Server device, control method of server device, and program

Abstract

To provide a server device that reduces a burden on a user who uses a service provided using a learning model.SOLUTION: A server device includes request acquisition means, reply acquisition means, and reply provision means. The request acquisition means acquires a request related to a specific theme from a user, which is a first request to at least two or more learning models. The reply acquisition means acquires replies from each of the at least two or more learning models by using at least a scene showing a situation related to the specific theme and the first request. The reply provision means provides the user with the replies acquired from each of the at least two or more learning models.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a server device, a control method for the server device, and a program.

Background Art

[0002] There are technologies for assisting communication between users.

[0003] For example, Patent Document 1 describes an information providing method that enables smooth and good communication by understanding how one's own personality and preferences differ from those of the other party and then using appropriate words or receiving words appropriately. In Patent Document 1, starting from the date of birth, the tendencies of the parts related to communication in human personality and preferences are clearly defined. The trouble tendencies and countermeasures in communication between people with different personalities are also clearly defined. The information providing method of Patent Document 1 is a database of these, where the user inputs the other party's date of birth, implementation date, and scene information, and can address the other party with words that resonate according to the scene.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] As described in paragraph

[0028] of Patent Document 1, there is a service that sets an artificial intelligence (a learning model obtained by machine learning) as a conversation partner and responds to consultations such as the user's worries. Here, since the artificial intelligence (learning model) outputs an answer in response to a question input by the user, if the user's question is ambiguous, there is a possibility of outputting an answer unintended by the user. However, accurately inputting one's own worries and the like in order to draw an appropriate answer from the learning model places a great burden on the user.

[0006] A main object of the present invention is to provide a server device, a control method for the server device, and a program that contribute to reducing the burden on a user who uses a service provided using a learning model.

Means for Solving the Problems

[0007] According to a first aspect of the present invention, there is provided a server device including: a request acquisition unit that acquires a first request for at least two or more learning models, which is a request related to a specific theme from a user; a response acquisition unit that acquires a response from each of the at least two or more learning models using at least a scene indicating a situation related to the specific theme and the first request; and a response providing unit that provides the user with the responses obtained from each of the at least two or more learning models.

[0008] According to a second aspect of the present invention, there is provided a control method for a server device including: a request acquisition step of acquiring a first request for at least two or more learning models, which is a request related to a specific theme from a user; a response acquisition step of acquiring a response from each of the at least two or more learning models using at least a scene indicating a situation related to the specific theme and the first request; and a response providing step of providing the user with the responses obtained from each of the at least two or more learning models.

[0009] According to a third aspect of the present invention, a computer mounted on a server device is caused to execute a request acquisition process for acquiring a first request regarding a specific theme from a user, the first request being a request for at least two or more learning models, and an answer acquisition process for acquiring an answer from each of the at least two or more learning models by using at least a scene indicating a situation related to the specific theme and the first request, and an answer providing process for providing the user with the answers obtained from each of the at least two or more learning models. A program for causing the computer to execute these processes is provided.

Advantages of the Invention

[0010] According to each aspect of the present invention, a server device, a control method for the server device, and a program are provided, which contribute to reducing the burden on a user who uses a service provided using a learning model. Note that the effects of the present invention are not limited to the above. Instead of or together with the above effects, other effects may be achieved by the present invention.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Figure 23

Figure 24

Figure 25

[0012] First, an outline of an embodiment will be described. Note that the reference numerals in the drawings attached to this outline are added to each element for convenience as an example to assist understanding, and the description of this outline is not intended to be limiting in any way. Also, unless otherwise specified, the blocks shown in each drawing represent a configuration in terms of functional units, not hardware units. The connection lines between the blocks in each figure include both bidirectional and unidirectional ones. The one-way arrow schematically shows the flow of the main signal (data) and does not exclude bidirectionality. In this specification and the drawings, elements that can be similarly described may be given the same reference numerals to omit redundant description.

[0013] A server device 100 according to an embodiment includes a request acquisition unit 101, a response acquisition unit 102, and a response providing unit 103 (see FIG. 1). The request acquisition unit 101 acquires a first request related to a specific theme from a user and addressed to at least two or more learning models (step S1 in FIG. 2). The response acquisition unit 102 acquires responses from at least two or more learning models respectively using at least a scene indicating a situation related to the specific theme and the first request (step S2). The response providing unit 103 provides the responses obtained from at least two or more learning models respectively to the user (step S3).

[0014] The server device 100 provides, for example, a service that accepts questions from users and has a learning model answer those questions. At that time, the server device 100 generates a query to be input to the learning model based on a scene and a request that indicate a situation where the user wants to ask a question (a request for an answer from the learning model) to the learning model. That is, the server device 100 generates a query to be input to the learning model so that the answer from the learning model does not deviate significantly from the question from the user. For example, by selecting a scene related to the matter the user wants to ask the learning model, the user can obtain an appropriate answer from the learning model without inputting detailed background, circumstances, etc. related to their question. That is, the burden on the user who uses the service provided using the learning model is reduced.

[0015] Specific embodiments will be described in more detail below with reference to the drawings.

[0016] [First Embodiment] The first embodiment will be described in more detail with reference to the drawings.

[0017] [Configuration of the System] As shown in FIG. 3, the information processing system according to the first embodiment includes a server device 10.

[0018] The server device 10 is operated by a business operator (service provider) that provides a service for giving advice on users' doubts, questions, etc. regarding a specific theme. In the first embodiment, the server device 10 will be described by taking the case of providing a support service for child-rearing as an example.

[0019] The server device 10 is a server that realizes the main operations of a service provider that provides a child-rearing support service. The server device 10 may be installed inside the building of the service provider or may be installed on the network (in the cloud).

[0020] The user possesses the terminal 20. The user operates the terminal 20 to input various information into the server device 10 or the like, or to acquire various information from the server device 10 or the like.

[0021] Each device shown in FIG. 3 is connected to each other. Specifically, the server device 10 and the terminal 20 are connected by wired or wireless communication means and are configured to be able to communicate with each other.

[0022] The configuration of the information processing system shown in FIG. 3 is an example and is not intended to limit the configuration. For example, the information processing system may include a plurality of server devices 10. Load distribution and redundancy may be achieved by the plurality of server devices 10.

[0023] [Schematic Operation] Subsequently, the schematic operation of the information processing system according to the first embodiment will be described.

[0024] [Preparation of Learning Model] The server device 10 provides a childcare support service using an AI (Artificial Intelligence) model (learning model) obtained by machine learning. A service provider (such as an employee of the service provider) prepares a learning model to be used for the childcare support service.

[0025] The service provider prepares a plurality of learning models that each give different answers (outputs) to questions (inputs). The service provider gives characteristics (personalities) to the answers of each learning model by appropriately selecting the learning data used for generating the learning model.

[0026] For example, the service provider uses the statements of childcare experts (statements of experts in TV programs, etc.) as learning data (text data) to generate a learning model that outputs answers reflecting the thinking of the experts.

[0027] Alternatively, the service provider generates a learning model that outputs responses reflecting the thinking of a scholar (e.g., a university professor) using learning data obtained from books and the like of scholars related to early childhood education and the like.

[0028] Alternatively, the service provider generates a learning model that outputs responses reflecting the experiences and thinking of child-rearing experiencers using learning data such as the statements of child-rearing experiencers on the Internet. Note that the learning data related to child-rearing experiencers is collected from weblogs (so-called blogs), SNSs (Social Networking Services), and the like.

[0029] Note that the service provider prepares each of the above multiple learning models as a large language model (LLM; Large Language Model).

[0030] <Definition of Scenes> The service provider defines a situation related to a specific theme as a "scene". In the first embodiment, various situations that can occur during child-rearing are defined as "scenes". That is, the service provider pre-defines as "scenes" the scenarios and situations that many child-rearers will experience.

[0031] A scene indicates an environment set to simulate (reproduce pseudo) the scene of child-rearing in daily life. The scenes defined by the service provider have a specific purpose or theme, and some or all of the scenes reflect the troubles related to actual child-rearing that a child-rearer may encounter.

[0032] By appropriately defining the scenes, the information processing system can provide the user with a framework for sharing the troubles of child-rearing among child-rearers and finding solutions.

[0033] For example, "putting to sleep" is exemplified as a specific example of a scene. In the putting-to-sleep scene, a situation where a caregiver puts a child to sleep is simulated. Alternatively, "meal management" is exemplified as a specific example of a scene. In the meal management scene, a situation related to a child's meal is simulated.

[0034] When a plurality of learning models (a plurality of learning models characterized by learning data) are implemented in the server device 10 and a scene is defined by a service provider, the server device 10 can provide a childcare support service.

[0035] <Account generation> A user (e.g., a caregiver) generates an account with the service provider in order to receive the childcare support service. Specifically, the user operates the terminal 20 to access the server device 10. The user inputs login information (ID, password), name, gender, date of birth, address, phone number, email address, etc. into the user registration site provided by the server device 10.

[0036] The server device 10 that has obtained the user's name, etc. generates a user ID for identifying the user. The server device 10 stores the user ID, login information, name, date of birth, etc. in association with each other. The server device 10 stores the user ID, login information, name, biometric information, etc. in the user management database. Details of the user management database will be described later.

[0037] By generating an account with the service provider, the user can receive the childcare support service from the service provider.

[0038] <Start of using the childcare support service> A user who wants to use the childcare support service logs in to their own account. A user who wishes to use the childcare support service first selects a scene corresponding to the troubles they have. For example, the server device 10 uses a GUI (Graphical User Interface), etc. to obtain a scene corresponding to the troubles the user wants to solve.

[0039] When a scene is selected, the user selects a learning model that answers questions regarding their concerns and the like. For example, the server device 10 displays a list of each learning model characterized by learning data, and enables the user to select a learning model.

[0040] At that time, the user selects at least two or more learning models from among the plurality of learning models installed in the server device 10.

[0041] For example, the user selects two or more learning models from among a learning model (expert model) that reflects the ideas of childcare experts and the like, a learning model (scholar model) that reflects the ideas of scholars of early childhood education and the like, and a (user model) that reflects the ideas of childcare experiencers and the like.

[0042] When a scene and a learning model are selected, the user starts a conversation regarding the selected scene, targeting the two or more selected learning models.

[0043] <Conversation with the learning model> First, the user operates the terminal 20 to input to the server device 10 the concerns and the like that they want to resolve. The server device 10 outputs an answer to the user's concerns. The user requests an answer from the server device 10 until they feel that the concerns and the like have been resolved.

[0044] As functions for realizing the above childcare support service, the server device 10 includes a basic function, an intervention function, a flip function, and the like. The server device 10 realizes these functions using the plurality of learning models selected by the user.

[0045] <Basic function> First, the basic function will be described.

[0046] The basic function is to sequentially present answers from multiple learning models in response to a user's question (request for an answer). In the basic function, the second and subsequent learning models answer while referring to the answers from each of the learning models that have answered so far.

[0047] The request (user's question) and answer (response of the learning model) in the basic function are referred to as the first request and the first answer. For one first request, a first answer is obtained from each of a plurality of learning models. The plurality of first answers can be different from each other.

[0048] Note that the first answer (answer of the first learning model) to the first request is denoted as the first request [1]. Similarly, the second answer to the first request is denoted as the first request [2], and the third answer as the first answer [3]. The first answer from the k-th learning model is denoted as the first answer [k] (k is a positive integer, the same hereinafter).

[0049] Furthermore, the plurality of first answers are collectively referred to as the first answer set.

[0050] The server device 10 acquires the first request from the user. The server device 10 inputs the acquired first request into one of the plurality of learning models selected by the user.

[0051] Specifically, the server device 10 selects one learning model from among the plurality of learning models selected by the user. Furthermore, the server device 10 generates a query based on the first request and the scene selected by the user so that an answer corresponding to the scene selected by the user can be obtained. The server device 10 inputs the generated query into the one selected learning model.

[0052] For example, when the user selects three models: an expert model, a scholar model, and an experienced person model, the server device 10 selects the expert model from among the three learning models. The server device 10 inputs the query into the selected expert model.

[0053] When the server device 10 obtains the output data (the first response [1]) from the learning model, it selects one learning model from among the plurality of learning models selected by the user and for which no query has been input. For example, in the above example, the scholar model is selected from among the scholar model and the experienced model.

[0054] Based on the first request, the scene selected by the user, and the output data (the first response) of the learning model for which the response has been obtained, the server device 10 generates a query for input to the second selected learning model. In the above example, in response to the first request, the scene, and the first response [1] of the expert model, a query for input to the scholar model (a query for the scholar model) is generated.

[0055] The server device 10 inputs the generated query to the second selected learning model to obtain output data (the first response [2]). In the above example, the server device 10 obtains the first response [2] from the scholar model.

[0056] When the first response [2] is obtained from the second learning model, the server device 10 selects the third learning model for which the response is to be obtained. In the above example, the experienced model is selected.

[0057] When the third learning model is selected, the server device 10, in the same manner as in the case of the second learning model, generates a query for the third learning model based on the first request, the scene, and the output data (the first response [1], the first response [2]) of the learning models for which the response has been obtained. The server device 10 uses the generated query to obtain the first response [3] from the third learning model. In the above example, the server device 10 obtains the first response [3] from the experienced model.

[0058] The input and output regarding the above three learning models are summarized as shown in FIG. 4. As shown in FIG. 4, the server device 10 adds the user's question (response request) and the responses (output data) of the learning models for which the response has been obtained to the scene and inputs them to the subsequent learning model, thereby obtaining the response of the learning model that refers to the response of the previous learning model.

[0059] When the server device 10 obtains the first responses [1] to [3] from each learning model selected by the user in response to the first request of the user, it displays the first responses (first response set) of the respective learning models on the terminal 20 (see FIG. 5).

[0060] Note that FIG. 5 shows an example of the first request and the first responses when the user selects the "putting to bed scene" and selects three learning models: the expert model, the scholar model, and the experienced model.

[0061] In the drawings including FIG. 5, the person (icon) on the left represents the user (questioner). Also, the person (icon) on the right represents the learning model (answerer). Further, the server device 10 makes it easy for the user to identify the answerer by changing the color and the like of the icon corresponding to each learning model. For example, in FIG. 5, the icons on the right correspond to the expert model, the scholar model, and the experienced model in order from the top.

[0062] The user who thinks that the worries and the like have been resolved by the first responses (first response set) from each learning model ends the childcare support service.

[0063] Here, the functions that respond to the user's desire for more answers and answers with deeper content are the "intervention function" and the "flip function".

[0064] <Intervention function> The intervention function is a function of sequentially obtaining new answers from a plurality of learning models by the user adding (intervening) questions to the first responses (first response set) obtained by the basic function. The intervention function is a function of answering additional questions by the user.

[0065] In the intervention function, from the second and subsequent learning models, new answers that refer to the new answers of each learning model obtained so far are obtained.

[0066] The requirements and new responses added in the intervention function are described as the second requirements and the second responses. For each of the second requirements, a second response is obtained from each of the plurality of learning models. The plurality of second responses are collectively described as the second response set.

[0067] Note that the intervention function may be further executed for responses after the second response. The new requirements and new responses in the intervention function performed for the nth response are described as the (n + 1)th requirements and the (n + 1)th responses, and the plurality of (n + 1)th responses are collectively described as the (n + 1)th response set.

[0068] For example, in the state shown in FIG. 5 (the state in which the first response set is displayed), the user inputs an additional question (the second requirement) to the server device 10 (see FIG. 6A).

[0069] When acquiring the additional question (the second requirement) from the user, the server device 10 generates a query from the second requirement of the user, the scene selected by the user, the first requirement, and / or the first response set. The server device 10 inputs the generated query into one of the plurality of learning models selected by the user.

[0070] For example, the server device 10 inputs the query generated by the learning model first selected in the basic function (the expert model in the above example). The server device 10 obtains the second response [1] from the learning model.

[0071] Similar to the basic function, the server device 10 generates a query for input to the second selected learning model. The server device 10 generates a query based on the second requirement, the scene, the first requirement, and / or the first response set, and the second response [1] of the first selected learning model. The server device 10 inputs the generated query into the second learning model (the scholar model in the above example).

[0072] The server device 10 performs the same processing for the third learning model (the experienced model).

[0073] When the server device 10 obtains a second response (second response set) from each learning model selected by the user in response to the user's second question, the server device 10 displays the second response on the terminal 20 (see FIG. 6B).

[0074] The inputs and outputs related to the three learning models in the intervention function are summarized as shown in FIG. 7. In FIG. 7, the first response set obtained by the basic function is used for generating each query.

[0075] In this way, the user can repeatedly ask questions to the server device 10 (the learning models selected by the user) using the intervention function.

[0076] <Flip function> The flip function is a function that allows the user to change (flip) the order in which a plurality of learning models are used for the response obtained by the basic function.

[0077] In the flip function, for at least one learning model, a new first response is obtained by referring to the already obtained first responses from each learning model whose order after the change is earlier than that of the learning model.

[0078] Here, the order of the learning models after being changed by the flip function is denoted as k(new). For example, the server device 10 causes one learning model [k(new)] to refer to the first responses [1] to [k(new)-1] obtained from each of the learning models [1] to [k(new)-1] whose order after the change is earlier than that of the learning model [k(new)]. Thereby, the server device 10 acquires a new first response [k(new)] corresponding to the first request from the learning model [k(new)]. Note that the server device 10 may acquire a new nth response based on an order change instruction for the nth response set.

[0079] For example, a user who has confirmed the answers of each learning model shown in FIG. 5 wants to know more about the answer of the expert model. Specifically, the user wants to know the answer of the expert model based on the answers of other learning models (scholar model, experienced model) to their own questions.

[0080] In this case, the user gives an "order change instruction" to the server device 10 so that the expert model answers last. The specific operation content of the order change instruction will be described later. As shown in FIG. 8A, the user performs an operation to move the first answer [1] of the expert model to the first answer [3] of the last experienced model.

[0081] Upon receiving the order change instruction, the server device 10 generates a query for input to the learning model whose order has been changed by the user. Specifically, the server device 10 generates a query based on the first request, the scene, the first answers [2] and [3] of the already answered learning models.

[0082] The server device 10 inputs the generated query to the learning model whose order has been changed by the user. As described above, when the order is changed so that the expert model answers last, the server device 10 obtains the first answer [3(new)] from the expert model. The server device 10 displays the obtained first answer [3(new)] on the terminal 20 (see FIG. 8B).

[0083] The inputs and outputs related to the three learning models in the flip function are summarized as shown in FIG. 9. As shown in FIG. 9, the expert model selected by the server device 10 as the learning model that answers first in the basic function has its order changed to be the learning model that answers last by the user. That is, the answer order of the expert model, scholar model, and experienced model is changed to the answer order of the scholar model, experienced model, and expert model by the user using the flip function.

[0084] Subsequently, the details of each device included in the information processing system according to the first embodiment will be described.

[0085] [Server device] FIG. 10 is a diagram showing an example of the processing configuration (processing modules) of the server device 10 according to the embodiment disclosed in the present application. Referring to FIG. 10, the server device 10 includes a communication control unit 201, a learning model management unit 202, a scene management unit 203, an account control unit 204, a service control unit 205, and a storage unit 206.

[0086] The communication control unit 201 is a means for controlling communication with other devices. For example, the communication control unit 201 receives data (packets) from the terminal 20. Also, the communication control unit 201 transmits data to the terminal 20. The communication control unit 201 delivers the data received from other devices to other processing modules. The communication control unit 201 transmits the data acquired from other processing modules to other devices. In this way, other processing modules perform data transmission and reception with other devices via the communication control unit 201. The communication control unit 201 has a function as a receiving unit for receiving data from other devices and a function as a transmitting unit for transmitting data to other devices.

[0087] The learning model management unit 202 is a means for controlling and managing the learning model.

[0088] The learning model management unit 202 acquires a plurality of learning models (a plurality of learning models characterized by learning data) prepared by a system administrator or the like.

[0089] At that time, the learning model management unit 202 acquires information such as the model name, features, and information regarding the learning data used for generating the learning model for each learning model from a system administrator or the like. For example, the learning model management unit 202 acquires the information of each learning model using a GUI or the like. Alternatively, the learning model management unit 202 may acquire the information of each learning model via a USB (Universal Serial Bus) memory or the like.

[0090] The learning model management unit 202 stores the information of the acquired learning model in the learning model management database (see Fig. 11). Note that the learning model management database shown in Fig. 11 is an example and is not intended to limit the items to be stored.

[0091] For example, a learning model specialized for a specific area (scene) of child-rearing may be registered in the server device 10.

[0092] For example, a bedtime support model specialized for "putting a child to bed" may be generated and registered in the server device 10. The bedtime support model provides (answers) detailed information on techniques and routines for putting a child to bed. Alternatively, a relaxation support model specialized for "relaxation" may be generated. The relaxation support model provides detailed information on a child's relaxation and mindfulness. Alternatively, a behavior analysis model specialized for behavior analysis related to putting a child to bed may be generated. The behavior analysis model is generated based on behavior analysis of a child's behavior before going to bed and makes proposals for improvement related to putting a child to bed. Alternatively, a nighttime story creation model specialized for the generation of stories when putting a child to bed may be generated. The nighttime story creation model provides stories and reading aloud when putting a child to bed.

[0093] Subsequently, the learning model and the generation of the learning model will be outlined.

[0094] A learning model is a model that generates an answer corresponding to a request. As an example, when a query generated based on a request from a user or the like is input, the learning model outputs an answer corresponding to the query.

[0095] As an example, a learning model is composed of a language model. The language model may be, for example, but not limited to, what is called an LLM (Large Language Models).

[0096] A language model is a machine learning model (also called a generative model) that takes language as input and outputs language. A language model learns the relationships between words in a sentence and is a model that generates related strings related to the target string from the target string. By using a language model that has learned sentences and texts in various contexts, it is possible to generate related strings with appropriate content related to the target string.

[0097] For example, the case of using a language model in question answering will be described. The language model receives, as the target string, the input of the question "What kind of country is Japan?". The language model generates a string such as "Japan is an island country in the Northern Hemisphere." as an answer to the question.

[0098] The learning method of the language model is not particularly limited. As an example, it may be one that is learned to output at least one sentence including the input string. For a specific example, the language model is GPT (Generative Pre-Transformer) that outputs a sentence including the input string by predicting a string with a high probability following the input string.

[0099] In addition to this, for example, T5 (Text-to-Text Transfer Transformer), BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately), etc. are also language models.

[0100] Alternatively, the language model may output a natural language corresponding to a string input in an artificial language. Also, the content generated by the language model is not limited to strings. The language model may generate, for example, image data, video data, audio data, or other data formats corresponding to the input string.

[0101] A learning model (language model) is generated based on learning data. For example, when generating a learning model that reflects the thinking of experts, various learning data (data resources) such as public statements by experts in the media, related industry articles, and reports are utilized for training (learning) the learning model.

[0102] Alternatively, an existing learning model (language model) may be utilized to generate a learning model with specific characteristics. For example, transfer learning (fine-tuning) may be performed by training the weights of a pre-generated learned model with new learning data. Specifically, by using an existing language model and conducting additional training with unique learning data (dataset), the learning model may be given characteristics. By preparing the statements of experts, scholars, and parents with child-rearing experience as learning data respectively and performing additional learning of these learning data on a basic learning model, the learning model may be characterized (personalized).

[0103] The method of "feedback" may be used in the generation of the learning model. For example, an evaluator (the generator of the learning model) etc. determines the appropriateness of the feedback for the output from the learning model. The judgment result is used as learning data and re-learning is executed. As a result, the performance (output accuracy) of the learning model is improved. For example, when generating a learning model (expert model) that reflects the thinking of experts, the evaluator (the generator of the learning model) may provide feedback so that the output result of the expert model "becomes closer to a more specific expert".

[0104] Regarding the characterization (personalization) of the learning model, in addition to generating the learning model from learning data, the learning model may be characterized by so-called "prompt engineering". That is, by devising questions and instructions (prompts) input to the learning model, the output of the learning model may be guided (regulated) in a specified manner. For example, when it is desired to obtain an answer like that of an expert (when it is desired to utilize an existing learning model as an expert model), a fixed phrase such as "Please think like expert A" may be input to the learning model. Alternatively, related statements of an expert, etc. may be acquired from a database, and the statements obtained from the database may be input to the learning model together with the above fixed phrase.

[0105] The scene management unit 203 is a means for controlling and managing the scenes when providing the childcare support service.

[0106] The scene management unit 203 acquires information about the scene from a system administrator or the like. For example, the scene management unit 203 acquires information about the scene using a GUI or the like. The scene management unit 203 acquires information such as the scene name, the description of the scene, and the purpose of the scene.

[0107] The scene management unit 203 stores the acquired scene information in the scene management database (see FIG. 12). Note that the scene management database shown in FIG. 12 is an example and is not intended to limit the items to be stored or the like.

[0108] Note that the scene creator generates a scene (information about the scene) in cooperation with childcare experts and educators. The scene creator defines in advance the situations regarding communication with children that are important in daily life. For example, scenes regarding "putting to bed" and "meal management" are defined, and the related information is input to the server device 10.

[0109] The account control unit 204 is a means for controlling the user's account.

[0110] The account control unit 204 acquires the user's login ID (ID, password), name, gender, address, date of birth, phone number, email address, etc. on the user registration site.

[0111] When acquiring the name, etc., the account control unit 204 generates a user ID for identifying the user. The user ID can be any information as long as it can uniquely identify the user. For example, the account control unit 204 may assign a unique value each time an account is generated and use it as the user ID.

[0112] The account control unit 204 stores the name, gender, etc. in the user management database (see Fig. 13). Note that the user management database shown in Fig. 13 is an example and is not intended to limit the items to be stored.

[0113] The account control unit 204 authenticates the user who logs in to the account using the login information.

[0114] The service control unit 205 is a means for executing control related to the user support (child-rearing support) service.

[0115] The service control unit 205 has a function as a request acquisition unit, a function as a response acquisition unit, and a function as a response provision unit.

[0116] The request acquisition unit acquires a first request for at least two or more learning models, which is a request related to a specific theme (e.g., child-rearing) from the user. The response acquisition unit acquires responses from at least two or more learning models respectively using at least a scene (e.g., bedtime scene) indicating a situation related to the specific theme and the first request. The response provision unit provides the responses obtained from at least two or more learning models respectively to the user.

[0117] In addition, the service control unit 205 has a function as a scene selection control unit that acquires the scene selected by the user from among a plurality of scenes. Further, the service control unit 205 has a function as a learning model selection control unit that acquires at least two or more learning models selected by the user from among a plurality of learning models.

[0118] Hereinafter, the operation of the service control unit 205 will be described in detail.

[0119] When a user who has logged in to an account wishes to receive a childcare support service, the service control unit 205 acquires the scene in which the user wishes to consult (have a conversation with AI). For example, the service control unit 205 acquires the scene desired by the user using a GUI or the like.

[0120] For example, the service control unit 205 displays a list of scenes stored in the scene management database and realizes the selection of a scene by the user (see FIG. 14).

[0121] After acquiring the scene, the service control unit 205 acquires at least two or more learning models to be the conversation partner of the user. The service control unit 205 acquires a plurality of learning models using a GUI or the like. The service control unit 205 displays a list of learning models stored in the learning model management database and realizes the selection of a learning model by the user (see FIG. 15).

[0122] After acquiring the scene desired by the user and a plurality of learning models, the service control unit 205 acquires a question (answer request; first request) from the user. When the service control unit 205 acquires the first request, it acquires the first answer of each learning model selected by the user by the basic function.

[0123] In the following description, the number of learning models selected by the user is denoted as "N" (N is an integer of 2 or more, the same hereinafter). For example, in the above example, N = 3.

[0124] Also, each learning model selected by the user is denoted as "learning model ML-N" using the above N. For example, in the above example, N = 3, and learning model ML-1, learning model ML-2, and learning model ML-3 are each learning models selected by the user. In the above example, the expert model corresponds to learning model ML-1, the scholar model corresponds to learning model ML-2, and the experienced model corresponds to learning model ML-3.

[0125] Furthermore, the order in which the learning model ML-N is used is predetermined at least in the basic functions. In the following description, it is assumed that each learning model answers in the order of learning model ML-1 (expert model), learning model ML-2 (scholar model), and learning model ML-3 (experienced model).

[0126] FIG. 16 is a diagram showing an example of the operation of the service control unit 205 according to an embodiment of the present disclosure. Referring to FIG. 16, the operation related to the basic function of the service control unit 205 will be described.

[0127] Step S101 is an example of input acquisition processing. In step S101, the service control unit 205 acquires the first request input by the user to the terminal 20. For example, the service control unit 205 acquires the first request "The child has trouble falling asleep." shown in FIG. 5.

[0128] Steps S102 and S103 in FIG. 16 are an example of at least a part of the response acquisition processing. The steps S102 and S103 are an example of processing for acquiring the first response [1] from the first learning model ML-1.

[0129] In step S102, the service control unit 205 generates a query for input to the first learning model ML-1 based on the first request and the scene selected by the user.

[0130] For example, the service control unit 205 may generate a query in a format that conforms to the learning model ML-1 from the first request and the scene selected by the user. The conforming format is a format for obtaining an appropriate answer from the learning model ML-1. For example, the service control unit 205 may convert the first request and the scene according to a predetermined conversion rule and use it as a query.

[0131] Also, for example, the service control unit 205 may generate the query in the conforming format by inputting the first request and the scene into a query generation model (AI model for query generation; not shown) that can generate a query in a format that conforms to the learning model ML-1. The query generation model may be a language model, but is not limited thereto.

[0132] When the user selects the bedtime scene, for example, as a query to be input to the learning model ML-1, a query "Bedtime scene; The user is troubled by the poor sleep of the child" may be generated from the bedtime scene and the first request "The child has trouble falling asleep".

[0133] In step S103, the service control unit 205 inputs the generated query into the learning model ML-1 to obtain the first answer [1] from the learning model ML-1.

[0134] In step S104, the service control unit 205 displays the obtained first answer [1] on the terminal 20. In the example of FIG. 5, the first answer "There may be no relaxation time." on the right side of the screen area is the first answer [1].

[0135] Subsequently, for each of k = 2, 3, ···, N, steps S105 to S107 are executed.

[0136] Steps S105 to S107 are an example of a series of processes for obtaining the first answer [k] from the second and subsequent learning models ML-k.

[0137] First, the case of k = 2 for obtaining the first response [2] will be described.

[0138] In step S105, the service control unit 205 generates a query for input to the second learning model ML-2 based on the first request, the scene, and the first response [1].

[0139] For example, the service control unit 205 may generate a query in a format suitable for the learning model ML-2 from the first request, the scene, and the first response [1]. For example, the service control unit 205 may generate the query in the suitable format by inputting the first request, the scene, and the first response [1] into the above-described query generation model. Since specific examples of the query generation model are as described above, detailed descriptions will not be repeated.

[0140] As an example, from the first request "The child has trouble falling asleep.", the bedtime scene, and the first response [1] "There may be no relaxation time.", a query "Bedtime scene; The user is troubled by the child's trouble falling asleep. The expert points out that there is no relaxation time." may be generated.

[0141] In step S106, the service control unit 205 inputs the generated query into the learning model ML-2 to obtain the first response [2] from the learning model ML-2.

[0142] In step S107, the service control unit 205 displays the obtained first response [2] on the terminal 20. In the example of FIG. 5, the second response "Academically, relaxation time is also important." on the right side of the screen area is the first response [2].

[0143] The first response [2] reflects the thinking of scholars and other content that forms the basis of the scholar model. The first response [2] may be different from the first response [1] corresponding to the same first request.

[0144] In the example of FIG. 5, the first response [2] of the scholar model includes "Academically, relaxation time is also important." The first response [2] is the response of the scholar model corresponding to the first request "The child has trouble falling asleep.", and is generated by the learning model ML-2 that refers to the first response [1] of the expert model "There may be no relaxation time."

[0145] Subsequently, the case of k = 3 for obtaining the first response [3] will be described.

[0146] In step S105, the service control unit 205 generates a query for input to the third learning model ML-3 based on the first request, the scene, the first response [1], and the first response [2].

[0147] Also, for example, the service control unit 205 may generate a query in a format suitable for the learning model ML-3 from the first request, the scene, the first response [1], and the first response [2]. For example, the service control unit 205 may generate the query in the suitable format by inputting the first request, the scene, the first response [1], and the first response [2] to the above-described query generation model.

[0148] As an example, from the first request "The child has trouble falling asleep.", the bedtime scene, the first response [1] "There may be no relaxation time.", and the first response [2] "Academically, relaxation time is also important.", a query "Bedtime scene; The user is troubled by the fact that the child has trouble falling asleep. The expert points out that there is no relaxation time. The scholar points out that relaxation time is also important academically." may be generated.

[0149] Steps S106 and S107 for the case of k = 3 are similarly described by replacing [2] with [3] in the description for the case of k = 2. As a result, the first response [3] obtained in step S106 reflects the thoughts of the child-rearing experienced person and may be different from the first response [1] and the second response [2] corresponding to the same first request.

[0150] In FIG. 5, the first response [3] of the experienced model includes "It is recommended to read picture books together." The first response [3] is the response of the experienced model corresponding to the first requirement "The child has trouble falling asleep.", and is generated by the learning model ML-3 with reference to the first response [1] and the first response [2].

[0151] Thereafter, while incrementing k by 1 each time, steps S105 to S107 are executed for each k. When these series of steps are completed for k = N, the basic function ends.

[0152] Regarding these series of steps when k ≥ 4, since they are almost the same as when k = 2 or 3, the detailed description will not be repeated. In the example of FIG. 5, since N = 3, the first response [1] to the first response [3] are displayed and the basic function has ended. The first response [1] to the first response [3] constitute an example of the first response set.

[0153] In this way, the service control unit 205 generates a first query to be input to the first learning model among at least two or more learning models based on the scene and the first requirement. The service control unit 205 obtains a first response from the first learning model by inputting the generated first query to the first learning model. The service control unit 205 generates a second query to be input to the second learning model among at least two or more learning models based on the scene, the first requirement, and the first response. The service control unit 205 obtains a second response from the second learning model by inputting the generated second query to the second learning model. The service control unit 205 provides the first and second responses to the user.

[0154] FIG. 17 is a diagram showing an example of the operation of the service control unit 205 according to the embodiment disclosed in the present application. With reference to FIG. 17, the operation regarding the intervention function of the service control unit 205 will be described.

[0155] The intervention function is executed following the basic function. For example, the intervention function starts in the state shown in FIG. 5. As shown in FIG. 17, the intervention function includes steps S201 to S207.

[0156] Each of steps S201 to S207 is described in substantially the same way by reading the first request in steps S101 to S107 of FIG. 16 as the second request and the first response as the second response. However, the following points are different.

[0157] In step S202, the service control unit 205 generates a query for input to the first learning model ML-1 based not only on the second request but also on one or both of the first requests and the first response set obtained so far.

[0158] Also, in step S205, the service control unit 205 generates a query for input to the k-th learning model ML-k based not only on the second request and the second responses [1] to [k-1] but also on one or both of the first requests and the first response set.

[0159] For example, the bottom row of FIG. 6A includes the second request "It is difficult to read a picture book together." obtained in step S201.

[0160] In the example of FIG. 6A, in step S202, in addition to the second request, a query is generated based on the first request, the scene, and the first response set. For example, the query "Bedtime scene; The user is having trouble putting the child to bed. The user has been proposed to read a picture book with the child, but the user seems to find it difficult to read a picture book with the child." is generated.

[0161] FIG. 6B shows the second responses of the learning models ML-1 to ML-3 considering the user's second request and the first response set obtained by the basic function.

[0162] Specifically, Figure 6B includes the second response [1] "Listening to music is also recommended." displayed in step S204. The second response [1] is the response of the expert model corresponding to the second requirement "It is difficult to read picture books together."

[0163] Also, Figure 6B includes the second response [2] "Music is also important for brain development." displayed in step S207 (k = 2). The second response [2] is the response of the scholar model corresponding to the second requirement "It is difficult to read picture books together." The second response [2] is generated by the scholar model with reference to the second response [1] "Listening to music is also recommended." of the expert model.

[0164] Also, Figure 6B includes the second response [3] "I think it is good to listen to slow music." displayed in step S207 (k = 3). The second response [3] is the response of the experienced model corresponding to the second requirement "It is difficult to read picture books together." The second response [3] is generated by the experienced model with reference to the second response [1] of the expert model and the second response [2] of the scholar model.

[0165] In the example of Figure 6B, since N = 3, the second responses [1] to [3] are displayed and the intervention function has ended. The second responses [1] to [3] constitute an example of the second response set.

[0166] In this way, the service control unit 205 acquires a second request from the user who has provided the first and second responses. The service control unit 205 generates a third query to be input to the first learning model based on the scene, the second request, and the first and / or second responses. The service control unit 205 inputs the generated third query to the first learning model to obtain a third response (second response [1]) from the first learning model. The service control unit 205 generates a fourth query to be input to the second learning model based on the scene, the second request, the third response (second response [1]), and the first and / or second responses. The service control unit 205 inputs the generated fourth query to the second learning model to obtain a fourth response (second response [2]) from the second learning model.

[0167] As described above, the service control unit 205 acquires the second request input by the user in response to the display of the first response set. The service control unit 205 sequentially uses each of the plurality of learning models to obtain a second response corresponding to the second request from each learning model. At this time, the service control unit 205 causes each of the second and subsequent learning models to refer to the second response obtained from at least one learning model used before the learning model. The service control unit 205 displays a second response set including the second responses obtained from each of the plurality of learning models on the terminal 20.

[0168] For this reason, the user can make a second request in response to and intervening in the responses obtained from the plurality of learning models. Also, in response to the user's intervention, a second response set can be obtained all at once from each learning model. Also, the second response is generated after referring to other second responses obtained from other learning models up to that point, similar to the first response. Therefore, a more useful second response set can be efficiently obtained compared to simply obtaining responses all at once from a plurality of learning models.

[0169] FIG. 18 is a diagram showing an example of the operation of the service control unit 205 according to the embodiment disclosed in the present application. Referring to FIG. 18, the operation regarding the flip function of the service control unit 205 will be described.

[0170] The flip function is executed following the basic function or the intervention function. For example, the intervention function starts in the state shown in FIG. 5. As shown in FIG. 18, the flip function includes steps S301 to S305.

[0171] In step S301, the service control unit 205 acquires the order change instruction input by the user in response to the display of the first response set. As described above, the order change instruction represents an instruction to change the order of using a plurality of learning models.

[0172] For example, as the order change instruction, the user may drag an arbitrary first response area displayed on the terminal 20 to the area indicating the changed order.

[0173] In FIG. 5, the user desires to obtain the first response from the expert model last. That is, the user desires to obtain a response from the expert model after referring to the first responses of the scholar model and the experienced model.

[0174] In this case, as shown in FIG. 8A, the user drags the first response [1] of the expert model so as to overlap the first response [3] of the experienced model obtained last at the current time. Thereby, an order change instruction for changing the order of using a plurality of learning models from the order of the expert model, the scholar model, and the experienced model to the order of the scholar model, the experienced model, and the expert model is input. Note that the operation for inputting the order change instruction is not limited to the above example.

[0175] In step S302 of FIG. 18, the service control unit 205 identifies the first response [k(new)] that needs to be updated by applying the changed order. Here, as the first response [k(new)] that needs to be updated, each response (first response [1] to first response [N]) may be identified. In this case, in other words, the basic function is repeated by applying the changed order.

[0176] Also, as the first response [k(new)] that needs to be updated, a part of the first responses [1] to [N] may be specified. For example, as the first response [k(new)] that needs to be updated, the first response operated by the user for giving an order change instruction may be specified. This is because it is highly likely that the user wants to know the new first response to which the changed order is applied for the first response operated by the user.

[0177] Also, the user may be able to select whether the first response [k(new)] that needs to be updated is all or a part of the first responses [1] to [N]. Here, the description will continue assuming that the first response [k(new)] that needs to be updated is the one in which the first response operated by the user is specified. In the example of Fig. 8A, the first response of the expert model operated by the user is specified as the first response [k(new)] that needs to be updated.

[0178] In step S303, the service control unit 205 generates a query based on the first request, the scene, and the first responses [1(new)] to [k(new)-1]. [1(new)] indicates that the changed order is the first. The first responses [1(new)] to [k(new)-1] may be those included in the previous first response set.

[0179] For example, in the example of Fig. 8A, the changed order of the expert model (learning model ML-1) is the third, and k(new)=3. Also, the first response [1(new)] is the first response [2] already obtained from the scholar model (learning model ML-2) whose changed order is the first. Also, the first response [k(new)-1]=the first response [2(new)] is the first response [3] already obtained from the experienced model (learning model ML-3) whose changed order is the second.

[0180] Based on the first request, scene, first response [2], and first response [3], the service control unit 205 generates a query. For example, a query "Bedtime scene; The user is troubled by the child's poor sleep. Scholars point out that relaxation time is also important academically. Experienced people suggest reading picture books with the child." is generated.

[0181] In step S304, the service control unit 205 inputs the generated query into the learning model ML[k(new)]. The service control unit 205 obtains a new first response [k(new)] from the learning model ML[k(new)].

[0182] In step S305, the service control unit 205 displays the obtained first response [k(new)] on the terminal 20. For example, the service control unit 205 may add it to the already displayed first response set and display the new first response [k(new)] on the terminal 20.

[0183] Also, the service control unit 205 may replace the first response [k] with the new first response [k(new)] in the already displayed first response set and display it. In this case, the service control unit 205 may rearrange the display order of the first response [1(new)] to the first response [k(new)] including the updated new first response [k(new)] in the already displayed first response set according to the changed order.

[0184] An example of the new first response [k(new)] displayed on the terminal 20 will be described with reference to FIG. 8B.

[0185] Compared with FIG. 8A, FIG. 8B is different in that it includes the first response [3(new)] "Not limited to reading picture books aloud, it is important to create a routine for relaxation." instead of the first response [1].

[0186] The first response [3(new)] is the response of the expert model corresponding to the first request "The child has trouble falling asleep." It is an updated version of the first response [1] of the first expert model. The first response [3(new)] is generated by referring to the first response [2] of the scholar model and the first response [3] of the experienced model, which were not accessible before the order was changed.

[0187] Also, in FIG. 8B, the display order of the first response [2] of the scholar model, the first response [3] of the experienced model, and the first response [3(new)] of the expert model is rearranged according to the changed order.

[0188] In the examples of FIGS. 8A and 8B, when the user is interested in the expert's thoughts, the user can change the order of using the expert model to a later order (e.g., the last). Thereby, the user can obtain a more useful first response [3(new)] of the expert model that refers to the first response [2] and the first response [3] of others instead of the first response [1] of the expert model that has not referred to the first response [2] and the first response [3] of others until then.

[0189] Note that the user is not limited to the order change instruction to make the order of the learning model of interest later as shown in FIG. 8A, and may also give an order change instruction to change the order of the learning model of no interest to an earlier order.

[0190] In this way, the service control unit 205 obtains an instruction (order change instruction) indicating to change the answer order of the first and second learning models from the user who has received the first and second responses. The service control unit 205 generates a third query to be input to the first learning model based on the scene, the first request, and the second response (the response of the second model before the order change). The service control unit 205 inputs the generated third query to the first learning model to obtain a third response (the response of the first learning model after the order change) from the first learning model.

[0191] As described above, the service control unit 205 acquires an instruction to change the order of using a plurality of learning models, which is a question (request) input by the user in response to the display of the first answer set. The service control unit 205 causes at least one of the plurality of learning models to refer to the first answers obtained from the learning models whose order after the change is earlier than that of the learning model. As a result, the service control unit 205 acquires a new first answer corresponding to the first request from the learning model. The service control unit 205 displays the new first answer obtained from the learning model on the terminal 20.

[0192] In this way, the user can change the order of using each learning model. As a result, the user can obtain a more useful new first answer from the learning model of interest. For example, the user can change the order of the learning model of interest to a later order (for example, the last), so that the learning model of interest can refer to the first answers of more other learning models. As a result, the user can efficiently obtain a more useful first answer from the learning model of interest.

[0193] The storage unit 206 is a means for storing information necessary for the operation of the server device 10. The storage unit 206 stores a plurality of scenes defined in advance. The storage unit 206 stores a plurality of learning models characterized by learning data.

[0194] [Terminal] A detailed description of the terminal 20 will be omitted. Examples of the terminal 20 include portable terminal devices such as smartphones, mobile phones, game machines, and tablets, and computers (personal computers, notebook personal computers), etc. The terminal 20 can be any device or apparatus as long as it can receive the user's operation and communicate with the server device 10.

[0195] Subsequently, a modification according to the first embodiment will be described.

[0196] <Modification 1 according to the first embodiment> The server device 10 may generate a history of scenes and learning models (scene selection history, etc.) selected by the user and a conversation history between the user and the learning model. The server device 10 may utilize the generated scene selection history, etc. and conversation history to provide services to the user.

[0197] When generating the scene selection history, etc., the service control unit 205 accumulates information about the user of the childcare support service (for example, user ID, gender, age) and the combination of the scene and learning model selected by the user. The service control unit 205 generates a scene selection history, etc. as shown in FIG. 19.

[0198] Alternatively, when generating the conversation history, the service control unit 205 stores the occurrence event and its content. For example, when storing the start of the service, the service control unit 205 stores user information, the selected scene, the selected learning model, etc. In addition, the service control unit 205 stores the utterances of the user and the learning model. Furthermore, the service control unit 205 stores the user's operations (intervention function, flip function) and their content. For example, the service control unit 205 generates a conversation history as shown in FIG. 20.

[0199] <Modification Example 2 According to the First Embodiment> In the above embodiment, the case where a system administrator or the like defines scenes in advance and registers the defined scenes in the server device 10 has been described. However, the server device 10 may generate new scenes. For example, the server device 10 may analyze the history information (for example, scene selection history, conversation history, etc.) obtained as a result of providing services (childcare support services) to a large number of users to generate (define) new scenes.

[0200] The scene management unit 203 analyzes the conversation history between the stored user and the learning model. For example, the scene management unit 203 calculates the average value of the number of questions (user requests) for each scene in one service provision. The scene management unit 203 determines whether there is a scene in which the calculated average value is greater than a predetermined value (threshold).

[0201] When the number of questions for a scene is greater than a predetermined value, the scene management unit 203 determines that the scene needs to be subdivided. The fact that many questions are required to obtain the answer expected by the user from the learning model indicates that there is room for improvement in the scene.

[0202] For example, when the average number of questions for the playtime scene is higher than the threshold value, the scene management unit 203 subdivides and redefines the playtime scene. At this time, the scene management unit 203 may subdivide the scene using the attribute information (for example, gender and age) of the user who selected the scene to be subdivided.

[0203] For example, even in the same scene, if the number of questions is significantly different between men and women, the scene management unit 203 may redefine (generate) the scene for each gender. Alternatively, if the number of questions varies depending on the age of the user, the scene management unit 203 may redefine (generate) the scene for each age group.

[0204] For example, the scene management unit 203 may generate scenes such as the playtime scene (father) and the playtime scene (mother) for the playtime scene.

[0205] Alternatively, when the account control unit 204 acquires information such as the occupation and interests of the user, the scene management unit 203 may generate a new scene using this attribute information. For example, the scene management unit 203 may generate scenes such as the playtime scene (full-time housewife), the playtime scene (self-employed), and the playtime scene (company employee).

[0206] In this way, the scene management unit 203 may generate a new scene based on the attribute information of the user, the conversation history between the user and the learning model, and the like.

[0207] <Modification Example 3 according to the First Embodiment> When the user selects a scene, the service control unit 205 may perform sorting of the scenes to be displayed in a list and filtering (selection) of the scenes to be displayed in a list according to the attribute information of the user and the like.

[0208] For example, the service control unit 205 may determine the order when displaying a list of scenes according to the age of the user. For example, the service control unit 205 reads out the age of the user who attempts to select a scene from the user management database (calculates the age from the date of birth).

[0209] The service control unit 205 analyzes the selection history of scene selection and the like, and calculates the number of selections by other users belonging to the age of the user for each scene. The service control unit 205 displays a list of scenes according to the calculated number of selections of the scenes. For example, the service control unit 205 displays a list of scenes in descending order of the number of selections (in the order of the most number of selections).

[0210] Alternatively, the service control unit 205 may determine whether to display a list of scenes according to the gender of the user. For example, as described above, when the play time scenes are separated by gender, the service control unit 205 may select the play time scenes to be displayed according to the gender of the user. Specifically, the service control unit 205 excludes the play time scenes for the gender different from the gender of the user.

[0211] Here, the service control unit 205 may use a pre-prepared learning model (scene selection model) regarding the selection and order of scenes presented to the user. The service control unit 205 may input the attribute information of the user (the user who has started using the childcare support service) into the scene selection model, and obtain the scenes and their order according to the attributes of the user from the scene selection model.

[0212] Note that, as the learning data used for the scene selection model, the selection history of scenes and the like and the conversation history of the childcare support service can be used. For example, the selection history of scenes and the like, which is a combination of a large amount of accumulated attribute information and the selected scenes, can be utilized as the learning data of the scene selection model.

[0213] Even when a large amount of learning data cannot be obtained, the server device 10 utilizes online optimization techniques represented by Bayesian optimization or the like, recommends a scene to the user at the start of service utilization by the user, and enables learning while obtaining feedback. Specifically, the server device 10 analyzes the conversation history between the user and the learning model (the user's instructions and feedback from the learning model), and adaptively selects a scene to propose to the user. Alternatively, the server device 10 may statistically select an optimal scene from the scenes already presented to the user and the selection rate of each scene, and present the selected scene to the user.

[0214] In this way, the user may select a scene according to their concerns or the like from the scenes pre-registered in the server device 10. Alternatively, the server device 10 may propose a scene to the user according to the user's attributes or the like.

[0215] <Modification Example 4 according to the First Embodiment> In the above embodiment, it has been described that the user selects a learning model to be the question partner when enjoying the childcare support service. However, this selection may not be made. The server device 10 may set each learning model pre-mounted on the device itself as the learning model that answers the user's questions.

[0216] Alternatively, the server device 10 may select a learning model to be the question partner for the user. For example, the service control unit 205 may select a plurality of learning models (combinations of learning models) to be the question partner for the user according to the user's attribute information (for example, gender, age). Alternatively, the service control unit 205 may select a plurality of learning models according to the scene selected by the user.

[0217] Alternatively, the server device 10 (service control unit 205) may narrow down the learning models presented to the user according to the user's attribute information, or may determine the order when displaying the learning models in a list. For example, the service control unit 205 may analyze the selection history of scenes and the like, and preferentially present learning models (learning models selected by many other users) that are popular among users of the same gender and age as the user.

[0218] <Modification Example 5 according to the First Embodiment> The server device 10 may calculate an index for evaluating each learning model according to the conversation with the user and the evaluation from the user. For example, the service control unit 205 acquires a confidence level (probability) together with the output (answer) from each learning model. The service control unit 205 stores the output and confidence level of the learning model in the conversation history.

[0219] For the scene selected by the user, the service control unit 205 refers to the conversation history and calculates the average value of the confidence levels output by each learning model. For a learning model whose average value is lower than the threshold, the service control unit 205 determines that it is inappropriate for the learning model to answer the scene selected by the user. The service control unit 205 makes the learning model determined to be inappropriate unavailable for the user to select or displays it at the lower position when displaying it in a list.

[0220] Alternatively, when the user ends the service, the service control unit 205 may acquire the evaluation of the learning model selected by the user. The service control unit 205 may conduct a questionnaire on the learning model. For example, the service control unit 205 acquires the user's evaluation of each learning model selected by the user in three levels: "good", "ordinary", and "bad".

[0221] The service control unit 205 quantifies the acquired evaluation and reflects the quantified evaluation in the learning model management database. For example, the service control unit 205 quantifies the evaluation such that good = 1 point, ordinary = 0 point, and bad = -1 point. The service control unit 205 stores the quantified evaluation in association with the scene (adds or subtracts it from the evaluation score of the scene).

[0222] When the service control unit 205 proposes a learning model to the user, it excludes from the proposal target the learning models with a low evaluation of the scene selected by the user.

[0223] In this way, when the server device 10 proposes a learning model to the user, it may select the learning model to be proposed to the user according to the evaluation of each learning model (confidence level for each scene, evaluation from the user).

[0224] <Modification Example 6 according to the First Embodiment> In the above embodiment, it has been described that the server device 10 realizes the conversation between the user and the learning model using the basic function, the intervention function, and the flip function. However, the server device 10 may realize the conversation between the user and the learning model using other functions. The server device 10 may provide a child-rearing support service to the user using other functions.

[0225] For example, the service control unit 205 of the server device 10 may be provided with a "give-up function", a "debate function", a "sum-up function", a "direct function", an "interaction function", and the like.

[0226] The give-up function is a function that, when the answer obtained from a certain learning model is not useful to the user, does not present (gives up) the answer. That the answer is not useful means a case where the confidence level of the answer output by the learning model is extremely low, etc.

[0227] The debate function is a function that obtains a second answer set by referring to the first answer set from each of a plurality of learning models. Note that the debate function may be further executed by referring to the second answer set and subsequent answer sets. By executing the debate function by referring to the nth answer set, the (n + 1)th answer set can be obtained. Note that the nth answer set in the debate function and the nth answer set in the above-described intervention function may be different because the information to be referred to by the learning model is different. Also, in the debate function, the second answer set and subsequent answer sets can be obtained without user intervention, whereas in the intervention function, the second answer set and subsequent answer sets can be obtained by user intervention.

[0228] The summary function is a function that presents a summary of the answers from a plurality of learning models. In the summary function, at least the first answer set is summarized, and when the nth answer set has been obtained, the first answer set to the nth answer set are summarized.

[0229] The direct function is a function that allows the user to directly specify a learning model and the answer that the learning model refers to. Specifically, the user selects one answer from among the answers from a plurality of learning models. Further, the user selects a learning model that answers (a learning model different from the learning model that gave the selected answer) by referring to the selected answer. The selected learning model outputs a new answer by referring to the answers of other learning models selected by the user. For example, in FIG. 5, consider the case where the user selects the expert model and the expert model answers by referring to the answer of the scholar model. In this case, the user drags the icon corresponding to the expert model to the icon of the scholar model or the answer of the scholar model. The expert model answers the query generated based on the user's first request, the scenario, and the first answer of the scholar model. Note that in the direct function, unlike the flip function, the order of the answers of the learning models does not change. In the example of FIG. 5 and the like, the order of the answers of the expert model, the scholar model, and the experienced model is maintained.

[0230] The dialogue function is a function that allows an operator related to any learning model to directly interact with the user on behalf of the learning model. The dialogue function may be started by the user's request or by the operator's request.

[0231] As described above, the situations that can occur regarding child-rearing are defined by a plurality of scenes, and the plurality of scenes are stored in the server device 10. The server device 10 acquires, from a user who wishes to use the child-rearing support service, a scene related to their concerns or the like. When obtaining an answer from the learning model selected by the user, the server device 10 generates a query to be input to the learning model based on the scene. That is, the server device 10 restricts the answer of the learning model by the scene. As a result, it is possible to prevent the answers from each learning model from deviating to topics or the like unintended by the user. In particular, generative AI such as large language models has the ability to answer even in situations that have little relevance to the situations assumed by the user, and without restrictions by the scene, there is a possibility of giving answers that the user does not expect. In this regard, the server device 10 guides (induces) the answer of the learning model to the answer expected by the user using the scene. Also, since the user can obtain an appropriate answer from the learning model by selecting a scene once, there is no need to input the background of the question in detail when inputting a question to the learning model. As a result, the convenience of the user using the child-rearing support service is improved.

[0232] In addition, the server device 10 answers the user's request using a plurality of learning models selected by the user. The plurality of learning models are characterized by learning data. The server device 10 can provide the user with a variety of answers based on different positions and different ideas. As a result, the user can refer to the answer that is most suitable for their situation or the like from among the plurality of answers.

[0233] [Second Embodiment] Subsequently, the second embodiment will be described in detail with reference to the drawings.

[0234] In the first embodiment, the case where the user and the learning model have a conversation regarding a specific theme and scene was described. In the second embodiment, the case where the user and a person have a conversation will be described. In the second embodiment, the user can actually have a conversation with a supporter who has previously consented to support the user and solve problems such as worries.

[0235] Note that since the configuration of the information processing system according to the second embodiment can be the same as that of the first embodiment, the description corresponding to FIG. 3 is omitted.

[0236] Hereinafter, the description will focus on the differences between the first embodiment and the second embodiment.

[0237] In the second embodiment, a history (conversation history) regarding the conversation between the user and the learning model is stored in the storage unit 206. The service control unit 205 stores, as a conversation history, requests and responses exchanged between the user and at least two or more learning models in the storage unit 206.

[0238] Also, the learning model management database according to the second embodiment stores information regarding candidates for supporters who support the user for each of the plurality of learning models. For example, the learning model management database stores the contact information of the sender of the learning data used for generating the learning model, who has consented to support the user.

[0239] For example, in the example of FIG. 11, when expert A consents to support the user (consents to become a supporter), the phone number, email address, etc. of the expert A are stored in the learning model management database as the contact information of the supporter candidate.

[0240] Alternatively, the learning model management database may store the profile (attribute information) of each supporter candidate. For example, the name, gender, date of birth, occupation, etc. of the supporter candidate may be stored in the learning model management database.

[0241] FIG. 21 is a diagram showing an example of a processing configuration (processing module) of the server device 10 according to the embodiment disclosed in the present application. Referring to FIG. 21, an assistance control unit 207 is added to the configuration of the server device 10 according to the first embodiment.

[0242] The assistance control unit 207 is a means for executing control related to the assistance of users of the childcare assistance service. For example, the assistance control unit 207 assists a user who wants to talk to an actual person (assistant) rather than talk to a learning model. More specifically, the assistance control unit 207 analyzes the conversation history to determine an assistant who assists the user from among a plurality of assistant candidates stored in the learning model management database.

[0243] The assistance control unit 207 displays a button for the user to press when the user wishes to talk to an assistant on the terminal 20 (see FIG. 22). For example, when the "Talk to Assistant" button shown in FIG. 22 is pressed, the assistance control unit 207 performs control to realize the conversation between the user and the assistant.

[0244] FIG. 23 is a flowchart showing an example of the operation of the assistance control unit 207 according to the embodiment disclosed in the present application. Referring to FIG. 23, the operation of the assistance control unit 207 will be described.

[0245] The assistance control unit 207 vectorizes the questions (requests) of the user (the user who wishes to talk to an assistant) stored in the conversation history (step S401).

[0246] Furthermore, the assistance control unit 207 identifies the learning models corresponding to the assistant candidates who have "agreed to be an assistant" in the learning model management database (step S402). For example, in the example of FIG. 11, if expert A and childcare experienced person C are set as assistant candidates, the expert model and the experienced person model corresponding to the two assistant candidates are identified.

[0247] The support control unit 207 vectorizes the answer of the identified learning model among the history of learning models stored in the conversation history (step S403). The answer used for vectorizing the said answer may be the answer output by the learning model to the user who wishes to have a conversation with the supporter, or may be the answer output to other users.

[0248] Note that since existing technologies can be applied to vectorize text data, detailed explanations are omitted. For example, existing embedding models such as Word2Vec, FastText, GloVe, etc., or transformer models such as BERT, RoBERTa, GPT, etc. can be used to vectorize the text data of users and learning models.

[0249] The support control unit 207 calculates the similarity between the vector generated from the user's question and at least one vector generated from the answer of each learning model (step S404). The support control unit 207 calculates metrics such as cosine similarity and Euclidean distance as the said similarity. By calculating the similarity, the support control unit 207 calculates the similarity between the troubles of the user and the answers of the supporter candidates.

[0250] The support control unit 207 determines a supporter who supports the user using the calculated similarity (step S405). For example, the supporter candidate corresponding to the learning model with the largest calculated similarity is determined as the supporter.

[0251] The support control unit 207 notifies the user of the contact information (phone number, email address) of the determined supporter (step S406). For example, the support control unit 207 displays the contact information of the supporter on the terminal 20.

[0252] The user calls the notified phone number or sends an email to the notified email address. The user contacts the supporter and consults the supporter about troubles and the like.

[0253] In this way, the support control unit 207 calculates a first vector (vector related to the user's request) related to the user's request stored in the conversation history. The support control unit 207 calculates at least two or more second vectors (vectors related to the responses of at least two or more learning models stored in the conversation history) related to the responses of at least two or more learning models stored in the conversation history. The support control unit 207 calculates the similarity between the first vector and each of the at least two or more second vectors. The support control unit 207 determines a supporter who supports the user from among the stored supporter candidates based on the calculated similarity. More specifically, the support control unit 207 determines, as the supporter who supports the user, the supporter candidate associated with the learning model corresponding to the second vector having the highest similarity to the first vector.

[0254] For a question from a user who wishes to talk to a supporter, a supporter corresponding to a learning model that gave a highly similar answer is likely to be able to provide appropriate advice or the like regarding the user's troubles and the like. Therefore, the support control unit 207 analyzes the conversation history between the user and the learning model, extracts the optimal supporter as the user's conversation partner, and notifies the user of the contact information etc. of the extracted supporter.

[0255] Subsequently, a modification according to the second embodiment will be described.

[0256] <Modification 1 according to the second embodiment> In the above embodiment, the case where the provider (sender) of the learning data that is the basis for learning model generation and the supporter candidate are the same person has been described. However, the provider (sender) of the learning data and the supporter candidate may be different persons.

[0257] For example, a person who has a similar idea to the provider (source of the learning data) of the learning data, or a person who agrees with the idea of the provider of the learning data may be stored in the learning model management database as a supporter candidate.

[0258] For example, a plurality of persons such as supporter candidate A1 and supporter candidate A2 who resonate with the ideas of expert A may be stored in the learning model management database as supporter candidates corresponding to the expert model. Alternatively, friends, acquaintances, etc. who belong to the same community as the childcare experienced person C may be stored in the learning model management database as supporter candidates.

[0259] When a plurality of supporter candidates are set for the learning model with the highest calculated similarity, the support control unit 207 may notify the user of the contact information of each of the plurality of supporter candidates. Alternatively, the support control unit 207 may notify the user of the contact information of the supporter candidate selected by the user from among the plurality of supporter candidates. At that time, the support control unit 207 may also display the profiles of each supporter candidate on the terminal 20.

[0260] <Modification Example 2 according to the Second Embodiment> When determining a supporter candidate or a supporter, the support control unit 207 may use the attribute information of the user and the attribute information of the supporter candidate.

[0261] For example, the support control unit 207 extracts supporter candidates having the same attributes as the attributes (such as gender and age) of the user from the learning model management database. The support control unit 207 may vectorize the answers of the learning models corresponding to the extracted supporter candidates and determine the supporter.

[0262] <Modification Example 3 according to the Second Embodiment> The server device 10 may provide an interface that presents a plurality of supporter candidates to the user and allows the user to select a final supporter from among the presented plurality of supporter candidates.

[0263] For example, the support control unit 207 extracts supporter candidates having the highest similarity among the calculated similarities. While displaying the extracted supporter candidates on the terminal 20, the support control unit 207 accepts the selection of a supporter from the user. The support control unit 207 displays the contact information of the supporter candidate selected by the user on the terminal 20.

[0264] <Modification Example 4 According to the Second Embodiment> When vectorizing the answer of the learning model, the support control unit 207 may weight the learning model itself or the answer of the learning model based on the user's operation or the like. For example, the support control unit 207 may calculate a vector by making the weight of the learning model whose answer order has been changed by the user's flip operation heavier than the weights of other learning models. In this case, the support control unit 207 makes the weight of each answer of the learning model heavier than the weights of the answers of other learning models.

[0265] Alternatively, the support control unit 207 may increase the weight of the answer of the learning model after the order change. The support control unit 207 may increase the weight only for the answer of the learning model performed immediately after the order change.

[0266] As described above, when calculating at least two or more second vectors, the support control unit 207 may assign weights to each of the at least two or more learning models. At that time, when the user changes the answer order of at least two or more learning models, the support control unit 207 may make the weight given to the learning model whose answer order has been changed by the user heavier than the weight given to the learning model whose answer order has not been changed by the user.

[0267] The learning model whose answer order has been changed by the flip function is a learning model with a high degree of attention from the user. For example, the learning model whose answer order is set last by the flip function is a learning model that the user is strongly interested in. Therefore, when vectorizing the answers of each learning model, the support control unit 207 increases the weight of the learning model that the user is strongly interested in, so as to more accurately extract supporters who are likely to solve the user's troubles and the like.

[0268] <Modification Example 5 According to the Second Embodiment> The server device 10 may provide a service to support a supporter who answers questions from the user.

[0269] Supporters, like the users who have contacted them, utilize the childcare support service provided by the server device 10. For example, when a supporter verifies or reinforces an answer to a question received from a user, the supporter utilizes the childcare support service.

[0270] For example, as shown in FIG. 24, the user and the supporter are having a conversation using each other's terminals 20. The user consults the supporter, saying, "I feel that my child's speech is slow. I'm concerned about this delay."

[0271] For example, the supporter answers the user's question with, "Have you consulted a doctor?"

[0272] At this time, the supporter inputs "Introduce a specialist doctor for developmental disorders" to the server device 10. The supporter selects an answer suitable for the user's situation from among the multiple answers (answers output by multiple learning models) obtained from the server device 10 and conveys it to the user. Also, the server device 10 may automatically generate a question based on the content input from the conversation history between the supporter and the user. For example, for the supporter's answer "Have you consulted a doctor?", the server device 10 may output to the supporter an answer to the automatically generated question "Introduce a specialist doctor for developmental disorders". With this function, the supporter can automatically obtain the necessary answers, enabling more rapid and accurate support.

[0273] Alternatively, the user may utilize the childcare support service when having a conversation with the supporter (during the conversation).

[0274] In this way, by the supporter utilizing the childcare support service, it becomes possible to provide more beneficial information to the user.

[0275] <Modification Example 6 according to the Second Embodiment> The support control unit 207 may not only introduce (notify the contact information) a supporter to a user who has troubles or the like, but also introduce products or services useful for solving the troubles or the like of the user.

[0276] For example, when the support control unit 207 detects a predetermined operation by the user (e.g., pressing a product introduction button), it introduces products or services corresponding to the user's past questions, etc. to the user.

[0277] For example, in the same manner as above, the support control unit 207 vectorizes the user's question and vectorizes the answer of the learning model. The support control unit 207 identifies the learning model that gave the answer closest to the user's concerns, etc. from the vectors of the user and at least one or more learning models.

[0278] The support control unit 207 generates a query for obtaining a product proposal corresponding to the user's past questions for the identified learning model. For example, the support control unit 207 generates a query such as "My child has trouble falling asleep. Please tell me recommended products." The support control unit 207 inputs the generated query to the identified learning model, so that a proposal of a product suitable for the user can be made from the learning model that gave an accurate answer to the user's concerns, etc.

[0279] As described above, the server device 10 according to the second embodiment supports the user using the conversation history, etc. between the user and the learning model. In particular, the server device 10 supports a user who wants to talk to a real person (supporter) rather than a learning model. At that time, the server device 10 analyzes the conversation history and introduces a supporter who is likely to give the most appropriate answer to the user's question (request). As a result, the user can obtain appropriate advice, etc. from a supporter who is familiar with the user's concerns, etc. That is, the convenience of the user is improved.

[0280] Subsequently, the hardware of each device constituting the information processing system will be described. FIG. 25 is a diagram showing an example of the hardware configuration of the server device 10.

[0281] The server device 10 can be configured by an information processing device (so-called computer) and has the configuration illustrated in FIG. 25. For example, the server device 10 includes a processor 311, a memory 312, an input / output interface 313, a communication interface 314, and the like. The components such as the processor 311 are connected by an internal bus or the like and are configured to be able to communicate with each other.

[0282] However, the configuration shown in FIG. 25 is not intended to limit the hardware configuration of the server device 10. The server device 10 may include hardware not shown, or may not include the input / output interface 313 if necessary. Also, the number of components such as the processor 311 included in the server device 10 is not intended to be limited to the example shown in FIG. 25. For example, a plurality of processors 311 may be included in the server device 10.

[0283] The processor 311 is, for example, a programmable device such as a CPU (Central Processing Unit), MPU (Micro Processing Unit), DSP (Digital Signal Processor), TPU (Tensor Processing Unit), GPU (Graphics Processing Unit). Alternatively, the processor 311 may be a device such as an FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit). The processor 311 executes various programs including an operating system (OS; Operating System).

[0284] The memory 312 is a RAM (Random Access Memory), ROM (Read Only Memory), HDD (Hard Disk Drive), SSD (Solid State Drive), or the like. The memory 312 stores an OS program, an application program, and various data.

[0285] The input / output interface 313 is an interface for a display device and an input device (not shown). The display device is, for example, a liquid crystal display or the like. The input device is a device that accepts user operations such as a keyboard and a mouse, for example.

[0286] The communication interface 314 is a circuit, a module, or the like that communicates with other devices. For example, the communication interface 314 includes a NIC (Network Interface Card) or the like.

[0287] The functions of the server device 10 are realized by various processing modules. The processing modules are realized, for example, when the processor 311 executes a program stored in the memory 312. Further, the program can be recorded on a computer-readable storage medium. The storage medium can be a non-transitory one such as a semiconductor memory, a hard disk, a magnetic recording medium, and an optical recording medium. That is, the present invention can also be embodied as a computer program product. Further, the above program can be downloaded via a network or updated using a storage medium storing the program. Furthermore, the above processing module may be realized by a semiconductor chip.

[0288] Note that the terminal 20 can also be configured by an information processing device in the same manner as the server device 10, and the basic hardware configuration thereof is the same as that of the server device 10, so the description thereof is omitted.

[0289] The server device 10, which is an information processing device, is equipped with a computer, and the functions of the server device 10 can be realized by causing the computer to execute a program. Further, the server device 10 executes a control method of the server device 10 according to the program.

[0290] [Modification Example] Note that the configurations, operations, and the like of the information processing system described in the above embodiment are examples, and are not intended to limit the configurations of the system and the like.

[0291] In the above embodiment, the operation of the information processing system according to the present disclosure was described by taking the childcare support service as an example. However, the information processing system according to the present disclosure is also applicable to services different from the childcare support service. For example, the information processing system according to the present disclosure is applicable to any service in which users such as job hunting support services, job transfer support services, and legal consultation services input questions, etc., and the learning model answers.

[0292] In the above embodiment, the case where one expert model is generated from the statements of one expert was described. However, it goes without saying that a plurality of expert models may be generated from the statements of a plurality of experts. Similarly, a plurality of scholar models may be generated from the statements of a plurality of scholars, or a plurality of experienced models may be generated from the statements of a plurality of childcare experienced persons.

[0293] Alternatively, one expert model may be generated from the statements of a plurality of experts. For example, a plurality of experts with similar thinking tendencies may be grouped into one group, and an expert model may be generated for each group.

[0294] When a plurality of learning models are selected, the server device 10 may provide an interface that allows the user to determine the order of answers for the plurality of learning models. That is, the user may determine the learning model for which they want to answer first or the learning model for which they want to answer last.

[0295] Before executing the basic functions, the server device 10 may execute control for determining the order of responses for a plurality of learning models selected by the user. Specifically, the server device 10 inputs a query based on the user's request and scene to each of the plurality of learning models selected by the user. The server device 10 obtains the confidence levels for the same query from each learning model. The server device 10 sets the order of the learning models with low confidence levels to the order of the learning models adopted by the basic model. For example, if the confidence levels of the expert model, scholar model, and experienced model are in the order of the expert model, experienced model, and scholar model from high to low, the server device 10 (service control unit 205) sets the response order of the learning models to the response order of the scholar model, experienced model, and expert model. The server device 10 may input the same query to each learning model, set the learning model with the lowest possibility of giving an appropriate response for the scene selected by the user as the first learning model, and set the learning model with the highest possibility of giving an appropriate response as the last learning model.

[0296] In the above embodiment, the case where a user management database or the like is configured inside the server device 10 has been described, but the database may be constructed in an external database server or the like. That is, some functions of the server device 10 may be implemented on another server. More specifically, it is sufficient that any device included in the system implements the "service control unit (service control means)" or the like described above.

[0297] The form of data transmission and reception between each device (for example, the server device 10 and the terminal 20) is not particularly limited, but the data transmitted and received between these devices may be encrypted. Between these devices, personal information of the user and the like is transmitted and received, and in order to appropriately protect this information, it is desirable that encrypted data is transmitted and received.

[0298] In the flowcharts (flowcharts, sequence diagrams) used in the above description, a plurality of steps (processes) are described in order, but the execution order of the steps executed in the embodiments is not limited to the order of the description. In the embodiments, for example, the order of the illustrated steps can be changed within a range that does not substantially affect the content, such as executing each process in parallel.

[0299] The above embodiments have been described in detail for ease of understanding of the present disclosure, and it is not intended that all the configurations described above are necessary. Also, when a plurality of embodiments are described, each embodiment may be used alone or in combination. For example, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace a part of the configuration of one embodiment with another configuration.

[0300] From the above description, the industrial applicability of the present invention is clear, and the present invention is suitably applicable to an information processing system that supports users, etc.

[0301] Some or all of the above embodiments may be described as follows in the following supplementary notes, but are not limited thereto.

[0302] [Supplementary Note 1] Request acquisition means for acquiring a first request for at least two or more learning models, which is a request related to a specific theme from a user; Answer acquisition means for acquiring an answer from each of the at least two or more learning models using at least a scene indicating a situation related to the specific theme and the first request; Answer providing means for providing the user with the answers obtained from each of the at least two or more learning models; A server device comprising the above. [Supplementary Note 2] The answer acquisition means Based on the scene and the first requirement, generate a first query to be input into a first learning model among the at least two or more learning models, and obtain a first answer from the first learning model by inputting the generated first query into the first learning model. Based on the scene, the first requirement, and the first answer, generate a second query to be input into a second learning model among the at least two or more learning models, and obtain a second answer from the second learning model by inputting the generated second query into the second learning model. The answer providing means is the server device according to Appendix 1 that provides the first and second answers to the user. [Appendix 3] The requirement acquisition means acquires a second requirement from the user to whom the first and second answers are provided. The answer acquisition means Based on the scene, the second requirement, the first and / or second answers, generate a third query to be input into the first learning model, and obtain a third answer from the first learning model by inputting the generated third query into the first learning model. The server device according to Appendix 2, which generates a fourth query to be input into the second learning model based on the scene, the second requirement, the third answer, the first and / or second answers, and obtains a fourth answer from the second learning model by inputting the generated fourth query into the second learning model. [Appendix 4] Further comprising a change instruction acquisition means for acquiring an instruction indicating a change in the answer order of the first and second learning models from the user to whom the first and second answers are provided. The answer acquisition means The server device according to Appendix 2, which generates a third query to be input into the first learning model based on the scene, the first requirement, and the second answer, and obtains a third answer from the first learning model by inputting the generated third query into the first learning model. [Appendix 5] Scene selection control means for obtaining a scene selected by the user from among a plurality of pre-defined scenes; Learning model selection control means for obtaining at least two or more learning models selected by the user from among a plurality of learning models characterized by learning data; The server device according to any one of Appendices 1 to 4, further comprising: [Appendix 6] Storage means for storing information regarding candidates for supporters who support the user for each of the at least two or more learning models; Support control means for determining the supporter who supports the user from among the stored candidates for supporters by analyzing a conversation history including requests and responses exchanged between the user and the at least two or more learning models; The server device according to Appendix 1. [Appendix 7] The support control means: Calculating a first vector regarding the request of the user stored in the conversation history, and calculating at least two or more second vectors regarding the responses of each of the at least two or more learning models included in the conversation history; Calculating a similarity between the first vector and each of the at least two or more second vectors; The server device according to Appendix 6, wherein the supporter who supports the user is determined from among the stored candidates for supporters based on the calculated similarity. [Appendix 8] The support control means determines, as the supporter who supports the user, the candidate for supporter stored in association with the learning model corresponding to the second vector having the highest similarity with the first vector; The server device according to Appendix 7. [Appendix 9] The support control means assigns weights to each of the at least two or more learning models when calculating the at least two or more second vectors; The server device according to Appendix 8. [Appendix 10] The support control means: The server device according to Supplementary Note 9, wherein when the user changes the order of answers of at least two or more learning models, the weight given to the learning model whose answer order has been changed by the user is made heavier than the weight given to the learning model whose answer order has not been changed by the user. [Supplementary Note 11] A request acquisition step of acquiring a first request for at least two or more learning models, which is a request from a user regarding a specific theme; An answer acquisition step of acquiring an answer from each of at least two or more learning models using at least a scene indicating a situation related to the specific theme and the first request; An answer providing step of providing the answers obtained from each of at least two or more learning models to the user. A control method for a server device, comprising: [Supplementary Note 12] The answer acquisition step includes: Based on the scene and the first request, generating a first query to be input to a first learning model among at least two or more learning models, and inputting the generated first query to the first learning model to obtain a first answer from the first learning model; Based on the scene, the first request, and the first answer, generating a second query to be input to a second learning model among at least two or more learning models, and inputting the generated second query to the second learning model to obtain a second answer from the second learning model; The answer providing step is to provide the first and second answers to the user. A control method for a server device according to Supplementary Note 11. [Supplementary Note 13] The request acquisition step acquires a second request from the user to whom the first and second answers have been provided; The answer acquisition step includes: Based on the scene, the second request, the first and / or second answers, generating a third query to be input to the first learning model, and inputting the generated third query to the first learning model to obtain a third answer from the first learning model; Based on the scene, the second request, the third response, the first and / or second responses, generate a fourth query to be input into the second learning model, and obtain a fourth response from the second learning model by inputting the generated fourth query into the second learning model, the control method of the server device described in Supplementary Note 12. [Supplementary Note 14] Further include a change instruction acquisition step of obtaining an instruction indicating a change in the response order of the first and second learning models from the user who provided the first and second responses. The response acquisition step is Based on the scene, the first request, and the second response, generate a third query to be input into the first learning model, and obtain a third response from the first learning model by inputting the generated third query into the first learning model, the control method of the server device described in Supplementary Note 12. [Supplementary Note 15] A scene selection control step of obtaining the scene selected by the user from among a plurality of predefined scenes; A learning model selection control step of obtaining at least two or more learning models selected by the user from among a plurality of learning models characterized by learning data; The control method of the server device according to any one of Supplementary Notes 11 to 14, further comprising [Supplementary Note 16] A storage step of storing information regarding candidates for supporters who support the user for each of the at least two or more learning models; Further include a support control step of determining the supporter who supports the user from among the stored supporter candidates by analyzing the conversation history including requests and responses exchanged between the user and the at least two or more learning models, the control method of the server device described in Supplementary Note 11. [Supplementary Note 17] The support control step is Calculate a first vector related to the user's request stored in the conversation history, and calculate at least two or more second vectors related to the responses of each of the at least two or more learning models included in the conversation history. Calculate the similarity between the first vector and each of the at least two or more second vectors. The server device according to Supplementary Note 16, which determines the supporter who supports the user from among the stored candidates for supporters based on the calculated similarity. [Supplementary Note 18] The support control step determines, as the supporter who supports the user, the candidate for the supporter stored in association with the learning model corresponding to the second vector having the highest similarity with the first vector, according to the control method of the server device described in Supplementary Note 17. [Supplementary Note 19] The support control step gives weights to each of the at least two or more learning models when calculating the at least two or more second vectors, according to the control method of the server device described in Supplementary Note 18. [Supplementary Note 20] The support control step When the user changes the order of responses of the at least two or more learning models, the weight given to the learning model whose response order has been changed by the user is made heavier than the weight given to the learning models whose response order has not been changed by the user, according to the control method of the server device described in Supplementary Note 19. [Supplementary Note 21] On a computer mounted on a server device A request acquisition process for acquiring a first request regarding a specific theme from the user and addressed to at least two or more learning models. An answer acquisition process for acquiring answers from each of the at least two or more learning models using at least a scene indicating a situation related to the specific theme and the first request. An answer providing process for providing the user with the answers obtained from each of the at least two or more learning models. A program for causing the above to be executed. [Supplementary Note 22] The response acquisition process generates a first query to be input to a first learning model among the at least two or more learning models based on the scene and the first request, and obtains a first response from the first learning model by inputting the generated first query to the first learning model, generates a second query to be input to a second learning model among the at least two or more learning models based on the scene, the first request, and the first response, and obtains a second response from the second learning model by inputting the generated second query to the second learning model, The response providing process is a program described in Supplementary Note 21 that provides the first and second responses to the user. [Supplementary Note 23] The request acquisition process obtains a second request from the user to whom the first and second responses are provided, The response acquisition process generates a third query to be input to the first learning model based on the scene, the second request, and the first and / or second responses, and obtains a third response from the first learning model by inputting the generated third query to the first learning model, A program described in Supplementary Note 22 that generates a fourth query to be input to the second learning model based on the scene, the second request, the third response, and the first and / or second responses, and obtains a fourth response from the second learning model by inputting the generated fourth query to the second learning model. [Supplementary Note 24] Further execute a change instruction acquisition process that obtains an instruction indicating a change in the response order of the first and second learning models from the user to whom the first and second responses are provided, The response acquisition process generates a third query to be input to the first learning model based on the scene, the first request, and the second response, and obtains a third response from the first learning model by inputting the generated third query to the first learning model, which is a program described in Supplementary Note 22. [Appendix 25] A scene selection control process for obtaining a scene selected by the user from among a plurality of predefined scenes, and A learning model selection control process for obtaining at least two or more learning models selected by the user from among a plurality of learning models characterized by learning data, and The program according to any one of Appendices 21 to 24, which further causes the above to be executed. [Appendix 26] A storage process for storing information regarding candidates for supporters who support the user for each of the at least two or more learning models, and A support control process for determining the supporter who supports the user from among the stored candidates for supporters by analyzing a conversation history including requests and responses exchanged between the user and the at least two or more learning models, and The program according to Appendix 21, which further causes the above to be executed. [Appendix 27] The support control process includes: Calculating a first vector regarding the request of the user stored in the conversation history, and calculating at least two or more second vectors regarding the responses of each of the at least two or more learning models included in the conversation history, Calculating the similarity between the first vector and each of the at least two or more second vectors, and The program according to Appendix 26, which determines the supporter who supports the user from among the stored candidates for supporters based on the calculated similarity. [Appendix 28] The support control process determines, as the supporter who supports the user, the candidate for supporter stored in association with the learning model corresponding to the second vector having the highest similarity with the first vector, according to the program of Appendix 27. [Appendix 29] The support control process gives weights to each of the at least two or more learning models when calculating the at least two or more second vectors, according to the program of Appendix 28. [Appendix 30] The support control process When the user changes the order of answers of at least two or more learning models, the weight given to the learning model whose answer order the user has changed is made heavier than the weight given to the learning models whose answer order the user has not changed, the program described in Supplementary Note 29.

[0303] In addition, part or all of the configurations described in Supplementary Notes 2 to 10 subordinate to Supplementary Note 1 described above may be subordinate to Supplementary Notes 11 and 21 in the same subordinate relationship as Supplementary Notes 2 to 10. Furthermore, not limited to Supplementary Notes 1, 11, and 21, within the scope not departing from the above-described embodiments, similarly for various hardware, software, various recording means for recording software, or systems, part or all of the configurations described as supplementary notes may be made subordinate.

[0304] Note that each disclosure of the above-cited prior art documents is incorporated herein by reference. As described above, although the embodiments of the present invention have been described, the present invention is not limited to these embodiments. It will be understood by those skilled in the art that these embodiments are merely examples and that various modifications are possible without departing from the scope and spirit of the present invention. That is, the present invention naturally includes all disclosures including the claims, various modifications and corrections that can be made by those skilled in the art according to the technical idea.

Explanation of Signs

[0305] 10 Server device 20 Terminal 100 Server device 101 Request acquisition means 102 Answer acquisition means 103 Answer providing means 201 Communication control unit 202 Learning model management unit 203 Scene management unit 204 Account control unit 205 Service control unit 206 Storage unit 207 Support Control Unit 311 Processor 312 Memory 313 Input / Output Interface 314 Communication Interface

Claims

1. A request acquisition means for acquiring a first request for at least two or more learning models, which is a request regarding a specific theme from a user; An answer acquisition means for acquiring an answer from each of the at least two or more learning models by using at least a scene indicating a situation related to the specific theme and the first request; An answer providing means for providing the answers obtained from each of the at least two or more learning models to the user; A server device comprising the above.

2. The answer acquisition means: Based on the scene and the first request, generate a first query to be input to a first learning model among the at least two or more learning models, and input the generated first query to the first learning model to obtain a first answer from the first learning model; Based on the scene, the first request, and the first answer, generate a second query to be input to a second learning model among the at least two or more learning models, and input the generated second query to the second learning model to obtain a second answer from the second learning model; The answer providing means provides the first and second answers to the user. The server device according to Claim 1.

3. The request acquisition means acquires a second request from the user to whom the first and second answers are provided; The answer acquisition means: Based on the scene, the second request, the first and / or second answers, generate a third query to be input to the first learning model, and input the generated third query to the first learning model to obtain a third answer from the first learning model; Based on the scene, the second request, the third answer, the first and / or second answers, generate a fourth query to be input to the second learning model, and input the generated fourth query to the second learning model to obtain a fourth answer from the second learning model. The server device according to Claim 2.

4. Further comprising a change instruction acquisition means for acquiring an instruction indicating to change the answer order of the first and second learning models from the user to whom the first and second answers are provided; The answer acquisition means: Based on the scene, the first request, and the second response, generate a third query to be input into the first learning model, and obtain a third response from the first learning model by inputting the generated third query into the first learning model. The server device according to claim 2.

5. Scene selection control means for obtaining the scene selected by the user from among a plurality of predefined scenes; Learning model selection control means for obtaining at least two or more learning models selected by the user from among a plurality of learning models characterized by learning data; The server device according to any one of claims 1 to 4, further comprising:

6. Storage means for storing information regarding candidates for supporters who assist the user for each of the at least two or more learning models; The server device according to claim 1, further comprising support control means for determining the supporter who assists the user from among the stored candidates for supporters by analyzing a conversation history including requests and responses exchanged between the user and the at least two or more learning models.

7. The support control means: Calculate a first vector regarding the request of the user stored in the conversation history, and calculate at least two or more second vectors regarding the responses of each of the at least two or more learning models included in the conversation history; Calculate the similarity between the first vector and each of the at least two or more second vectors; The server device according to claim 6, determining the supporter who assists the user from among the stored candidates for supporters based on the calculated similarity.

8. The support control means determines, as the supporter who assists the user, the candidate for supporter stored in association with the learning model corresponding to the second vector having the highest similarity with the first vector. The server device according to claim 7.

9. The support control means assigns weights to each of the at least two or more learning models when calculating the at least two or more second vectors. The server device according to claim 8.

10. The support control means: The server device according to claim 9, wherein when the user changes the order of responses of the at least two or more learning models, the weight given to the learning model whose response order has been changed by the user is made heavier than the weight given to the learning model whose response order has not been changed by the user.

11. A request acquisition step of acquiring a first request for at least two or more learning models, which is a request regarding a specific theme from a user; A response acquisition step of acquiring a response from each of the at least two or more learning models using at least a scene indicating a situation related to the specific theme and the first request; A response providing step of providing the user with the responses obtained from each of the at least two or more learning models. A control method for a server device, comprising:

12. A program for causing a computer mounted on a server device to execute a request acquisition process of acquiring a first request for at least two or more learning models, which is a request regarding a specific theme from a user; execute a response acquisition process of acquiring a response from each of the at least two or more learning models using at least a scene indicating a situation related to the specific theme and the first request; execute a response providing process of providing the user with the responses obtained from each of the at least two or more learning models. ​

Images