System, program and information processing method

Abstract

To provide a system, program and information processing method which are capable of utilizing dialogue information of chats.SOLUTION: A system includes acquisition means 20, model generation means 38, and answer generation means 40. The acquisition means 20 acquires dialogue information in a chat executed by a user terminal 42. The model generation means 38 generates a trained model LLM34 that takes questions as input and provides answers as output by performing machine learning using the acquired dialogue information as training data. The answer generation means 40 generates an answer to an instruction sentence from a client terminal 46 on the basis of the generated trained model LLM 34 and transmits the generated answer to the client terminal 46.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a system, a program, and an information processing method. [Background technology]

[0002] In recent years, there has been an increase in AI chatbot services in which AI (artificial intelligence) answers questions from users. For example, Patent Document 1 discloses a dialogue system that uses AI technology to answer questions from users in customer support situations. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6218057 Summary of the Invention [Problem to be solved by the invention]

[0004] It is desirable to utilize chatbot dialogue information conducted on user terminals and chat dialogue information between multiple user terminals.

[0005] The present disclosure has been made in consideration of the above points, and aims to provide a system, a program, and an information processing method that can utilize chat dialogue information. [Means for solving the problem]

[0006] The system of the present disclosure comprises: A system comprising an acquisition means, a model generation means, and an answer generation means, the acquiring means acquires conversation information in a chat being executed on the user terminal; the model generation means performs machine learning using the acquired dialogue information as training data to generate a trained model in which a question is input and an answer is output; The answer generation means is characterized in that it transmits an answer generated based on the generated trained model to the client terminal.

[0007] In the system of the present disclosure, The answer generation means may generate an answer based on the trained model in response to an instruction from the client terminal, and transmit the generated answer to the client terminal.

[0008] The system of the present disclosure comprises: further comprising authorization means; the authorization means grants authorization to a predetermined client terminal; The answer generation means may transmit an answer generated based on the generated trained model to the client terminal to which authority has been granted by the authorization granting means.

[0009] The system of the present disclosure comprises: further comprising authorization means; the authorization means grants authorization to a predetermined client terminal; The answer generation means may generate an answer based on the generated trained model in response to an instruction statement from the client terminal to which authority has been granted by the authorization granting means, and transmit the generated answer to the client terminal.

[0010] The system of the present disclosure comprises: Further comprising a user registration means, the user registration means registers users who are to carry out chats at the user terminals; the model generation means generates the trained model for each user registered by the user registration means, The answer generation means may transmit an answer generated based on the trained model generated for each user to the client terminal.

[0011] The system of the present disclosure comprises: Further comprising a user registration means, the user registration means registers users who are to carry out chats at the user terminals; the model generation means generates the trained model for each user registered by the user registration means, The answer generation means may generate an answer based on the trained model generated for each user in response to an instruction sentence for each user from the client terminal, and transmit the generated answer to the client terminal.

[0012] The system of the present disclosure comprises: Further comprising a user registration means, the user registration means registers a user who will execute a chat on the user terminal, and sets attributes of the user at this time; the model generation means generates the trained model for each user attribute, The answer generation means may transmit an answer generated based on the trained model for each generated user attribute to the client terminal.

[0013] The system of the present disclosure comprises: Further comprising a user registration means, the user registration means registers a user who will execute a chat on the user terminal, and sets attributes of the user at this time; the model generation means generates the trained model for each user attribute, The answer generation means may generate an answer based on the trained model generated for each user attribute in response to an instruction sentence for each user attribute from the client terminal, and transmit the generated answer to the client terminal.

[0014] The system of the present disclosure comprises: the user registration means registers a user who will execute a chat on the user terminal, and sets attributes of the user at this time; The model generation means may generate a third trained model that takes a question as input and an answer as output by performing machine learning using dialogue information of a specific user and dialogue information of a user having a specific attribute as training data.

[0015] The system of the present disclosure comprises: The user registration means sets attributes of the user, The model generation means may generate a third trained model that inputs a question and outputs an answer by performing machine learning on the trained model for each user using dialogue information of a user with specific attributes as training data.

[0016] In the system of the present disclosure, The model generation means may generate a third trained model that inputs a question and outputs an answer by performing machine learning on the trained model for each user attribute using dialogue information of a specific user as training data.

[0017] The system of the present disclosure comprises: further comprising a permission obtaining means and an authorization means; the permission acquisition means transmits to the user terminal a confirmation command that confirms that the user registered by the user registration means is authorized to use the trained model by a specific client, and then receives permission information from the user terminal to acquire the permission of the user who operates the user terminal; the authorization means grants the authorization to use the trained model corresponding to the user to the client terminal of the client for which user permission has been obtained; The answer generation means may generate an answer based on the generated trained model in response to an instruction statement from the client terminal to which authority has been granted by the authorization granting means, and transmit the generated answer to the client terminal.

[0018] The system of the present disclosure comprises: further comprising a transmitting means; The acquisition means acquires information about a specific question from an administrator terminal, the transmission means transmits the acquired specific question to the user terminal, thereby outputting the specific question to a chat interface of the user terminal; The dialogue information acquired by the acquisition means may include a specific question and a user's answer to the specific question.

[0019] The system of the present disclosure comprises: further comprising a transmitting means; when the acquisition means acquires a question from a first user terminal, the transmission means transmits the question acquired from the first user terminal to a second user terminal, thereby allowing a chat to take place between the first user terminal and the second user terminal; When the acquisition means acquires dialogue information in a chat held between the first user terminal and the second user terminal, the model generation means may perform machine learning using the dialogue information acquired by the acquisition means as training data.

[0020] In the system of the present disclosure, The chat may be a chat with a chatbot.

[0021] In the system of the present disclosure, The chatbot may be tuned to output specific questions as dialogue information.

[0022] The system of the present disclosure comprises: Further comprising a reward granting means, The reward granting means may grant a reward to a user who provides dialogue information in a chat executed on the user terminal.

[0023] The system of the present disclosure comprises: Further comprising an evaluation inquiry means, the evaluation inquiry means transmits a notification to the second user terminal requesting an evaluation of dialogue information in a chat executed on the first user terminal; The model generation means may perform machine learning by also referring to an evaluation acquired from the second user terminal.

[0024] In the system of the present disclosure, Further comprising an evaluation inquiry means, The evaluation inquiry means sends a notification to the second user terminal requesting an evaluation of the answer of the trained model, The model generation means may perform machine learning by also referring to an evaluation acquired from the second user terminal.

[0025] In the system of the present disclosure, Further comprising an evaluation inquiry means, The evaluation inquiry means sends a notification to the second user terminal requesting an evaluation of the trained model's response to a command from the client terminal, The model generation means may perform machine learning by also referring to an evaluation acquired from the second user terminal.

[0026] The system of the present disclosure comprises: Further comprising a reward granting means, the reward granting means grants a reward to a user who provides dialogue information in a chat executed on the user terminal; The reward given by the reward giving means may be determined based on an evaluation of dialogue information in a chat executed at the first user terminal, the evaluation information being acquired from the second user terminal.

[0027] The program of the present disclosure is A program that causes one or more computers to function as an acquisition means, a model generation means, and an answer generation means, the acquiring means acquires conversation information in a chat being executed on the user terminal; the model generation means performs machine learning using the acquired dialogue information as training data to generate a trained model in which a question is input and an answer is output; The answer generation means is characterized in that it transmits an answer generated based on the generated trained model to the client terminal.

[0028] The information processing method of the present disclosure includes: 1. An information processing method executed by one or more computers, comprising: a step of the computer acquiring conversation information in a chat being executed on a user terminal; a step of causing the computer to perform machine learning using the acquired dialogue information as training data, thereby generating a trained model in which a question is input and an answer is output; A step in which the computer transmits an answer generated based on the generated trained model to a client terminal; The present invention is characterized by the following. [Effects of the Invention]

[0029] According to the system, program, and information processing method of the present disclosure, chat dialogue information can be utilized. [Brief explanation of the drawings]

[0030] [Figure 1] 1A is a diagram showing a schematic configuration of an exemplary information processing system according to an embodiment of the present disclosure, and FIG. 1B is a diagram showing an example of a trained model in a second large-scale language model server. [Figure 2] FIG. 10 is a diagram illustrating an example of the flow of information processing during a chat between a user and a chatbot in an information processing system and an information processing method according to an embodiment of the present disclosure. [Figure 3] 1 is a diagram illustrating an example of the flow of information processing during a conversation between users in an information processing system and an information processing method according to an embodiment of the present disclosure. [Figure 4] FIG. 10 is a diagram illustrating an example of the flow of information processing when a question is posed to a user in an information processing system and an information processing method according to an embodiment of the present disclosure. [Figure 5] 1 is a diagram illustrating an example of the flow of information processing when a client terminal acquires information in an information processing system and an information processing method according to an embodiment of the present disclosure. [Figure 6] FIG. 1 is a diagram illustrating an example of the flow of information processing in an information processing method according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0031]

[0023] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings, but the invention according to the present disclosure is not limited thereto. Figures 1 to 6 are diagrams showing an information processing system 1 and an information processing method according to the present embodiment.

[0032] [Information Processing System 1] The information processing system 1 according to the present disclosure shown in FIG. 1 includes a server (computer) 10, a user terminal 42, an administrator terminal 44, a first large-scale language model server (calculation unit) 32, a second large-scale language model server (calculation unit) 34, and a client terminal 46. The information processing system 1 may include a plurality of each component. The server 10 is an administrative server that can be used via a communication network (not shown) such as the Internet. User terminals 42 and the like are communicatively connected to the server 10 via the communication network. The information processing system 1 provides a platform that enables dialogue between users, dialogue between users and chatbots, and dialogue between clients and chatbots (FIGS. 2 to 5). Each component of the information processing system 1 will be described below.

[0033] A "chatbot" refers to a program or application that interacts with people using text, voice, etc., and preferably utilizes a large-scale language model. A "user" refers to a person who interacts with a chatbot (e.g., the trained model LLM32 described below) using the server 10 platform and provides the interaction information. An "administrator" refers to a person (including a corporation) who manages the server 10. A "client" refers to a person who interacts with a chatbot (e.g., the trained model LLM34 described below) using the server 10 platform and utilizes the user's interaction information. Typical client businesses and operations include survey creation, product development, travel planning, matchmaking / matching services, advertising, and sales. In this way, trained models such as LLM34 can be used for marketing and recommendation systems.

[0034] <Server (Computer) 10> The configuration of a server (computer) 10 according to the present disclosure will be described with reference to Fig. 1(A). Fig. 1(A) is a schematic diagram illustrating the configuration of the server 10 according to the present disclosure. The server 10 according to this embodiment is configured from an industrial computer, a personal computer, or the like, and includes a control unit 12, a storage unit 28, and a communication unit 30, as shown in Fig. 1(A).

[0035] (Control unit 12) The control unit 12 is configured with a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc., and controls the operation of the server 10. Specifically, the control unit 12 executes programs stored in a storage unit 28 (described later) to function as a user registration means 14, a permission acquisition means 16, an authorization means 18, an acquisition means 20, a transmission means 22, an evaluation inquiry means 24, a reward granting means 26, etc. In the embodiment shown in FIG. 1(A), dialogue information (FIGS. 2 to 4, "question and answer sets") is transmitted from the server 10 to the second large-scale language model server 34, and the control unit 36 ​​of the second large-scale language model server 34 functions as a model generation means 38 and an answer generation means 40. However, the present disclosure is not limited to this example, and the control unit 12 may also function as the model generation means 38 and the answer generation means 40. Each of these means will be described later. Note that these functions may be realized by executing one or more independent programs or applications. Furthermore, these programs and applications may be provided on a single terminal (server 10) or distributed across multiple terminals (including server 10 and second large-scale language model server 34), and in the latter case, may be interconnected via wired cables or a communication network.

[0036] (Storage unit 28) The storage unit 28 is configured, for example, with an HDD (Hard Disk Drive), RAM (Random Access Memory), ROM (Read Only Memory), SSD (Solid State Drive), etc. Furthermore, the storage unit 28 is not limited to being built into the server 10, but may be a storage medium (for example, a USB memory) that can be detachably attached to the server 10. Furthermore, instead of providing the storage unit 28, various pieces of information may be stored in other storage means (such as a cloud server).

[0037] The storage unit 28 is capable of storing, in addition to the programs executed by the control unit 12, user information such as registration information and attributes for each user's account, client information such as registration information for each client's account and whether or not the client has permission to use the trained model, information on the character (avatar) that the user interacts with in the chatbot, dialogue information, prompts, etc. In this specification, the term "character" refers to a unique real person or deceased person, such as a celebrity, famous person, or ordinary person, who appears on the platform (chatbot) of the server 10, or their personality, character, personality, etc., and is a separate concept from a user. The user information may further be associated with trained models 34a, 34b, and 34c (FIG. 1(B)) corresponding to each user and license information.

[0038] "Dialogue information" includes question information, answer information, etc., and may also include audio information and text information of these. The latest question information from the user terminal 42 can also be stored in the storage unit 28 at any time. In this specification, unless otherwise specified, "question information" refers to dialogue information from the user terminal 42 or the client terminal 46, and "answer information" refers to dialogue information from a character (a chatbot; in the example of FIG. 2, a trained model in the first large-scale language model server 32). For example, even if the content of the dialogue information from a character is a question to the user, it is referred to as answer information for convenience.

[0039] In this specification, the term "attribute" is used in its usual sense to refer to characteristics shared by multiple people. Examples include, but are not limited to, age (generation), gender, blood type, family structure, prefecture of residence, place of origin, occupation, hobbies and interests (including ongoing interests such as extracurricular activities and qualification courses, as well as temporary interests such as news scandals that people want to exchange opinions about with others or chatbots), and personality test results. Although "attribute" is a concept that groups multiple people, if differences of opinion are observed, attributes can be further subdivided to identify commonalities. For example, reading as a hobby can be categorized into genres such as fiction and nonfiction. Furthermore, if users' preferences for a particular sport are divided into groups A and B, attributes can be subdivided, or an attribute related to a player X who is particularly favored within group A can be further established. User information including such attributes can be extracted from conversations, but can also be registered by the user at any time without conversation. It is preferable to store each attribute information in association with a user, as this allows for efficient extraction of data (users and their corresponding individual models 34a).

[0040] (Communication unit 30) The communication unit 30 connects the control unit 12 to an external device (e.g., a user terminal 42) wirelessly or via a wired connection so that the control unit 12 can communicate with the external device. The control unit 12 is capable of transmitting and receiving signals to and from the external device via the communication unit 30.

[0041] (others) In addition to the administrator terminal 44, the server 10 may also be provided with an operation unit (input unit) such as a keyboard that allows the administrator of the server 10 to give various commands to the control unit 12, and a display unit such as a monitor that displays various screens in response to display command signals from the control unit 12. In one embodiment, a display operation unit such as a touch panel that integrates the operation unit and display unit may be used. Furthermore, the server 10 may also be provided with a calculation unit corresponding to the trained models of the first large-scale language model server 32 and the second large-scale language model server 34.

[0042] <First large-scale language model server 32, second large-scale language model server 34> The first large-scale language model server 32 and the second large-scale language model server 34 are servers such as industrial computers. Each of the first large-scale language model server 32 and the second large-scale language model server 34 includes a trained model that receives a question sentence as input and outputs a response sentence, and preferably includes a large-scale language model (hereinafter, the language model in the first large-scale language model server 32 is also referred to as LLM32, and the language model in the second large-scale language model server 34 is also referred to as LLM34), an acoustic model, a WaveNet model, or the like. The LLM32 generates question information for a character and corresponding answer information by the character as training data. On the other hand, the LLM34 generates question information for a user and corresponding answer information by the user as training data (FIGS. 2 to 4, "Question and Answer Set"). The training method is not particularly limited, and a method described in Patent Application No. 2023-197471 by the present applicant may be used. Examples of LLMs 32 and 34 include BERT, XLNeT, T5, and GPT, with multimodal GPT-4 being particularly preferred. LLMs 32 and 34 calculate answer information in response to question information from a user terminal 42 or a client terminal 46 on a website (platform) where a program is executed by the server 10 according to the present disclosure. Furthermore, LLMs 32 and 34 may have a function for converting the format of the answer information (text, audio, video, etc.). In this case, speech recognition, natural language processing, etc. can be performed to analyze the question content, and then audio information or text information of an answer according to the question content can be calculated and output. Other methods may also be employed.

[0043] In the present disclosure, as shown in FIG. 1(B), three types of trained models 34a, 34b, and 34c can be generated as the LLM 34. The "individual (user) model 34a" is a trained model related to one user. The "attribute model 34b" is a trained model related to one attribute, and is generated based on dialogue information (question information and answer information) of multiple users related to (having) that attribute. The "attribute + user model (third model) 34c" is generally a model that combines a certain user model and a certain attribute model (the user in the case of the third model 34c may be one or multiple users).

[0044] The third model 34c can be classified into three patterns as shown in FIG. 1(B) depending on the order of learning the training data (interaction information of a specific user and interaction information of users with specific attributes). In the "user → attribute pattern 34c1," a trained model 34a is generated based on the interaction information of a specific user, and then the trained model 34a learns the interaction information of multiple users with specific attributes to generate the third model 34c (34c1). In the "attribute → user pattern 34c2," a trained model 34b is generated based on the interaction information of a user with specific attributes, and then the trained model 34b learns the interaction information of the specific user to generate the third model 34c (34c2). In the "simultaneous pattern 34c3," the third model 34c (34c3) is generated based on the interaction information of a specific user and the interaction information of users with specific attributes.

[0045] The third model 34c in the "user → attribute pattern 34c1" and the "attribute → user pattern 34c2" can be generated by, for example, online learning, and the third model 34c in the "simultaneous pattern 34c3" can be generated by, for example, batch learning, but is not limited to this. Note that even in batch learning, there may be a time lag between the input of dialogue information of a specific user and the input of dialogue information of a user with a specific attribute. Even in such a case, if both pieces of information are input and then learned together, this falls under the "simultaneous pattern 34c3."

[0046] In the embodiment of the present disclosure, it is assumed that the first large-scale language model server 32 stores trained models related to characters, and the second large-scale language model server 34 stores LLMs 34 related to users, but there are no limitations on which trained models the first large-scale language model server 32 and the second large-scale language model server 34 store. One of the first large-scale language model server 32 and the second large-scale language model server 34 may have all trained models, and the other server may not be part of the information processing system 1.

[0047] <User terminal 42, administrator terminal 44, client terminal 46> The user terminal 42 (including 42a and 42b), the administrator terminal 44, and the client terminal 46 are terminals operated by a user, an administrator of the server 10, and a client, respectively. The user terminal 42, the administrator terminal 44, and the client terminal 46 may include a personal computer, a tablet terminal, a mobile terminal, etc., but are not particularly limited as long as they are capable of transmitting and receiving information to and from the server 10. The user terminal 42, the administrator terminal 44, and the client terminal 46 do not need to be the same terminal; the same user (user, administrator, client) may transmit and receive information using different terminals at different times, as long as each terminal is identified by the same registration information (account information). In the present invention, interactions between users and interactions with a trained model are not limited to being performed on a website, but may be executed by so-called apps (application software) installed on the user terminal 42 or the client terminal 46. Furthermore, the program according to the present disclosure may be executed by both a website and an app.

[0048] (Details of the control unit 12 and the control unit 36) (User registration means 14) The user registration means 14 registers users who will chat on the user terminal 42. The user registration means 14 can also set (register) user attributes when registering a user. Multiple attributes may be set for each user.

[0049] In this specification, "registration" refers to, but is not limited to, assigning an identification number or the like to a user and setting up an account on the server 10, thereby enabling the user to interact with a character (trained model) on the platform. The user registration means 14 can also perform a similar registration process for clients. As a registration method on this platform, for example, a registrant (user, etc.) may express a predetermined intention to be registered (including, for example, consenting to management terms and conditions, providing name, etc., and verifying identity), after which the registrant is assigned an account, identification number, etc. that enables interaction with the trained model.

[0050] (Method of obtaining permission 16) The permission acquisition means 16 transmits a confirmation command to the user terminal 42 to confirm that the user registered by the user registration means 14 is authorized to use the trained model LLM 34 (at least one of 34a, 34b, and 34c) by a specific client. The permission acquisition means 16 acquires the permission of the user operating the user terminal 42 by accepting permission information from the user terminal 42. That is, when permission information is accepted from the user terminal 42, it can be considered that permission has been acquired to use the trained model corresponding to the user. The user's permission may be granted for each trained model or for each client. Note that when an LLM is created by acquiring dialogue information in a chat between users (between the first user terminal 42a and the second user terminal 42b) or in the case of LLMs 34b and 34c, the permission of all users who generated dialogue information as training data for these models may also be required.

[0051] (Authorization Means 18) The authorization means 18 grants authorization to a predetermined client terminal 46 (FIG. 5, "Access Authorization"). That is, the LLM 34 can be caused to calculate (generate) an answer only in response to a command (question) from a predetermined client terminal 46 that has been authorized to use the LLM 34. As described above, the authorization means 18 may grant authorization to a client terminal 46 of a client whose user's authorization has been acquired (opted in) via the authorization acquisition means 16 to use the trained model (at least one of 34a, 34b, 34c) corresponding to that user.

[0052] (Acquisition method 20) The acquisition means 20 acquires dialogue information from a chat held on a user terminal 42. As will be described later, when the acquisition means 20 acquires dialogue information from a chat held between a first user terminal 42a and a second user terminal 42b, the model generation means 38 performs machine learning using the dialogue information acquired by the acquisition means 20 as training data. Here, the "chat" may be a chat with a chatbot (FIG. 2) or a chat between users (FIG. 3). The chat may be conducted using voice or text. In the case of a chat between users, dialogue information (question information, answer information) acquired on a social networking site or the like can also be used as training data to generate the trained model (LLM 34).

[0053] The acquisition unit 20 may acquire specific question information from the administrator terminal 44, thereby enabling the dialogue information acquired by the acquisition unit 20 to include the specific question and the user's answer to the specific question (see FIG. 4, "Question and Answer Set"). Such specific questions can also be stored in the storage unit 28 as needed. The "specific question" is not particularly limited and may be a sentence containing fixed questions, such as a questionnaire, and can be created by the administrator or client as appropriate. Examples of specific questions include direct questions related to the client's products or services (e.g., confirming the client's willingness to purchase a certain product, comparing products, future travel destinations, etc.) and indirect questions not directly related to the client's products or services (e.g., questions about the user's attributes, how they spend their holidays, etc.). The specific question and answer set can be acquired as learning data for the LLM 34 by, for example, asking specific questions to users with certain attributes, such as Generation Z, and having them respond to the questions verbally. The specific questions can be asked by directly outputting a list of questions to the screen of the user terminal 42 or by listening to the questions during a "dialogue" with a character (LLM 32) or the like. Alternatively, without setting specific questions, users with certain attributes can be asked to talk about a specific topic (such as "currently trending words"), which can then be transcribed and used as learning data for the specific question. Having users respond via voice is preferable, as it allows for the collection of "live" spoken language. In the case of voice input, specifically, users' voices can be collected via a web interface (chat interface), or questions can be displayed on the user terminal 42, and users can be asked to record their voice responses.

[0054] (Transmission means 22) The transmitting means 22, for example, transmits the specific question acquired by the acquiring means 20 to the user terminal 42, thereby outputting the specific question to a chat interface of the user terminal 42 (FIG. 4). The "chat interface" is the output section of the user terminal 42, and may be a chat screen (display, etc.) when text, images, or videos are output, or a speaker of the user terminal 42 when audio information is output.

[0055] In addition, when the acquisition means 20 acquires a question from the first user terminal 42a, the transmission means 22 can transmit the question acquired from the first user terminal 42a to the second user terminal 42b, thereby allowing a chat to take place between the first user terminal 42a and the second user terminal 42b.

[0056] (Evaluation inquiry means 24) The evaluation inquiry means 24 transmits a notification to the second user terminal 42b requesting an evaluation of the dialogue information in the chat executed on the first user terminal 42a. The evaluation inquiry means 24 may also transmit a notification to the second user terminal 42b requesting an evaluation of the response of the learned model (LLM 34) (including the response of the LLM 34 to the instruction from the client terminal 46). The notification requesting an evaluation may be sent to multiple users, not just one user. As described above, the learned model LLM 34 can be generated by taking into account (referring to) evaluations of the dialogue information by other user terminals (such as the second user terminal 42b). For example, weighting based on the evaluation results can be performed during learning to reflect the correct sense of attributes in the attribute model 34b. In particular, if the response by the LLM 34 is highly evaluated, the reliability of the learned models 34b and 34c can be improved by using this response information as training data.

[0057] There are no particular limitations on the evaluation method, and for example, the user may be asked to give a graded evaluation such as "I agree," "I understand how you feel!", "I empathize," or "I don't agree," or "I think it's a little different," and only the number indicating the evaluation may be sent to the server 10 (control unit 12) (quantitative evaluation), or an example answer based on what was the deciding factor in the evaluation or the attributes may be presented (qualitative evaluation).

[0058] (Reward granting means 26) The reward granting means 26 grants a reward to a user who provides dialogue information in a chat session executed on the user terminal 42. The "reward" is not particularly limited, and may be money or a monetary equivalent (so-called points, a free dialogue coupon with the LLM 32, etc.). The reward (amount, quantity, etc.) granted by the reward granting means 26 may be determined based on the evaluation (for the dialogue information in the chat session executed on the first user terminal 42a) acquired from the second user terminal 42b in response to the notification from the evaluation inquiry means 24, as described above. The reward may be set so that the higher the evaluation, the larger the reward. Alternatively, the reward may be set so that a larger total amount of dialogue information is provided based on the user's dialogue history on the platform. In the case of the trained models 34b and 34c, the reward granted to a specific user may vary depending on the proportion of the dialogue information of that specific user among the dialogue information from all users.

[0059] Furthermore, the reward granting means 26 may grant a reward to the user (of the second user terminal 42b) who has made an evaluation in response to a notification from the evaluation inquiry means 24. In the case of mutual evaluation between users, a higher reward may be given to a user who has made a higher evaluation. "Mutual evaluation" refers to, for example, another user further evaluating a qualitative or quantitative evaluation made by a certain user (such as "Like!"), and the user's evaluation, empathy, influence, etc. may be quantified. Furthermore, evaluations by users with higher evaluations may be weighted more heavily during learning.

[0060] (Model generation means 38) The model generation means 38 performs machine learning using the acquired dialogue information as training data to generate a trained model LLM34, which takes questions as input and answers as output. Examples of "machine learning" include supervised learning and deep learning. Furthermore, various techniques can be used to train the LLM34, such as fine-tuning the LLM34, cost-effective tuning using an adapter, learning using RAG, and inference based on a summary of the dialogue history. Tuning the character chatbot (LLM32) to output specific questions as dialogue information can also be performed. A prompt can be created for the LLM32 to create the specific question itself, or to change the content of the specific question (including changing the tone of voice) depending on the user (Figure 4). The methods described above can be used to create and collect training data for "specific questions."

[0061] The model generation means 38 can generate a trained model 34a for each user registered by the user registration means 14. The model generation means 38 may generate a trained model 34b for each user attribute. The model generation means 38 can also generate a third trained model 34c1 in which a question is input and an answer is output by performing machine learning on the trained model 34a for each user using dialogue information of a user with a specific attribute as training data. Furthermore, the model generation means 38 can also generate a third trained model 34c2 in which a question is input and an answer is output by performing machine learning on the trained model 34b for each user attribute using dialogue information of a specific user (which may be one or more users) as training data. The model generation means 38 can also generate a third trained model 34c3 in which a question is input and an answer is output by performing machine learning on dialogue information of a specific user and dialogue information of users with a specific attribute as training data. The same question can be asked to each model and the answers can be compared. For example, the attribute model 34b and the third model 34c can provide generally valid answers that cannot be obtained from the individual model 34a, and are more important in terms of marketing. The third model 34c can provide answers that are more personalized than the attribute model 34b. Answers that are more personalized are suitable for creating a recommendation system. The individual model 34a is easy to create because it has less data. The individual model 34a can also be used in marketing activities by comparing it with the attribute model 34b or with other individual models 34a.

[0062] (Answer generation means 40) The answer generation means 40 generates an answer based on the generated trained model LLM34 in response to an instruction from the client terminal 46 and transmits the generated answer to the client terminal 46. Generally, a trained model is also capable of dialogue with itself (questions and answers), and the answer generation means 40 can also transmit dialogue information such as spontaneous speech (so-called monologue) by the trained model LLM34 (34a, 34b, 34c, etc.) as an "answer" to the client terminal 46. As an "instruction," the client can ask questions such as, "What's your family structure?", "Where do you go for fun on the weekend?", "What are your food preferences?", and "If you were to go on a trip, would you go to Hokkaido or Okinawa?", but more specific questions are also possible. The client can also introduce and explain recommended products and see the reaction.

[0063] Furthermore, the answer generation means 40 may generate an answer based on the generated trained model 34a for each user in response to an instruction statement for each user from the client terminal 46, and transmit the generated answer to the client terminal 46. Furthermore, the answer generation means 40 may generate an answer based on the generated trained model 34b for each user attribute or the third trained model 34c (including 34c1 to 34c3) in response to an instruction statement for each user attribute from the client terminal 46, and transmit the generated answer to the client terminal 46. The answer generation means 40 may transmit answers generated based on the LLM 34 (34a, 34b, 34c, etc.) only to client terminals 46 that have been granted authority by the authorization means 18, or may generate answers only in response to instruction statements from client terminals 46 that have been granted authority by the authorization means 18, and transmit the answers to the client terminals 46.

[0064] [Information processing method] Next, an information processing method in the above-described information processing system 1 will be described with reference to the drawings, but the information processing method according to the present disclosure is not limited to the following example. In the following description, components with the same reference numerals are the same as the components described above, and duplicated descriptions will be omitted as appropriate. Note that the processing described below is performed by the control unit 12 (server 10) executing a program stored in the storage unit 28, and an example will be described in which the dialogue information is text information.

[0065] First, with reference to FIG. 2, an example of the flow of information processing during a chat between a user and a chatbot in the information processing system 1 will be described. Here, first, a user operating a user terminal 42 logs in with a user ID to a website (platform) on which the program of the present disclosure is executed. Then, display instruction information for displaying a chatbot screen (text information) on the display of the user terminal 42 is transmitted from the server 10. Then, when the user inputs a question on the chatbot screen, the question information is transmitted to the server 10. Next, an instruction statement for calculating an answer corresponding to the question is generated in the server 10, and this instruction statement is transmitted to the first large-scale language model server 32 (LLM32). Then, answer information is calculated by the LLM32, and this answer information is transmitted to the user terminal 42 via the server 10 and displayed on the chatbot screen. Then, these sets of questions and answers are transmitted as training data from the server 10 to the second large-scale language model server 34 (LLM34).

[0066] Next, with reference to FIG. 3, an example of the flow of information processing during a conversation between users in the information processing system 1 will be described. First, a first user and a second user operating a first user terminal 42a and a second user terminal 42b log in with their user IDs to a website where the program of the present disclosure is executed. Then, display instruction information for displaying a chatbot screen on the displays of the first user terminal 42a and the second user terminal 42b is transmitted from the server 10. Next, a question to the second user is transmitted from the first user terminal 42a via the server 10 and displayed on the chatbot screen of the second user terminal 42b. Then, an answer to the first user is transmitted from the second user terminal 42b via the server 10 and displayed on the chatbot screen of the first user terminal 42a. Thus, while a chat is taking place between the first user and the second user, a set of the question and answer is transmitted from the server 10 to the second large-scale language model server 34 (LLM 34) as training data.

[0067] Furthermore, as shown in the center right of Figure 3, audio information collected (recorded) by a microphone of a conversation about a specific topic between multiple users who actually gather physically together may be used as training data for training the second large-scale language model server 34 (LLM34).

[0068] Next, with reference to FIG. 4, an example of the flow of information processing when a question is posed to a user in the information processing system 1 will be described. Here, first, a specific question is transmitted from the administrator terminal 44 to the server 10 and saved. As indicated by the dotted arrow, this specific question can be changed by the first large-scale language model server 32 (LLM32) depending on the user. Next, a user operating the user terminal 42 logs in with a user ID to a website (platform) on which the program of the present disclosure is executed. Then, display instruction information for displaying a chatbot screen on the display of the user terminal 42 is transmitted from the server 10, and the specific question is transmitted and displayed. Next, the user inputs a response sentence on the chatbot screen, and this response information is transmitted to the server 10. In this way, a set of the specific question and the response is transmitted from the server 10 to the second large-scale language model server 34 (LLM34) as training data.

[0069] Next, an example of the flow of information processing when the client terminal 46 in the information processing system 1 acquires information will be described with reference to Fig. 5. Here, first, the server 10 (authorization means 18) transmits information for granting the client terminal 46 access authorization to the LLM 34. Then, the client terminal 46 transmits an instruction sentence (question information) to the second large-scale language model server 34 (LLM 34). Then, the LLM 34 calculates answer information, which is transmitted to the client terminal 46 and displayed on the chatbot screen. Note that when the monologue (spontaneous speech) of the LLM 34 is transmitted to the client terminal 46 as answer information, no instruction sentence is required from the client terminal 46.

[0070] Next, an example of an information processing method in the information processing system 1 will be described with reference to Fig. 6. First, the control unit 12 (user registration means 14) registers a user who will have a chat on the user terminal 42 (step S10).

[0071] Next, the control unit 12 (permission acquisition means 16) sends a confirmation command to the user terminal 42 to confirm that the user registered by the user registration means 14 is permitted to use the trained model (LLM34a in this example) by a specific client (step S20).

[0072] Next, the control unit 12 (permission acquisition means 16) receives permission information from the user terminal 42 to acquire the permission of the user who operates the user terminal 42 (Opt-in, step S30).

[0073] Next, the control unit 12 (authorization means 18) grants the client terminal 46 of the client for which user permission has been obtained the authority to use the trained model 34a corresponding to that user (step S40).

[0074] Next, the control unit 12 (acquisition means 20) acquires information on the specific question from the administrator terminal 44 (step S50).

[0075] Next, the control unit 12 (transmission means 22) transmits the specific question acquired by the acquisition means 20 to the user terminal 42 so that the specific question is output to the chat interface of the user terminal 42 (step S60; see FIG. 4).

[0076] Next, the control unit 12 (acquisition means 20) acquires conversation information in the chat being executed on the user terminal 42 (step S70).

[0077] Next, the control unit 12 (model generation means 38) performs machine learning for each user registered by the user registration means 14 using the acquired dialogue information as training data, thereby generating a trained model 34a in which a question is input and an answer is output (step S80).

[0078] Next, the control unit 12 (answer generating means 40) generates an answer to a question from each user from the client terminal 46 based on the generated trained model 34a (step S90).

[0079] Next, the generated response is sent to the client terminal 46 (step S100).

[0080] Next, the control unit 12 (evaluation inquiry means 24) sends a notification to the second user terminal 42b requesting an evaluation of the response of the trained model 34a to the instruction statement from the client terminal 46 (step S110). If the evaluation is high, the response information can be used for additional training of the trained models 34b and 34c.

[0081] The information processing method described above is merely an example, and the processing flow is not limited to the above. For example, the processing of steps S20 to S40 (from the transmission of the confirmation command by the permission acquisition means 16 to the granting of authorization by the authorization granting means 18) does not have to be performed immediately after user registration (step S10), but may be performed before the answer generation means 40 receives an instruction (question) from the client terminal 46 (step S90). Furthermore, as described above, if trained models 34b and 34c for each attribute are also generated, answer information can be calculated for all trained models 34a, 34b, and 34c and transmitted to the client terminal 46. In another embodiment, when a spontaneous utterance (including a question, an answer, or monologue) is made by the trained model 34a generated in step S80, a notification requesting an evaluation of the spontaneous utterance may be transmitted to the second user terminal 42b, as in step S110. For example, whether spontaneous speech is natural or not may be evaluated from the perspective of another user who has the same attributes as the user corresponding to the trained model 34a (for example, the method of expression such as wording, the content of the speech, etc.).

[0082] In the information processing system 1, computer (server 10), program, and information processing method of this embodiment configured as described above, the acquisition means 20, model generation means 38, and answer generation means 40 are provided as described above. The acquisition means 20 acquires dialogue information in a chat executed on a user terminal 42. The model generation means 38 performs machine learning using the acquired dialogue information as training data to generate a trained model LLM 34 that takes a question as input and outputs an answer. The answer generation means 40 transmits an answer generated based on the generated trained model LLM 34 to the client terminal 46.

[0083] Various types of chatbots have been created, and conversational information from online chats between chatbots and users, as well as conversational information between users, has been accumulated. According to the information processing system 1, computer (server 10), program, and information processing method of the present embodiment, this conversational information can be acquired as "user Q&A data" and trained to create a "user language model (LLM 34a, 34b, 34c, etc.)." Third-party companies (clients) can use such language models for marketing and recommendation systems. In particular, according to the present invention, monologues based on the user's language model (LLM 34a, 34b, 34c, etc.) can also be continuously provided to clients. Rather than simply searching for and using user opinions and preferences from the conversational information, processing the user's conversational information into a language model as disclosed herein allows clients to efficiently extract desired information.

[0084] Furthermore, in the information processing system 1, computer (server 10), program, and information processing method according to this embodiment, the answer generation means 40 can generate an answer based on the learned model LLM34 in response to an instruction statement from the client terminal 46, and transmit the generated answer to the client terminal 46. In this way, a client operating the client terminal 46 can create an instruction statement and expect to obtain an answer that is useful for marketing or building a recommendation system.

[0085] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include an authorization means 18 that grants authorization to a predetermined client terminal 46, and the answer generation means 40 may transmit an answer generated based on the generated trained model LLM 34 to the client terminal 46 authorized by the authorization means 18, or the answer generation means 40 may generate an answer based on the generated trained model LLM 34 in response to a command from the client terminal 46 authorized by the authorization means 18, and transmit the generated answer to the client terminal 46. In this way, by limiting access authorization to the trained model LLM 34 to a certain range of people, user privacy can be protected.

[0086] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include a user registration means 14 for registering users who will chat on a user terminal 42. In this case, the model generation means 38 generates a trained model LLM34a for each user registered by the user registration means 14, and the answer generation means 40 can transmit an answer generated based on the generated trained model LLM34a for each user to the client terminal 46. In particular, the answer generation means 40 can generate an answer based on the generated trained model LLM34a for each user in response to a user-specific instruction from the client terminal 46 and transmit the generated answer to the client terminal 46. This allows for cases where a client wishes to ask a question to a specific user (the trained model LLM34a). Furthermore, as described above, the client can use the trained model LLM34a for each user as a comparison model for marketing.

[0087] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include user registration means 14 for registering users who will chat on the user terminal 42 and setting the user's attributes. In this case, the model generation means 38 generates a trained model LLM 34b for each user attribute, and the answer generation means 40 can transmit to the client terminal 46 an answer generated based on the trained model LLM 34b for each user attribute. In particular, the answer generation means 40 can generate an answer based on the trained model LLM 34b for each user attribute in response to an instruction from the client terminal 46 for each user attribute, and transmit the generated answer to the client terminal 46. In this way, in addition to the LLM 34a corresponding to each individual user, an attribute model 34b trained on general information about many users can be created. For example, if a "Generation Z LLM" is trained on dialogue data with Generation Z, it is possible to conduct marketing interviews with Generation Z using this LLM.

[0088] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment further include user registration means 14 for registering users who will chat on user terminals 42 and setting user attributes at this time. The model generation means 38 can generate a third trained model 34c3 that inputs a question and outputs an answer by performing machine learning using dialogue information of a specific user and dialogue information of users with specific attributes as training data. In another embodiment, the user registration means 14 sets user attributes, and the model generation means 38 can generate a third trained model 34c1 that inputs a question and outputs an answer by performing machine learning using dialogue information of users with specific attributes as training data on a trained model LLM 34a for each user. In another embodiment, the model generation means 38 can generate a third trained model 34c2 that inputs a question and outputs an answer by performing machine learning using dialogue information of a specific user as training data on a trained model LLM 34b (attribute model 34b) for each user attribute. In this way, various third models 34c (34c1 to 34c3) can be generated according to the situation and needs and used for comparison, etc., allowing for providing a variety of answers to clients. In particular, even if the dialogue information of a specific user additionally learned when generating the third model 34c2 is also used when generating the attribute model 34b, the additionally learned information of the specific user can be reflected with higher priority, making it possible to extract information of the specific user from the third model 34c2 rather than the attribute model 34b. Furthermore, with the third model 34c2, answers that cannot be obtained from the dialogue information of a specific user can be calculated from dialogue information of the same attribute.

[0089] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include a permission acquisition means 16 and an authorization granting means 18 as described below. That is, the permission acquisition means 16 transmits a confirmation command to the user terminal 42 to confirm that the user registered by the user registration means 14 is authorized to use the trained model LLM 34 (at least one of 34a, 34b, and 34c) by a specific client, and then receives authorization information from the user terminal 42 to acquire the authorization of the user operating the user terminal 42. The authorization granting means 18 grants the client terminal 46 of the client from which the user's authorization has been obtained the authorization to use the trained model LLM 34 corresponding to the user. In this case, the answer generation means 40 can generate an answer based on the generated trained model LLM 34 in response to a command from the client terminal 46 authorized by the authorization granting means 18 and transmit the generated answer to the client terminal 46. In this way, if a user permits the use of the learned model LLM34 for each client, the user's privacy can be further protected.

[0090] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include a transmitting means 22 as described below. In this case, the acquiring means 20 acquires information on a specific question from the administrator terminal 44, and the transmitting means 22 transmits the acquired specific question to the user terminal 42, thereby outputting the specific question on a chat interface of the user terminal 42, and the dialogue information acquired by the acquiring means 20 may include the specific question and the user's answer to the specific question. By providing a specific question in this way, the client can more efficiently extract the information he or she wants to know.

[0091] Furthermore, the information processing system 1, the computer (server 10), the program, and the information processing method according to this embodiment may further include a transmission means 22 as described below. In this case, when the acquisition means 20 acquires a question from the first user terminal 42a, the transmission means 22 transmits the question acquired from the first user terminal 42a to the second user terminal 42b, thereby causing a chat between the first user terminal 42a and the second user terminal 42b. When the acquisition means 20 acquires dialogue information from the chat held between the first user terminal 42a and the second user terminal 42b, the model generation means 38 may perform machine learning using the dialogue information acquired by the acquisition means 20 as training data. In this way, a desired trained model LLM (including 34a, 34b, and 34c) can also be generated from dialogue information between users.

[0092] Furthermore, in the information processing system 1, the computer (server 10), the program, and the information processing method according to this embodiment, the chat from which the acquisition unit 20 acquires dialogue information may be a chat with a chatbot. In this way, by using an AI character (e.g., LLM32) that the user is likely to like as a chatbot, it is expected that the user's true feelings and niche information can be elicited from the user apart from drinking parties.

[0093] Furthermore, in the information processing system 1, computer (server 10), program, and information processing method according to this embodiment, the chatbot (LLM 32) is tuned to output specific questions as dialogue information. The content of the chatbot's (LLM 32, AI character) conversation may be tuned to elicit more "true feelings." Furthermore, the chatbot may be tuned to proactively ask specific questions during specific periods (e.g., asking in-depth questions about children before Christmas).

[0094] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include a reward granting means 26 that grants a reward to a user who provides dialogue information in a chat executed on a user terminal 42. In this way, if a reward is granted for providing dialogue information, the user will be motivated to participate in the platform according to the present disclosure and interact with a desired character (chatbot LLM 32). Furthermore, if the responses by the LLM 34 reflect dialogue information from many users, clients will also be motivated to participate in (invest in) the platform and issue instructions.

[0095] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include an evaluation inquiry means 24 that sends a notification to the second user terminal 42b requesting an evaluation of dialogue information in a chat session executed on the first user terminal 42a. The model generation means 38 can perform machine learning by also referring to the evaluation acquired from the second user terminal 42b. In this way, a question-and-answer pair obtained from a user can be presented to other users, who can evaluate it as "agree" or "disagree," for example. Weighting based on such evaluation results during learning can generate a more accurate attribute model 34b.

[0096] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include evaluation inquiry means 24 that sends a notification to the second user terminal 42b requesting an evaluation of the trained model LLM34's response (which may include the trained model LLM34's response to an instruction from the client terminal 46), and the model generation means 38 can perform machine learning by also referring to the evaluation acquired from the second user terminal 42b. In this way, highly reliable trained models LLM34 (particularly 34b and 34c) can be realized by using, as training data, responses from trained models LLM34 that have received high evaluations from users, in addition to dialogue information from users.

[0097] Furthermore, the information processing system 1, computer (server 10), program, and information processing method according to this embodiment may further include a reward granting means 26 that grants a reward to a user who provides dialogue information in a chat executed on a user terminal 42, and the reward granted by the reward granting means 26 may be determined based on an evaluation of the dialogue information in a chat executed on a first user terminal 42a, which is obtained from a second user terminal 42b. For example, if an LLM 34b, 34c such as a "Generation Z LLM" generates business revenue, a portion of that revenue may be returned to the respondent. In this case, rather than returning a uniform amount to the respondent, the return rate can be adjusted using a method including the above-mentioned "weighting," allowing for the generation of more accurate LLMs 34b, 34c.

[0098] The information processing system 1, the computer (server 10), the program, and the information processing method according to the present embodiment are not limited to the above-described aspects and combinations, and various modifications can be made. [Explanation of symbols]

[0099] 1. Information Processing Systems 10 Servers (computers) 12 Control Unit 14 User Registration Method 16 Means of obtaining permission 18 Authorization Instruments 20 Acquisition method 22 Transmission means 24 Evaluation inquiry method 26 Rewarding Means 28 Memory section 30 Communications Department 32 First large-scale language model server (computation unit) 34 Second large-scale language model server (computation unit) 34a Trained model (individual / user model) 34b Trained model (attribute model) 34c, 34c1, 34c2, 34c3 Trained models (attribute + individual model, third model) 36 Control Unit 38 Model Generation Methods 40 Answer generation means 42, 42a, 42b User terminal 44 Administrator terminal 46 client terminals

Claims

1. A system comprising an acquisition means, a model generation means, and an answer generation means, the acquiring means acquires conversation information in a chat being executed on the user terminal; the model generation means performs machine learning using the acquired dialogue information as training data to generate a trained model in which a question is input and an answer is output; The answer generation means transmits an answer generated based on the generated trained model to a client terminal.

2. The system according to claim 1 , wherein the answer generation means generates an answer based on the trained model in response to an instruction from the client terminal and transmits the generated answer to the client terminal.

3. further comprising authorization means; the authorization means grants authorization to a predetermined client terminal; The system according to claim 1 , wherein the answer generation means transmits an answer generated based on the generated trained model to the client terminal to which authorization has been granted by the authorization granting means.

4. further comprising authorization means; the authorization means grants authorization to a predetermined client terminal; The system according to claim 1, wherein the answer generation means generates an answer based on the generated trained model in response to an instruction statement from the client terminal to which authorization has been granted by the authorization granting means, and transmits the generated answer to the client terminal.

5. Further comprising a user registration means, the user registration means registers users who are to carry out chats at the user terminals; the model generation means generates the trained model for each user registered by the user registration means, The system according to claim 1 , wherein the answer generation means transmits to the client terminal an answer generated based on the trained model generated for each user.

6. Further comprising a user registration means, the user registration means registers users who are to carry out chats at the user terminals; the model generation means generates the trained model for each user registered by the user registration means, The system according to claim 1, wherein the answer generation means generates an answer based on the trained model generated for each user in response to an instruction sentence for each user from the client terminal, and transmits the generated answer to the client terminal.

7. Further comprising a user registration means, the user registration means registers a user who will execute a chat on the user terminal, and sets attributes of the user at this time; the model generation means generates the trained model for each user attribute, The system according to claim 1 , wherein the answer generation means transmits to the client terminal an answer generated based on the trained model for each generated user attribute.

8. Further comprising a user registration means, the user registration means registers a user who will execute a chat on the user terminal, and sets attributes of the user at this time; the model generation means generates the trained model for each user attribute, The system according to claim 1, wherein the answer generation means generates an answer based on the trained model for each user attribute generated in response to an instruction sentence for each user attribute from the client terminal, and transmits the generated answer to the client terminal.

9. Further comprising a user registration means, the user registration means registers a user who will execute a chat on the user terminal, and sets attributes of the user at this time; The system of claim 1, wherein the model generation means generates a third trained model that takes a question as input and an answer as output by performing machine learning using dialogue information of a specific user and dialogue information of a user having specific attributes as training data.

10. The user registration means sets attributes of the user, The system described in claim 5 or 6, wherein the model generation means performs machine learning on the trained model for each user using dialogue information of a user with specific attributes as training data, thereby generating a third trained model that inputs a question and outputs an answer.

11. The system described in claim 7 or 8, wherein the model generation means performs machine learning on the trained model for each user attribute using dialogue information of a specific user as training data to generate a third trained model that inputs a question and outputs an answer.

12. further comprising a permission obtaining means and an authorization means; the permission acquisition means transmits to the user terminal a confirmation command that confirms that the user registered by the user registration means is authorized to use the trained model by a specific client, and then receives permission information from the user terminal to acquire the permission of the user who operates the user terminal; the authorization means grants the authorization to use the trained model corresponding to the user to the client terminal of the client for which user permission has been obtained; The answer generation means generates an answer based on the generated trained model in response to an instruction statement from the client terminal to which authority has been granted by the authorization granting means, and transmits the generated answer to the client terminal. The system according to any one of claims 5 to 9.

13. further comprising a transmitting means; The acquisition means acquires information about a specific question from an administrator terminal, the transmission means transmits the acquired specific question to the user terminal, thereby outputting the specific question to a chat interface of the user terminal; 2. The system according to claim 1, wherein the dialogue information acquired by said acquisition means includes a specific question and a user's answer to the specific question.

14. further comprising a transmitting means; when the acquisition means acquires a question from a first user terminal, the transmission means transmits the question acquired from the first user terminal to a second user terminal, thereby allowing a chat to take place between the first user terminal and the second user terminal; The system described in claim 1, wherein when the acquisition means acquires dialogue information in a chat conducted between the first user terminal and the second user terminal, the model generation means performs machine learning using the dialogue information acquired by the acquisition means as training data.

15. The system of claim 1 , wherein the chat is a chat with a chatbot.

16. The system according to claim 15, wherein the chatbot is tuned to output specific questions as dialogue information.

17. Further comprising a reward granting means, 2. The system according to claim 1, wherein the reward providing means provides a reward to a user who provides dialogue information in a chat session executed on the user terminal.

18. Further comprising an evaluation inquiry means, the evaluation inquiry means transmits a notification to the second user terminal requesting an evaluation of dialogue information in a chat executed on the first user terminal; The system according to claim 1 , wherein the model generation means performs machine learning by also referring to an evaluation acquired from the second user terminal.

19. Further comprising an evaluation inquiry means, The evaluation inquiry means transmits a notification to the second user terminal requesting an evaluation of the answer of the trained model, The system according to claim 1 , wherein the model generation means performs machine learning by also referring to an evaluation acquired from the second user terminal.

20. Further comprising an evaluation inquiry means, The evaluation inquiry means transmits to the second user terminal a notification requesting an evaluation of the trained model's response to a command from the client terminal, The system according to claim 2 , wherein the model generation means performs machine learning by also referring to an evaluation acquired from the second user terminal.

21. Further comprising a reward granting means, the reward granting means grants a reward to a user who provides dialogue information in a chat executed on the user terminal; 19. The system according to claim 18, wherein the reward given by the reward giving means is determined based on an evaluation of dialogue information in a chat executed at the first user terminal, the evaluation information being acquired from the second user terminal.

22. A program that causes one or more computers to function as an acquisition means, a model generation means, and an answer generation means, the acquiring means acquires conversation information in a chat being executed on the user terminal; the model generation means performs machine learning using the acquired dialogue information as training data to generate a trained model in which a question is input and an answer is output; The answer generation means transmits an answer generated based on the generated trained model to a client terminal.

23. 1. An information processing method executed by one or more computers, comprising: a step of the computer acquiring conversation information in a chat being executed on a user terminal; a step of causing the computer to perform machine learning using the acquired dialogue information as training data, thereby generating a trained model in which a question is input and an answer is output; A step in which the computer transmits an answer generated based on the generated trained model to a client terminal; An information processing method comprising:

Images