​generation service 300 via the network 50. While it is preferable that the LLM be customized for the user, it is not limited to this, and a general-purpose LLM may also be used.

[0031] When the user terminal 100 receives a predetermined response from the generation AI service 300 via the network 50, it performs output processing. The user terminal 100 also transmits input information received from input devices, etc. (described later), to the generation AI service 300 via the network 50.

[0032] Furthermore, the user terminal 100 generates an opinion aggregation space on the support device 200 via the network 50, allows other users to participate in the opinion aggregation space using participating user terminals 100', and instructs the support device 200 to aggregate the opinions within the opinion aggregation space to obtain aggregated opinions. The opinion aggregation space is created in units similar to chat rooms, and conversations with the generation AI service 300 while participating in the opinion aggregation space are recognized as the opinions of the users in the opinion aggregation space and are shared together with the opinions of other participants.

[0033] The user terminal 100 comprises a processing unit 110, a storage unit 120, a communication unit 130, an input receiving unit 140, and an output unit 150. The processing unit 110 includes a dialogue processing unit 111, a generation AI connection processing unit 112, an opinion aggregation space creation unit 113, an opinion aggregation instruction unit 114, and an opinion aggregation participation unit 115. The storage unit 120 includes login information 121 and relationship information 122. The participating user terminal 100' has the same configuration as the user terminal 100.

[0034] The dialogue processing unit 111 receives input such as voice or text for the generation AI service 300 via the input reception unit 140, and passes this information to the generation AI service 300 as text or voice data. When the dialogue processing unit 111 receives a predetermined response from the generation AI service 300, it outputs it via the output unit 150. The output unit 150 controls screen display and voice output.

[0035] The generation AI connection processing unit 112 uses the login information 121 to control the start, continuation, and termination of communication to the generation AI service 300. The login information 121 stores configuration information such as information to identify the generation AI service 300 and information for accessing it (e.g., URL: Uniform Resource Locator), and information for user authentication (e.g., a user identifier such as an email address and a password).

[0036] The opinion aggregation space creation unit 113 creates an opinion aggregation space, which is a space for sharing conversation data, including the content of conversations between the user and the generative AI service 300 that uses a large-scale language model (LLM).

[0037] The opinion aggregation instruction unit 114 instructs the aggregation of opinions expressed in the conversation data shared within a predetermined opinion aggregation space.

[0038] The opinion aggregation participation unit 115 allows the user and the user's AI model 310 associated with that user to participate in a predetermined opinion aggregation space.

[0039] Relationship information 122 is information that manages the account information of other users (e.g., participating users) who have a relationship with the user, similar to an SNS (Social Network Service), and is used for sending messages to other users, etc.

[0040] The communication unit 130 communicates with the generation AI service 300 and the support device 200 via the network 50. The input receiving unit 140 receives input from the user of the user terminal 100 and passes it on to one of the following depending on the content of the received input: the dialogue processing unit 111, the opinion aggregation space creation unit 113, the opinion aggregation instruction unit 114, or the opinion aggregation participation unit 115. The input receiving unit 140 may receive voice input via a microphone and identify the text, or it may identify the text via eye-tracking input, etc., but is not limited to these, and may also receive text from, for example, a hardware keyboard or a software keyboard. When the output unit 150 receives predetermined response information that it has been instructed to output from the generation AI service 300 from the communication unit 130, it displays it on the display and outputs it as sound from the speaker.

[0041] The support device 200 is a so-called server device. However, it is not limited to this, and the support device 200 may be any kind of information processing device, such as a personal computer, smartphone, workstation, PDA (Personal Data Assistant), or tablet device.

[0042] When the support device 200 receives a request for the creation of an opinion aggregation space from a user terminal 100 via the network 50, it appropriately creates an opinion aggregation space in accordance with the request. Furthermore, when the support device 200 receives a participation request from a participating user terminal 100' via the network 50, it allows the terminal to participate in the opinion aggregation space in accordance with the request. Additionally, when the support device 200 receives a request for opinion aggregation regarding a conversation between a user terminal 100 participating in the opinion aggregation space and the generating AI service 300, it appropriately creates aggregated opinions in accordance with the request.

[0043] In this process, the support device 200 delegates the opinion aggregation process to the generation AI service 300 via the network 50.

[0044] The support device 200 includes a space provision unit 210, a generation AI connection relay unit 220, a support unit 230, and a communication unit 240. The space provision unit 210 creates an opinion aggregation space upon request and manages the users participating in the space.

[0045] The generation AI connection relay unit 220 stores conversation data, including the content of conversations between a user and a predetermined large-scale language model (LLM), in an opinion aggregation space on a predetermined topic. Multiple methods can be employed for acquiring conversation data. For example, one method is to log the conversation data of users participating in the opinion aggregation space in real time and store the conversation data in the support device 200, or to store the conversation data with the user in a user AI model 310 associated with the user participating in the opinion aggregation space.

[0046] The support unit 230 instructs the generation AI service 300 to organize the content of the user input information, i.e., the utterance information, from the conversation data acquired from users participating in the opinion aggregation space, and propose aggregated opinions on the topic.

[0047] The communication unit 240 controls communication (wired or wireless) between the support device 200 and other devices via the network 50.

[0048] By using the opinion aggregation support system 1, for example, a user can share a common topic, such as a predetermined agenda item, with other participants, and automatically record their opinions in response to their utterances during a voice chat conversation with the generation AI service 300 using a smartphone, which is the user terminal 100. The system can then aggregate the opinions automatically recorded in response to the utterances of other participants during similar conversations with the generation AI service 300 and have the generation AI service 300 propose a conclusion on the agenda item.

[0049] The Generative AI Service 300 is a service that provides the functions of so-called generative AIs such as GPT, Gemini, and Claude via an API (Application Programming Interface). The Generative AI Service 300 provides the Generative AI with commands (prompts) in natural language to generate the desired results. The Generative AI Service 300 comprises a user AI model 310, an integrated AI model 320, and a communication unit 330.

[0050] In this embodiment, when the generation AI service 300 receives instructions, for example via an API, it causes the user AI model 310, which is a custom LLM for each user, to generate response information, and sends the result to the source of the instructions as a return value of the API. At that time, the user AI model 310 receives information related to the response, reads this information as part of the information to be added and considered, and reflects it in the generation of the response. It is desirable that the user AI model 310 has learned the characteristics of the user's conversations, etc., based on the conversation history with the user, or continues to learn.

[0051] For example, when the generation AI service 300 receives spoken or text from a user terminal 100 on a topic, such as a matter for consultation, it creates a natural language response using the voice or text information, taking into account the input information provided up to that point, and sends it back to the user terminal 100. For example, if the topic received is a consultation about interpersonal relationships, the generation AI service 300 generates a response in natural language that continues the conversation in various ways, such as empathy, further exploration, and advice, and sends it back to the user terminal 100.

[0052] The user AI model 310 is not limited to this, and may also use a generative AI held by the user terminal 100. In that case, the generative AI may be any generative AI that uses a large-scale language model.

[0053] Generative AI is a pre-trained model created by machine learning or deep learning a large-scale language model using a neural network (NN) with language data. User AI model 310 is implemented using generative AI. Generative AI is not limited to neural networks (NN); other known methods may be used. Furthermore, it is desirable to use a generative AI that has been tuned to achieve higher accuracy through techniques such as few-shot learning and fine-tuning.

[0054] The integrated AI model 320 is a user-common LLM that exists integrally for information sharing and collaborative processing among user-specific AI models 310, which are LLMs customized for each user. User-specific AI models 310 basically operate in a closed environment for each user, and are optimized and learn more according to the user's usage environment. In contrast, the integrated AI model 320 oversees information sharing and collaborative processing among user-specific AI models 310. In other words, if the user-specific AI models 310 are panelists in a panel discussion, the integrated AI model 320 acts like a moderator that aggregates the opinions of the panelists. For example, the integrated AI model 320 has the individual user-specific AI models 310, which are modules, organize and output content about a topic using conversation data with the user. Then, the integrated AI model 320 organizes and integrates the content about the topic that the user-specific AI models 310 have organized, or proposes conclusions or ranked proposals on the agenda based on majority rule logic.

[0055] The communication unit 330 communicates with the user terminal 100 and the support device 200 via the network 50.

[0056] Figure 2 shows an example of the hardware configuration of user terminal 100. User terminal 100 includes an input device 101, a processor 102, storage 103, memory 104, output device 105, communication device 106, and a bus 107 connecting the devices. Participating user terminal 100' also has the same hardware configuration as user terminal 100.

[0057] The input device 101 is a variety of input devices such as a keyboard, mouse, touch panel, or microphone. The input receiving unit 140 of the user terminal 100 is realized by the input device 101 and the processor 102.

[0058] The processor 102 is a computing device such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), and it executes processing according to a program recorded in memory 104 or storage 103. This program is, for example, an application program that can be executed on an OS (Operating System) program. In the user terminal 100, processing is performed by the processor 102, which operates according to the program read from memory 104 or storage 103. The processing unit 110, the dialogue processing unit 111, the generation AI connection processing unit 112, the opinion aggregation space creation unit 113, the opinion aggregation instruction unit 114, and the opinion aggregation participation unit 115 each realize their respective functions by having the processor 102 execute programs.

[0059] Storage 103 is a writable and readable storage device. The functions of the storage unit 120, login information 121, and relation information 122 are realized by the storage 103 or memory 104. The functions of the storage unit 120 may also be realized by a storage device connected via the communication device 106. Memory 104 is a storage device such as RAM (Random Access Memory) or flash memory, and functions as a storage area where programs and data are temporarily read.

[0060] The output device 105 is, for example, a display device such as a liquid crystal display or an organic EL display, or a speaker. The output unit 150 is realized by the output device 105 and the processor 102.

[0061] The communication device 106 is an interface for connecting the user terminal 100 to an external device. For example, the communication device 106 is a network card that performs wired communication via a wired connection. Alternatively, the communication device 106 performs wireless communication using an antenna that can utilize predetermined radio waves (e.g., 5GHz band, 2.4GHz band, etc.) to establish a connection with the generated AI service 300 according to the Wi-Fi standard. The communication unit 130 is implemented by the communication device 106.

[0062] Furthermore, the processing of each component of the user terminal 100 may be performed on one piece of hardware or on multiple pieces of hardware. Also, the processing of each component of the user terminal 100 may be implemented by one program or by multiple programs.

[0063] The above is an example of the hardware configuration of user terminal 100.

[0064] Each configuration of the user terminal 100 can be further classified into many more components depending on the processing content. Alternatively, each component can be classified to perform even more processing tasks.

[0065] Furthermore, each processing unit (processing unit 110, dialogue processing unit 111, generation AI connection processing unit 112, opinion aggregation space creation unit 113, opinion aggregation instruction unit 114, opinion aggregation participation unit 115) may be constructed using dedicated hardware (ASIC, GPU, etc.) to realize its respective function. Also, the processing of each processing unit may be executed on a single piece of hardware or on multiple pieces of hardware.

[0066] Figure 3 shows an example of the hardware configuration of the support device. The support device 200 comprises a processor 201, memory 202, storage 203, communication device 204, and a bus 205 connecting the devices. In addition, the support device 200 may also include an input device.

[0067] The processor 201 is a computing device such as a CPU or GPU, and it performs processing according to a program recorded in memory 202 or storage 203. This program is, for example, an application program that can be executed on an OS program. In the support device 200, processing is performed by the processor 201, which operates according to the program read from memory 202 or storage 203. The space provision unit 210, the generation AI connection relay unit 220, and the support unit 230 each realize their respective functions by having the processor 201 execute a program.

[0068] Memory 202 is a storage device such as RAM (Random Access Memory) or flash memory, and functions as a storage area where programs and data are temporarily read. Storage 203 is a writable and readable storage device. The storage unit 120 may also function through a storage device connected via the communication device 204.

[0069] The communication device 204 is an interface for connecting the support device 200 to an external device. For example, the communication device 204 is a network card that performs wired communication via a wired connection. Alternatively, the communication device 204 performs wireless communication using an antenna that can utilize predetermined radio waves (e.g., 5GHz band, 2.4GHz band, etc.) to establish a connection with the generated AI service 300 and the user terminal 100 according to the Wi-Fi standard.

[0070] Furthermore, the processing of each component of the support device 200 may be performed on one piece of hardware or on multiple pieces of hardware. Also, the processing of each component of the support device 200 may be implemented by one program or by multiple programs.

[0071] The communication unit 240 of the support device 200 described above is implemented by the communication device 204. The above is an example of the hardware configuration of the support device 200.

[0072] Each component of the support device 200 can be further classified into many more components depending on the processing content. Alternatively, each component can be classified to perform even more processing.

[0073] Furthermore, each processing unit (space provisioning unit 210, generation AI connection relay unit 220, support unit 230) may be constructed using dedicated hardware (ASIC, GPU, etc.) to realize its respective function. Also, the processing of each processing unit may be executed on a single piece of hardware or on multiple pieces of hardware. In addition, the processing of each component of the support device 200 may be implemented by a single program or by multiple programs.

[0074] The above is an example of the hardware configuration of the support device 200.

[0075] Next, the operation of the opinion aggregation support system 1 in this embodiment will be described.

[0076] First, in this embodiment, a user of the opinion aggregation support system 1 uses a user terminal 100 to create a shared discussion space on the support device 200 for aggregating opinions with other users. Users participating in the opinion aggregation join this discussion space, forming a population for opinion aggregation. Each user belonging to the opinion aggregation population then freely engages in conversation with the generation AI service 300. After a predetermined period (for example, 3 days, 1 week, 1 month, or longer), the support device 200 organizes the content of opinions among the users participating in the discussion space based on the conversations during that period and makes a proposal.

[0077] This process includes two distinct processes: an opinion aggregation space creation process in which the user terminal 100 requests the support device 200 to create an opinion aggregation space on the support device 200, which will be a shared discussion space for aggregating opinions with other users on a predetermined topic (agenda), and allows the user and other users to participate in the opinion aggregation space on the support device 200 in response to requests from the user terminal 100; and an opinion aggregation process in which the content is organized using conversation data, including the content of conversations between the users who participated in the opinion aggregation space and the generating AI service 300, and a proposal is made for aggregating opinions on the topic.

[0078] Figure 4 shows an example of the flow of the opinion aggregation space creation process. The opinion aggregation space creation process is started at the user terminal 100 upon a start command from the user.

[0079] First, the opinion aggregation space creation unit 113 receives the agenda and the instruction to create an opinion aggregation space (step S101). Specifically, when the opinion aggregation space creation unit 113 receives the topic and the instruction to create an opinion aggregation space via the input reception unit 140, it transmits the topic and the instruction to create an opinion aggregation space to the support device 200.

[0080] The space provision unit 210 of the support device 200 creates an opinion aggregation space (step S102). Specifically, the space provision unit 210 creates a discourse space having a predetermined space identifier, associates the received topic as topic information of the discourse space, and associates the user terminal 100 as the owner. Then, the space provision unit 210 transmits the space identifier of the created discourse space to the user terminal 100.

[0081] Then, the opinion aggregation space creation unit 113 accepts the designation of invited users (step S103). Specifically, when the opinion aggregation space creation unit 113 receives the user identifiers of one or more invited users via the input reception unit 140, it transmits the space identifier received in step S102 and one or more user identifiers to the support device 200.

[0082] Then, the space provision unit 210 issues a license for the speech space (step S104). Specifically, the space provision unit 210 generates unique information for each invited user, such as a password, passkey, random key, or QR code (registered trademark), and registers it as a license for the speech space. Then, the space provision unit 210 transmits the created license for the speech space to the user terminal 100.

[0083] Then, the opinion aggregation participation unit 115 receives a spatial login instruction (step S105). Specifically, the opinion aggregation participation unit 115 receives a login instruction to the created discourse space via the input reception unit 140. The opinion aggregation participation unit 115 then transmits the spatial identifier of the created discourse space, the user identifier, the permission slip, and the login information 121 necessary for connecting the user's AI model 310 to the support device 200.

[0084] Then, the space provision unit 210 starts maintaining the connection from the user (step S106). Specifically, the space provision unit 210 connects to the user AI model 310 using the login information 121 received from the user terminal 100 in order to relay communication between the user terminal 100 and the user AI model 310. Then, the generated AI connection relay unit 220 relays the dialogue between the user terminal 100 and the user AI model 310 (step S107). The user and the user AI model 310 converse about the target topic. In this conversation, the user's thoughts and ideas about the topic are revealed.

[0085] The above is an example of the flow for creating an opinion aggregation space. According to this example of the flow for creating an opinion aggregation space, a user can create a discourse space for which opinions will be aggregated, and participation in that space, that is, the interaction with the user's AI model 310 during participation, can be made the subject of opinion aggregation.

[0086] Furthermore, other participating users, using their participating user terminal 100', receive a spatial identifier and a permission slip from their user terminal 100 by some means, for example, via an SNS (Social Network Service) using relation information 122, or in person. After receiving these, the opinion aggregation participation unit 115 of the participating user terminal 100' receives a spatial login instruction (step S108). Specifically, the opinion aggregation participation unit 115 receives a login instruction to the created discourse space via the input reception unit 140. The opinion aggregation participation unit 115 then transmits the spatial identifier of the discourse space, the user identifier, the permission slip, and the login information 121 necessary for connecting the participating user's user AI model 310 to the support device 200.

[0087] Then, the space provision unit 210 starts maintaining the connection from the participating user (step S109). Specifically, the space provision unit 210 connects to the user AI model 310 using the login information 121 received from the participating user terminal 100' in order to relay communication between the participating user terminal 100' and the user AI model 310. Then, the generated AI connection relay unit 220 relays the dialogue between the participating user terminal 100' and the user AI model 310 (step S110). The participating user and the user AI model 310 converse about the topic in question. In this conversation, the user's thoughts and ideas about the topic are revealed.

[0088] This process makes it possible to include the participation of participating users in the discourse space that is the target of opinion aggregation, that is, the dialogue with the user AI model 310 that took place during their participation, as the subject of opinion aggregation.

[0089] In this way, users who participate in the same discourse space with the same spatial identifier can use the conversational data from their interactions with the user AI model 310 during their participation as the subject of opinion aggregation.

[0090] Figure 5 shows an example of the opinion aggregation process flow. The opinion aggregation process is initiated at the user terminal 100 by a start command (opinion aggregation command) from the user.

[0091] First, the opinion aggregation instruction unit 114 receives an opinion aggregation instruction (step S201). Specifically, when the opinion aggregation instruction unit 114 receives the spatial identifier and the opinion aggregation instruction via the input reception unit 140, it transmits the spatial identifier and the opinion aggregation instruction to the support device 200.

[0092] Then, the support unit 230 of the support device 200 instructs the integrated AI model 320 to summarize the opinions of the participating users (step S202). At that time, the support unit 230 creates a user opinion aggregation prompt that includes the agenda (topic set in the discussion space) and the participating user identifiers (user identifiers of users participating in the discussion space), and instructs the integrated AI model 320 to do so. The user opinion aggregation prompt will be described later.

[0093] The integrated AI model 320, following instructions, identifies a user-specific AI model 310 for each user and instructs it to create an opinion summary with personal information masked (step S203). Specifically, following the instructions of the inter-user opinion aggregation prompt, the integrated AI model 320 identifies a user-specific AI model 310 for each user using the participating user identifier, removes privacy information from the conversation data, masks personal information, organizes the content of participants' statements, and instructs each user-specific AI model 310 to create an opinion summary that captures their perspectives. At this time, the integrated AI model 320 creates a prompt for each user-specific AI model 310 that includes a topic and a participating user identifier, and instructs each user-specific AI model 310 accordingly.

[0094] Then, each of the user AI models 310 follows instructions to mask personal information and summarize the user's opinion (step S204). Specifically, each of the user AI models 310 follows instructions to remove privacy information from conversation data, mask personal information, and create a text that summarizes the opinion of the participating user. Then, each of the user AI models 310 sends the participating user identifier and the summarized opinion to the integrated AI model 320.

[0095] Then, the integrated AI model 320 creates a compromise proposal among the participating users (step S205). Specifically, the integrated AI model 320 creates a compromise proposal among the participating users using the participating user identifiers and summarized opinions received from each of the user AI models 310. Then, the integrated AI model 320 transmits the compromise proposal among the participating users to the support device 200.

[0096] The support device 200 then presents the compromise proposal to the user terminal 100 (step S206), and the user terminal 100 displays the compromise proposal as the aggregated opinion (step S207). The support device 200 may also present the compromise proposal to other user terminals 100', and the other user terminals 100' may display the compromise proposal as the aggregated opinion.

[0097] The above is an example of the opinion aggregation process flow. According to this example of the opinion aggregation process flow, conversation data, including the content of conversations between users who participated in the opinion aggregation space and the generating AI service 300, can be used to organize the content and propose an opinion aggregation on the topic. In this type of opinion aggregation process, the statements of participating users that represent their opinions on the topic are used as input information from the conversation data, and the proposed aggregated opinion will include a conclusion on the topic.

[0098] Figure 6 shows an example of a user opinion aggregation prompt. Example 600 of the user opinion aggregation prompt includes a command 601 that uses the topic 602, participant attribute information 603 that identifies the participant's user identifier and the AI ​​being used, and constraint information 604 that includes masking personal information and creating a majority vote opinion if there is no confidence in the compromise, to query the AI ​​being used by the participant, organize the content of the participant's statements to understand their thinking, and propose a compromise (or a majority vote proposal or multiple prioritized proposals) among the participants' opinions. Example 600 of the user opinion aggregation prompt also includes an example output format 605.

[0099] The above is an example of an opinion aggregation support system 1 equipped with support functions that apply an embodiment according to one aspect of the present invention. According to the opinion aggregation support system 1 of this embodiment, it is possible to utilize LLM to support mutual understanding and conclusions in discussions among multiple users.

[0100] In the example of the opinion aggregation support system 1 equipped with the above-described embodiment, the integrated AI model 320 created a compromise (or majority-vote proposal) among the opinions of the participating users by utilizing information obtained by having each participating user's user-specific AI model 310 summarize the opinions of the participating users, but this is not limited to this. For example, the generation AI connection relay unit 220 relays the dialogue between the participating user terminal 100' and the user-specific AI model 310, and therefore maintains a history of dialogue information for each participating user. This can be used by the integrated AI model 320 to summarize the opinions of each participating user and create a compromise (or majority-vote proposal or multiple proposal) among the opinions of the participating users.

[0101] Figure 7 shows an example of the flow of opinion aggregation processing (using integrated AI) applied to such modified examples. Opinion aggregation processing (using integrated AI) is started at the user terminal 100 by a start instruction (opinion aggregation instruction) from the user.

[0102] First, the opinion aggregation instruction unit 114 receives an opinion aggregation instruction (step S301). Specifically, when the opinion aggregation instruction unit 114 receives the spatial identifier and the opinion aggregation instruction via the input reception unit 140, it transmits the spatial identifier and the opinion aggregation instruction to the support device 200.

[0103] Then, the support unit 230 of the support device 200 instructs the integrated AI model 320 to summarize opinions for each user (step S302). At that time, the support unit 230 creates a predetermined prompt for the integrated AI model 320 that includes the agenda (topic set in the discussion space), participating user identifiers (user identifiers of users participating in the discussion space), and conversation data, and instructs the integrated AI model 320 to create a summary of opinions for each user.

[0104] The integrated AI model 320, following instructions, performs opinion summaries for each user, masking personal information (step S303). Specifically, the integrated AI model 320 follows the instructions of the user opinion aggregation prompt (using integrated AI) and creates a document summarizing each user's opinion while masking personal information. The integrated AI model 320 then transmits the participating user identifier and the summarized opinion to the support device 200.

[0105] Then, the support unit 230 instructs the integrated AI model 320 to create a compromise proposal among the users (step S304). Specifically, the support unit 230 creates a user opinion aggregation prompt (using integrated AI) that includes the participating user identifiers and summarized opinions, and instructs the integrated AI model 320 to create a compromise proposal among the users. The user opinion aggregation prompt (using integrated AI) will be described later.

[0106] Then, the integrated AI model 320 creates a compromise among the participating users (step S305). Specifically, the integrated AI model 320 uses the participating user identifiers and the summarized opinions to create a compromise among the participating users (or a majority vote proposal or multiple proposals with priority). Then, the integrated AI model 320 transmits the compromise among the participating users to the support device 200.

[0107] Then, the support device 200 presents the compromise proposal to the user terminal 100 (step S306), and the user terminal 100 displays the compromise proposal as a summary opinion (step S307).

[0108] The above is an example of the opinion aggregation process (using integrated AI) flow. According to this example of the opinion aggregation process (using integrated AI) flow, conversation data including the content of conversations between users participating in the opinion aggregation space and the generating AI service 300 can be used to organize the content and propose opinion aggregation on a topic. In this type of opinion aggregation process, the integrated AI model 320 summarizes and aggregates the opinions, making it easier to generate logically consistent aggregated opinions.

[0109] Figure 8 shows an example of a user opinion aggregation prompt (using integrated AI). Example 700 of the user opinion aggregation prompt (using integrated AI) includes a command 701 that prompts participants to propose a compromise (or a majority-vote proposal or multiple prioritized proposals) using a topic 702, participant attribute information 703 that identifies the user identifier of each participant, opinions 704 for each participant, and constraint information 705 that includes creating a majority-vote opinion if there is no confidence in the compromise. Example 700 of the user opinion aggregation prompt (using integrated AI) also includes an example output format 706.

[0110] The above is a modified example in which the integrated AI model 320 summarizes the opinions of each participating user and creates a compromise (or a majority-vote proposal or multiple prioritized proposals) among the opinions of the participating users.

[0111] Figure 9 is an example of a block diagram of the grouping support system according to this embodiment. The grouping support system 1' basically has the same configuration as the opinion aggregation support system 1. The following will focus on the differences.

[0112] The grouping support system 1' includes a user terminal 170 (and participating user terminals 100' used by other users), a support device 250, and a generation AI service 300.

[0113] User terminal 170 is a terminal used by users of the grouping support system 1'. User terminal 170 is connected to the generation AI service 300 via network 50 for communication. User terminal 170 is also connected to the support device 250 via network 50 for communication.

[0114] The user terminal 170 is a so-called smartphone or personal computer. However, it is not limited to these; the user terminal 170 may also be various wearable devices such as smartwatches, smart glasses, or wireless earphones, or various information processing devices such as PDAs (Personal Data Assistants) or tablet devices.

[0115] The user terminal 170 receives input from the user via input devices such as voice, text, and other screen operations, and communicates appropriately with the generation AI service 300 or support device 200 according to the input.

[0116] Furthermore, the user terminal 170 generates a population on the support device 250 via the network 50, allows other users to join the population using the participating user terminal 100', and instructs the support device 250 to group the participants within the population to obtain grouping proposals. The population is created in units similar to chat rooms, and conversations with the generation AI service 300 while participating in the population are captured as the opinions of the users in the population and shared together with the opinions of other participants.

[0117] The user terminal 170 comprises a processing unit 110', a storage unit 120, a communication unit 130, an input receiving unit 140, and an output unit 150. The processing unit 110' includes a dialogue processing unit 111, a generation AI connection processing unit 112, a population creation unit 116, a grouping instruction unit 117, and a population participation unit 118.

[0118] The population creation unit 116 creates a population, which is a space for sharing conversation data, including the content of conversations between users and the generative AI service 300 that uses a large-scale language model (LLM).

[0119] The group assignment instruction unit 117 instructs the creation of group assignment plans based on opinions expressed in conversation data shared within a predetermined population.

[0120] The population participation unit 118 includes a user and an AI model 310 associated with that user in a predetermined population.

[0121] The support device 250 is a so-called server device. However, it is not limited to this, and the support device 250 may be any kind of information processing device, such as a personal computer, smartphone, workstation, PDA (Personal Data Assistant), or tablet device.

[0122] When the support device 250 receives a request for population creation from a user terminal 170 via the network 50, it appropriately creates a population in accordance with the request. Furthermore, when the support device 250 receives a participation request from a participating user terminal 100' via the network 50, it allows the terminal to join the population in accordance with the request. Additionally, when the support device 250 receives a request for grouping regarding a conversation between a user terminal 170 that has joined the population and the generation AI service 300, it creates an appropriate grouping in accordance with the request.

[0123] In this process, the support device 250 delegates the sorting process to the generation AI service 300 via the network 50.

[0124] The support device 250 includes a space provision unit 210, a generation AI connection relay unit 220, a support unit 230', and a communication unit 240. The space provision unit 210 defines a population targeting users who have expressed their intention to participate. The space provision unit 210 creates a population upon request and manages users who participate in the population.

[0125] The generation AI connection relay unit 220 stores conversation data, including the content of conversations between users and a predetermined large-scale language model (LLM), within a population related to a predetermined topic. Multiple methods can be employed for acquiring conversation data. For example, one method is to log the conversation data of users participating in the population in real time and store the conversation data in the support device 250, or to store the conversation data with users in a user AI model 310 associated with users participating in the population.

[0126] The support unit 230' instructs the generation AI service 300 to organize the content of the user input information, i.e., the utterance information, from the conversation data acquired about users participating in the population, and to propose grouping them according to topic.

[0127] The communication unit 240 controls communication (wired or wireless) between the support device 250 and other devices via the network 50.

[0128] By using the grouping support system 1', for example, a user can share a common topic with other participants, such as a predetermined agenda, and automatically record their thoughts and ideas in response to their utterances during a conversation with the generating AI service 300 via voice chat using their smartphone, which is the user terminal 170. The generating AI service 300 can then aggregate the thoughts and ideas automatically recorded in response to the utterances of other participants during their conversations with the generating AI service 300 and propose grouping plans.

[0129] When the generating AI service 300 receives a topic from the user terminal 170, such as spoken audio or text about forming a working team, it takes into account the input information provided up to that point, creates a natural language response using the audio or text information, and sends it back to the user terminal 170. For example, if the topic is forming a working team, the generating AI service 300 generates responses in natural language that continue the conversation in various ways, such as agreeing, delving deeper, or giving advice, and sends them back to the user terminal 170.

[0130] Next, the operation of the grouping support system 1' in this embodiment will be described.

[0131] First, in this embodiment, a user of the grouping support system 1' uses the user terminal 170 to generate a shared population on the support device 250 with other users, and the users participating in the population form the grouping population. Each user belonging to the population then freely engages in conversation with the generation AI service 300. After a predetermined period (for example, 3 days, 1 week, 1 month, or longer), the support device 250 makes grouping suggestions among the users participating in the population based on the conversation content during that period.

[0132] This process includes two distinct processes: a population creation process in which the user terminal 170 requests the support device 250 to create a population on the support device 250 for grouping with other users on a predetermined topic (agenda), and allows the user and other users' user terminals 100' to participate in the said population on the support device 250; and a grouping process in which the content is organized using conversation data, including the conversation content between the users who have participated in the said population and the generating AI service 300, and a grouping proposal is made for grouping on a topic.

[0133] Figure 10 shows an example of the population creation process flow. The population creation process is started on the user terminal 170 by a start command from the user.

[0134] First, the population creation unit 116 receives the grouping purpose and the population creation instruction (step S401). Specifically, when the population creation unit 116 receives the grouping purpose (which can also be called a topic) and the population creation instruction via the input reception unit 140, it transmits the grouping purpose and the population creation instruction to the support device 250.

[0135] The space provision unit 210 of the support device 250 creates a population (step S402). Specifically, the space provision unit 210 creates a population having a predetermined spatial identifier, associates the received grouping purpose as topic information of the population, and associates the user terminal 170 as the owner. Then, the space provision unit 210 transmits the spatial identifier of the created population to the user terminal 170. For example, the grouping purpose may relate to the relationships between multiple users.

[0136] Then, the population creation unit 116 accepts the designation of the invited users (step S403). Specifically, when the population creation unit 116 receives the user identifiers of one or more invited users via the input reception unit 140, it transmits the spatial identifier received in step S402 and one or more user identifiers to the support device 250.

[0137] Then, the space provision unit 210 issues a permit for the population (step S404). Specifically, the space provision unit 210 generates unique information such as a password, passkey, random key, or QR code (registered trademark) that is different or common to each invited user, and registers it as a permit for the population. Then, the space provision unit 210 transmits the created permit for the population to the user terminal 170.

[0138] Then, the population participation unit 118 receives a population login instruction (step S405). Specifically, the population participation unit 118 receives a login instruction to the created population via the input reception unit 140. The population participation unit 118 then transmits the spatial identifier of the created population, the user identifier, the authorization card, and the login information 121 necessary for connecting the user's AI model 310 to the support device 250.

[0139] Then, the space provision unit 210 starts maintaining the connection from the user (step S406). Specifically, the space provision unit 210 connects to the user AI model 310 using the login information 121 received from the user terminal 170 in order to relay communication between the user terminal 170 and the user AI model 310. Then, the generation AI connection relay unit 220 relays the dialogue between the user terminal 170 and the user AI model 310 (step S407). The user and the user AI model 310 converse on any topic or a specific topic. For example, it could be an everyday conversation or a conversation about the purpose of grouping. The user's thoughts and ideas are revealed in this conversation.

[0140] The above is an example of the population creation process flow. According to this example of the population creation process flow, the user can create a population to which they will be grouped, and their participation in that population, that is, the interaction with the user's AI model 310 during their participation, can be used as a basis for making decisions regarding grouping.

[0141] Furthermore, other participating users, using their participating user terminal 100', receive a spatial identifier and a permission slip from their user terminal 170 by some means, for example, via an SNS (Social Network Service) using relation information 122, or in person. After receiving these, the population participation unit 118 of the participating user terminal 100' receives a population login instruction (step S408). Specifically, the population participation unit 118 receives a login instruction to the created population via the input reception unit 140. The population participation unit 118 then transmits the population spatial identifier, user identifier, permission slip, and login information 121 necessary for connecting the participating user's user AI model 310 to the support device 250.

[0142] Then, the space provision unit 210 starts maintaining the connection from the participating user (step S409). Specifically, the space provision unit 210 connects to the user AI model 310 using the login information 121 received from the participating user terminal 100' in order to relay communication between the participating user terminal 100' and the user AI model 310. Then, the generation AI connection relay unit 220 relays the dialogue between the participating user terminal 100' and the user AI model 310 (step S410). The participating user and the user AI model 310 converse on any topic or a specific topic. For example, it may be an everyday conversation or a conversation about the purpose of grouping. The user's thoughts and ideas are revealed in this conversation.

[0143] This process allows participating users to use their participation in the target population for grouping, specifically their interactions with the user-facing AI model 310 during their participation, as a basis for determining grouping.

[0144] In this way, users who have joined the same population with the same spatial identifier can use the conversational data from their interactions with the user AI model 310 during their participation as a basis for determining their grouping.

[0145] Figure 11 shows an example of the sorting process flow. The sorting process is initiated by a start command (sorting command) from the user on the user terminal 170.

[0146] First, the grouping instruction unit 117 receives a grouping instruction (step S501). Specifically, when the grouping instruction unit 117 receives the spatial identifier and grouping instruction via the input receiving unit 140, it transmits the spatial identifier and grouping instruction to the support device 200.

[0147] Then, the support unit 230' of the support device 200 instructs the integrated AI model 320 to calculate the characteristics of each participating user's speech (step S502). At that time, the support unit 230' creates a grouping prompt for the integrated AI model 320 that includes the grouping purpose (topic set in the population) and the participating user identifier (user identifier of the user participating in the population), and instructs the integrated AI model 320. The grouping prompt will be described later.

[0148] The integrated AI model 320, following instructions, identifies a user-specific AI model 310 for each user and instructs it to calculate speech features (step S503). Specifically, following the instructions of the grouping prompt, the integrated AI model 320 identifies a user-specific AI model 310 for each user using the participating user identifier, and instructs the user-specific AI model 310 to calculate features based on information that has been organized from the content of the speech after removing privacy information from the conversation data. At this time, the integrated AI model 320 creates a prompt for each user-specific AI model 310 that includes a topic and a participating user identifier, and instructs each user-specific AI model 310 accordingly.

[0149] Then, each of the user AI models 310 calculates the characteristics of the user's utterances according to the instructions (step S504). Specifically, each of the user AI models 310 removes privacy information from the conversation data according to the instructions and obtains vector information based on the text that verbalizes the content of the participating users' utterances (for example, opinions on the grouping of multiple users) using existing technologies such as Word to Vec, thereby obtaining a vector that represents the characteristics of the utterances. Then, each of the user AI models 310 transmits the participating user identifier and the vector representing the characteristics of the utterances to the integrated AI model 320.

[0150] Then, the integrated AI model 320 groups users according to the purpose based on the characteristics of their speech (step S505). Specifically, the integrated AI model 320 creates a grouping proposal among participating users using the participating user identifiers and vectors received from each of the user AI models 310. This grouping proposal can be determined by the integrated AI model 320 by, for example, clustering the participating users until a suitable number of groups is obtained.

[0151] Alternatively, if each group suited to the purpose possesses unique characteristics (for example, a group of sales representatives and a group of customers in a role-playing exercise), the integrated AI model 320 may assign participating users to groups that closely match those unique characteristics, i.e., groups with a close vector distance. Alternatively, the AI ​​model 320 may assign participating users to groups so that they are evenly distributed across all groups, preventing a concentration of participants with a close vector distance in a particular group. The algorithm used in determining these group assignments may be determined by the integrated AI model 320 according to the purpose of the group assignment.

[0152] The integrated AI model 320 then transmits the proposed grouping of participating users to the support device 250.

[0153] Then, the support unit 230' of the support device 250 presents the grouping plan to the user terminal 170 (step S506), and the user terminal 170 displays the grouping plan (step S507).

[0154] The above is an example of the grouping process flow. According to this example of the grouping process flow, conversation data, including the content of conversations between users who participated in the population and the generating AI service 300, can be used to organize the content and propose groupings according to the topic. In this type of grouping process, the statements of participating users that represent their opinions on the topic are used as input information from the conversation data, and the proposed aggregated opinions will include grouping suggestions related to the topic.

[0155] Figure 12 shows an example of a grouping prompt. Example grouping prompt 800 includes a command 801 that uses the topic, which is the grouping objective 802, participant attribute information 803 that identifies the participant's user identifier and the AI ​​being used, and constraint information 804 that includes creating groups with highly similar members if there is no confidence in the proposed grouping or if there is no suitable grouping, to query the AI ​​being used by the participant, organize the content of the participant's statements to understand their thinking, and propose a grouping plan among the participants that suits the objective. Example grouping prompt 800 also includes an example output format 805.

[0156] The above is an example of a grouping support system 1' equipped with support functions that apply an embodiment according to one aspect of the present invention. According to the grouping support system 1' of this embodiment, it is possible to utilize LLM to support matching between multiple users.

[0157] In the example of the grouping support system 1' equipped with the support function to which the above embodiment is applied, the integrated AI model 320 causes the user AI model 310 used by each participating user to calculate a vector based on the participating user's utterance data, and uses this to create a grouping plan for the participating users, but this is not limited to this. For example, the generation AI connection relay unit 220 relays the dialogue between the participating user terminal 100' and the user AI model 310, and therefore maintains a history of dialogue information for each participating user. Using this, the integrated AI model 320 may generate a vector based on the utterances of each participating user and create a grouping plan for the participating users.

[0158] Figure 13 shows an example of the sorting process (using integrated AI) flow applicable to this modified example. The sorting process (using integrated AI) is started on the user terminal 170 by a start command (sorting command) from the user.

[0159] First, the grouping instruction unit 117 receives a grouping instruction (step S601). Specifically, when the grouping instruction unit 117 receives the spatial identifier and grouping instruction via the input receiving unit 140, it transmits the spatial identifier and grouping instruction to the support device 250.

[0160] Then, the support unit 230' of the support device 250 instructs the integrated AI model 320 to calculate the characteristics of each user's speech (step S602). At that time, the support unit 230' creates a predetermined prompt for the integrated AI model 320 that includes a participating user identifier (user identifier of a user participating in the population) and conversation data, and instructs the integrated AI model 320 to calculate the characteristics of each user's speech.

[0161] The integrated AI model 320 calculates the characteristics of each participant's speech according to the instructions (step S603). Specifically, the integrated AI model 320 removes privacy information from the conversation data according to the instructions and obtains vector information based on the text of the participant's speech (for example, their opinion on grouping multiple users) using existing technologies such as Word to Vec, thereby obtaining a vector that represents the characteristics of the speech. The integrated AI model 320 then transmits the participant user identifier and the vector representing the characteristics of the speech to the support device 250.

[0162] Then, the support unit 230' instructs the integrated AI model 320 to group users according to the purpose (step S604). Specifically, the support unit 230' creates a grouping prompt (using integrated AI) that includes the grouping purpose, participating user identifiers, and vectors, and instructs the integrated AI model 320 to create a user grouping proposal. The grouping prompt (using integrated AI) will be described later.

[0163] Then, the integrated AI model 320 groups users according to the purpose based on the characteristics of their statements (step S605). Specifically, the integrated AI model 320 uses participant user identifiers and vectors to create proposed groupings among participating users according to the purpose of the grouping. This proposed grouping can be determined by the integrated AI model 320 by clustering the participating users until a suitable number of groups is obtained.

[0164] Alternatively, if each group suited to the purpose possesses unique characteristics (for example, a group of sales representatives and a group of customers in a role-playing exercise), the integrated AI model 320 may assign participating users to groups that closely match those unique characteristics, i.e., groups with a close vector distance. Alternatively, the AI ​​model 320 may assign participating users to groups so that they are evenly distributed across all groups, preventing a concentration of participants with a close vector distance in a particular group. The algorithm used in determining these group assignments may be determined by the integrated AI model 320 according to the purpose of the group assignment.

[0165] The integrated AI model 320 then transmits the proposed grouping of participating users to the support device 250.

[0166] Then, the support unit 230' of the support device 250 presents the grouping plan to the user terminal 170 (step S606), and the user terminal 170 displays the grouping plan (step S607).

[0167] The above is an example of the grouping process (using integrated AI) flow. According to this example of the grouping process (using integrated AI) flow, conversation data, including the content of conversations between users who participated in the population and the generating AI service 300, can be used to organize the content and propose groupings according to the topic. In this type of grouping process, the statements of participating users that represent their opinions on the topic are used as input information from the conversation data, and the proposed aggregated opinions will include grouping proposals related to the topic.

[0168] Figure 14 shows an example of a grouping prompt (using integrated AI). Example 900 of the grouping prompt (using integrated AI) includes a command 901 that prompts the system to propose a grouping plan among participants to suit the purpose, using the topic, which is the grouping purpose 902, participant attribute information 903 that identifies the user identifiers and utterance feature vectors of the participants, and constraint information 904 that includes creating groups with highly similar members if the system is unsure of the proposed grouping plan or if there is no suitable grouping. Example 900 of the grouping prompt (using integrated AI) also includes an example 906 of the output format.

[0169] The above is a modified example in which the integrated AI model 320 determines the characteristics of each participating user's statements and creates a proposed grouping of participating users.

[0170] The present invention is not limited to the embodiments described above. The embodiments described above can be modified in various ways within the scope of the technical idea of ​​the present invention. For example, in the embodiments described above, the generative AI service 300 includes a generative AI that uses a large-scale language model (LLM), but it is not limited to this and may be created using other modeling methods.

[0171] Furthermore, the technical elements of the embodiments described above may be applied individually, or they may be divided into multiple parts, such as program components and hardware components, and applied accordingly.

[0172] The present invention has been described above, focusing on its embodiments. [Explanation of Symbols]

[0173] 1···Opinion aggregation support system, 1'···Grouping support system, 50···Network, 100,170···User terminals, 110,110'···Processing unit, 111···Dialogue processing unit, 112···Generating AI connection processing unit, 113···Opinion aggregation space creation unit, 114···Opinion aggregation instruction unit, 115···Opinion aggregation participation unit, 116···Population creation unit, 117···Grouping instruction unit, 118···Population participation unit, 120· ...Memory unit, 121...Login information, 122...Relation information, 130...Communication unit, 140...Input reception unit, 150...Output unit, 200, 250...Support devices, 210...Space provision unit, 220...Generating AI connection relay unit, 230, 230'...Support unit, 240...Communication unit, 300...Generating AI service, 310...User AI model, 320...Integrated AI model, 330...Communication unit.

Claims

1. A generation AI connection relay unit that stores conversation data including the content of conversations between a user and a predetermined large-scale language model (LLM) on a predetermined topic, A support unit instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, A communication unit that transmits the aforementioned proposal to a designated terminal, Equipped with, The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. The support unit issues the instruction to the integrated module, The instructions include causing the individual module to organize the content using the conversation data with the user, A support device characterized by the following features.

2. A generation AI connection relay unit that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic, A support unit instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, A communication unit that transmits the aforementioned proposal to a designated terminal, Equipped with, The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. The support unit issues the instruction to the integrated module, The instructions include causing the integration module to organize the content using the conversation data, A support device characterized by the following features.

3. A support device according to claim 1 or 2, Of the aforementioned conversation data, the aforementioned topic is a topic relating to a predetermined agenda, The user input information is an opinion on the aforementioned agenda item, The aforementioned proposal includes conclusions on the aforementioned agenda, A support device characterized by the following features.

4. A support device according to claim 1 or 2, The topic among the aforementioned conversation data is a topic relating to the relationships between multiple users. The user input information is an opinion on the grouping of multiple users. The aforementioned proposal includes the presentation of the aforementioned groupings, A support device characterized by the following features.

5. A support device according to claim 1 or 2, The support unit instructs the large-scale language model to remove privacy information from the conversation data, organize the content, and make suggestions. A support device characterized by the following features.

6. A support device according to claim 1 or 2, The support unit instructs the large-scale language model to organize the content and make multiple proposals. A support device characterized by the following features.

7. A support device according to claim 1 or 2, The support unit instructs the large-scale language model to organize the content, prioritize it, and make multiple proposals. A support device characterized by the following features.

8. A support device according to claim 1 or 2, The support unit instructs the large-scale language model to organize and propose content that conforms to majority rule, using input information from multiple users from the accumulated conversation data. A support device characterized by the following features.

9. The support device according to claim 4, It includes a space provision unit that defines the population of users who have expressed their intention to participate, The user input information is an opinion on the grouping of multiple users included in the population. The proposal includes presenting the grouping of the multiple users included in the population, A support device characterized by the following features.

10. In the processor, A generative AI connection relay process that stores conversation data including the content of conversations between a user and a predetermined large-scale language model (LLM) on a predetermined topic, Support processing that instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, Communication processing to transmit the above proposal to a predetermined terminal, They will implement this, The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. In the support process, the instruction is given to the integrated module, The instructions include causing the individual module to organize the content using the conversation data with the user, A support method characterized by the following features.

11. A generative AI connection relay step that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic, A support step instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, A display step that displays the aforementioned proposal, We will implement the following: The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. In the support step, the instruction is given to the integrated module, The instructions include causing the individual module to organize the content using the conversation data with the user, A support system characterized by the following features.

12. In the processor, A generative AI connection relay step that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic, A support step instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, A communication step of transmitting the above proposal to a predetermined terminal, They will implement this, The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. In the support step, the instruction is given to the integrated module, The instructions include causing the individual module to organize the content using the conversation data with the user, A support program characterized by the following features.

13. The processor includes: A generative AI connection relay process that stores conversation data including the content of conversations between a user and a predetermined large-scale language model (LLM) on a predetermined topic, Support processing that instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, Communication processing to transmit the above proposal to a predetermined terminal, They will implement this, The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. In the support process, the instruction is given to the integrated module, The instructions include causing the integration module to organize the content using the conversation data, A support method characterized by the following features.

14. A generation AI connection relay step that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic, A support step instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, A display step that displays the aforementioned proposal, We will implement the following: The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. In the support step, the instruction is given to the integrated module, The instructions include causing the integration module to organize the content using the conversation data, A support system characterized by the following features.

15. The processor includes: A generative AI connection relay step that stores conversation data including the content of a conversation between a user and a predetermined large-scale language model (LLM) on a predetermined topic, A support step instructs the large-scale language model to organize the content and make suggestions using the user's input information from the accumulated conversation data, A communication step of transmitting the above proposal to a predetermined terminal, They will implement this, The large-scale language model includes individual modules corresponding to each user and an integrated module common to all users. The aforementioned conversation data is conversation data between the user and the corresponding individual module. In the support step, the instruction is given to the integrated module, The instructions include causing the integration module to organize the content using the conversation data, A support program characterized by the following features.

Images