Information processor and information processing method and information processing program

Abstract

To realize an information processor that allows a user to obtain useful information more efficiently from multiple services.SOLUTION: The information processor comprises an input acquisition part that acquires a user's first request, a reply acquisition part that, when acquiring a first response corresponding to the first request from each of a first machine learning model and a second machine learning model, makes the first response obtained from the first machine learning model referred to by the second machine learning model, and a display control part that displays the first response set that includes the first response obtained from each of the first machine learning model and the second machine learning model on a display device.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a technique for generating answers in response to a user request. [Background technology]

[0002] There are known techniques for generating responses in response to user requests. As an example, Patent Document 1 describes an automatic response AI that analyzes the content of a user's comments and provides the user with a response to the content of the comments. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-49900 A Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, many organizations have been providing services that provide answers to user requests that are suited to their own organizations, using technology such as that described in Patent Literature 1. However, there is a problem in that when a user wants to compare answers from multiple services, the user cannot efficiently obtain answers from each service.

[0005] The present disclosure has been made in consideration of the above problems, and an exemplary objective thereof is to provide a technique that enables a user to more efficiently obtain useful information from multiple services. [Means for solving the problem]

[0006] An information processing device according to an exemplary aspect of the present disclosure includes an input acquisition means for acquiring a first request from a user, an answer acquisition means for causing a second machine learning model to refer to the first answer obtained from the first machine learning model when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, and a display control means for displaying a first answer set including the first answers obtained from each of the first machine learning model and the second machine learning model on a display device.

[0007] An information processing method according to an exemplary aspect of the present disclosure includes an input acquisition process in which at least one processor acquires a first request from a user; an answer acquisition process in which the at least one processor, when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, refers to the first answer obtained from the first machine learning model to the second machine learning model; and a display control process in which the at least one processor displays a first answer set including the first answers obtained from each of the first machine learning model and the second machine learning model on a display device.

[0008] A program according to an exemplary aspect of the present disclosure is a program that causes a computer to function as an information processing device, and causes the computer to function as an input acquisition means for acquiring a first request from a user, an answer acquisition means for referencing the first answer obtained from the first machine learning model to the second machine learning model when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, and a display control means for displaying a first answer set including the first answers obtained from each of the first machine learning model and the second machine learning model on a display device. Effect of the Invention

[0009] According to an exemplary aspect of the present disclosure, an exemplary effect is provided in that a technology can be provided that enables a user to more efficiently obtain useful information from multiple services. [Brief description of the drawings]

[0010] [Figure 1] 1 is a block diagram showing a configuration of an information processing device according to the present disclosure. [Diagram 2] 1 is a flow diagram showing the flow of an information processing method according to the present disclosure. [Diagram 3] 1 is a block diagram showing a configuration of an information processing system according to the present disclosure. [Figure 4] FIG. 2 is a block diagram showing a configuration of a user terminal according to the present disclosure. [Diagram 5] FIG. 2 is a block diagram showing a configuration of a server according to the present disclosure. [Figure 6] FIG. 2 is a block diagram showing a configuration of a management terminal according to the present disclosure. [Figure 7] 1 is a flow diagram illustrating the flow of an information processing method according to the present disclosure. [Figure 8] FIG. 13 is a diagram showing an example of a screen in a basic function according to the present disclosure. [Figure 9] 1 is a flow diagram illustrating the flow of an information processing method according to the present disclosure. [Figure 10] FIG. 13 is a diagram illustrating an example of a screen in an intervention function according to the present disclosure. [Figure 11] FIG. 13 is a diagram showing an example of a screen to which a give-up function according to the present disclosure is applied. [Figure 12] 1 is a flow diagram illustrating the flow of an information processing method according to the present disclosure. [Figure 13] 11A and 11B are diagrams illustrating example screens in a flip function according to the present disclosure. [Figure 14] 1 is a flow diagram illustrating the flow of an information processing method according to the present disclosure. [Figure 15] FIG. 13 is a diagram showing an example of a screen in a debate function according to the present disclosure. [Figure 16] 1 is a flow diagram illustrating the flow of an information processing method according to the present disclosure. [Figure 17] FIG. 13 is a diagram showing an example of a screen in a thumbs-up function according to the present disclosure. [Figure 18] FIG. 13 is a diagram showing an example of a screen in a thumbs-up function according to the present disclosure. [Figure 19] 1 is a flow diagram illustrating the flow of an information processing method according to the present disclosure. [Figure 20] FIG. 13 is a diagram showing an example of a screen in a dialogue function according to the present disclosure. [Figure 21] 1 is a block diagram showing a variation of the device configuration according to the present disclosure. [Figure 22] 1 is a block diagram showing a variation of the device configuration according to the present disclosure. [Diagram 23] 1 is a block diagram showing a variation of the device configuration according to the present disclosure. [Figure 24] FIG. 2 is a block diagram showing the hardware configuration of a computer that functions as each device according to the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, an embodiment obtained by appropriately combining the technical means employed in each of the exemplary embodiments shown below may also be included in the scope of the present invention. In addition, an embodiment obtained by appropriately omitting a part of the technical means employed in each of the exemplary embodiments shown below may also be included in the scope of the present invention. In addition, the effects mentioned in each of the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, an embodiment that does not exhibit the effects mentioned in each of the exemplary embodiments shown below may also be included in the scope of the present invention.

[0012] First Exemplary Embodiment A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is the basic form of each exemplary embodiment described later. The scope of application of each technical means adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means adopted in this exemplary embodiment can be adopted in other exemplary embodiments included in this disclosure to the extent that no particular technical obstacle occurs. In addition, each technical means shown in the drawings referred to for explaining this exemplary embodiment can also be adopted in other exemplary embodiments included in this disclosure to the extent that no particular technical obstacle occurs.

[0013] (Configuration of information processing device) The configuration of the information processing device 1 will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the information processing device 1. As shown in Fig. 1, the information processing device 1 includes an input acquisition unit 11, an answer acquisition unit 12, and a display control unit 13.

[0014] The input acquisition unit 11 acquires a first request from a user. When acquiring a first answer corresponding to the first request from each of the first machine learning model and the second machine learning model, the answer acquisition unit 12 causes the second machine learning model to refer to the first answer obtained from the first machine learning model. The display control unit 13 displays a first answer set including the first answers obtained from each of the first machine learning model and the second machine learning model on a display device.

[0015] (Effects of information processing device) As described above, in the information processing device 1, the input acquisition unit 11 acquires the first request of the user, and when the answer acquisition unit 12 acquires the first answer corresponding to the first request from each of the first machine learning model and the second machine learning model, the first answer acquired from the first machine learning model is referred to the second machine learning model, and the display control unit 13 displays the first answer set including the first answers acquired from each of the first machine learning model and the second machine learning model on the display device. Therefore, according to the information processing device 1, an effect is obtained in which the user can obtain useful information more efficiently from a plurality of services. Specifically, since the first answer acquired from the second machine learning model is generated with reference to the first answer acquired from the first machine learning model, a more useful first answer set can be obtained efficiently compared to the case where answers are simply acquired from the first machine learning model and the second machine learning model all at once.

[0016] (Flow of information processing method) The flow of the information processing method S1 will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the information processing method S1. As shown in Fig. 2, the information processing method S1 includes an input acquisition process S11, an answer acquisition process S12, and a display control process S13.

[0017] In an input acquisition process S11, at least one processor acquires a first request from a user.

[0018] In the answer acquisition process S12, when obtaining a first answer corresponding to a first request from each of the first machine learning model and the second machine learning model, the at least one processor refers to the first answer obtained from the first machine learning model in the second machine learning model.

[0019] In a display control process S13, the at least one processor displays a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model on a display device.

[0020] (Effects of information processing methods) As described above, in the information processing method S1, in the input acquisition process S11, at least one processor acquires a first request from a user, and in the answer acquisition process S12, when the at least one processor acquires a first answer corresponding to the first request from each of the first machine learning model and the second machine learning model, the at least one processor causes the second machine learning model to refer to the first answer obtained from the first machine learning model, and in the display control process S13, the at least one processor displays a first answer set including the first answers obtained from each of the first machine learning model and the second machine learning model on a display device. Therefore, according to the information processing method S1, an effect is obtained in which a user can obtain useful information more efficiently from a plurality of services. Specifically, since the first answer obtained from the second machine learning model is generated with reference to the first answer obtained from the first machine learning model, a more useful first answer set can be obtained efficiently compared to the case where answers are simply obtained from the first machine learning model and the second machine learning model all at once.

[0021] Second Exemplary Embodiment A second exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be given the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technical means adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means adopted in this exemplary embodiment can be adopted in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs. In addition, each technical means shown in each drawing referred to for explaining this exemplary embodiment can be adopted in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs.

[0022] (Overview of Information Processing System 1A) The information processing system 1A is a system that provides an answer service that presents answers from multiple machine learning models ML in response to a user request. The multiple machine learning models ML include the above-mentioned first machine learning model and second machine learning model. In other words, the number of machine learning models ML may be two, or may be three or more. When focusing on two of the multiple machine learning models ML, the exemplary embodiment 2 described below can be similarly described by replacing "multiple machine learning models ML" with "first machine learning model and second machine learning model."

[0023] The information processing system 1A includes a user terminal 10 and a plurality of servers 20. The inclusion of a plurality of servers 20 also plays a role in ensuring load balancing and / or redundancy. For example, each server 20 is managed by a company that provides services to users, and utilizes a machine learning model ML to provide part of the company's services. Each server 20 generates an answer corresponding to a request from a user using the machine learning model ML. The user terminal 10 functions as an interface between the user and the server 20. The user terminal 10 receives a request from the user, generates a query based on the request, and transmits the generated query to the server 20. In addition, the server 20 generates an answer corresponding to the query using the machine learning model ML, and transmits the generated answer to the user terminal 10. As a result, the answer corresponding to the request is transmitted to the user. The information processing system 1A can efficiently convey information to the user using answers from a plurality of machine learning models. As a result, the user can obtain the necessary information in a short time.

[0024] In this exemplary embodiment, the information processing system 1A provides a basic function of a reply service, an intervention function, a flip function, a give-up function, a debate function, a thumbs-up function, and an interaction function.

[0025] The basic function is a function that sequentially obtains answers from multiple machine learning models ML in response to a user request. In the basic function, answers are obtained from the second and subsequent machine learning models ML by referencing answers from each machine learning model ML obtained up to that point. The request and answer in the basic function are referred to as a first request and a first answer. A first answer is obtained from each of the multiple machine learning models ML in response to one first request. The multiple first answers may be different from each other. The multiple first answers are collectively referred to as a first answer set.

[0026] The intervention function is a function in which a user adds (intervenes in) a request to a first answer obtained by the basic function, and thereby obtains new answers from multiple machine learning models ML in sequence. In the intervention function, a new answer is obtained from the second and subsequent machine learning models ML by referring to the new answers from each machine learning model ML obtained up to that point. The request and new answer added in the intervention function are referred to as a second request and a second answer. A second answer is obtained from each of the multiple machine learning models ML for one second request. The multiple second answers are collectively referred to as a second answer set. Note that the intervention function may be further executed for the second answer and onwards. A new request and a new answer in the intervention function made for the nth answer are referred to as the n+1th request and the n+1th answer, and the multiple n+1th answers are collectively referred to as the n+1th answer set.

[0027] The flip function is a function that allows the user to change (flip) the order in which multiple machine learning models ML are used for an answer obtained by the basic function. In the flip function, for at least one machine learning model ML, a new first answer is obtained by referring to the first answers already obtained from each machine learning model ML that is earlier in the changed order than the machine learning model ML.

[0028] The give-up function is a function that does not present (give up) an answer obtained from a certain machine learning model ML if the answer is unfavorable. An unfavorable answer means that the answer is less favorable than the answers obtained from at least one of the machine learning models ML used up to that point.

[0029] The debate function is a function that acquires a second answer set by referring to the first answer set from each of a plurality of machine learning models ML. The debate function may be further executed by referring to the second answer set and onwards. The n+1th answer set is obtained by executing the debate function by referring to the nth answer set. The nth answer set in the debate function and the nth answer set in the intervention function described above may differ because the information referred to by the machine learning model ML is different. In addition, the debate function obtains the second answer set and onwards without the user's intervention, whereas the intervention function obtains the second answer set and onwards by the user's intervention.

[0030] Such a debate function can be used, for example, in a situation where a user is forced to choose between products offered by two specific companies. By using information processing system A, the user can have the machine learning models of each company debate the two products. The user can make a selection based not only on one-sided information provided by each company, but also on specific comparison points and relative advantages.

[0031] The thumbs-up function is a function that presents a summary (thumbs-up) of answers from multiple machine learning models ML. In the thumbs-up function, at least the first answer set is summarized, and if up to the nth answer set are obtained, the first through nth answer sets are summarized.

[0032] The dialogue function is a function in which an operator associated with any machine learning model ML directly dialogues with a user on behalf of the machine learning model ML. The dialogue function may be started at the request of a user or at the request of an operator. In this exemplary embodiment, the case where the dialogue function is started at the request of an operator will be mainly described.

[0033] (Configuration of information processing system 1A) The configuration of the information processing system 1A will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the information processing system 1A. The information processing system 1A includes a user terminal 10, multiple servers 20-i (i=1, 2, ..., N: N is a natural number equal to or greater than 2), and multiple management terminals 30-i. The user terminal 10 is an example of an information processing device. Note that, although Fig. 3 shows one user terminal 10, the number of user terminals 10 included in the information processing system 1A may be two or more.

[0034] The user terminal 10 and the server 20-i are connected to each other so as to be able to communicate with each other via a network NW. The network NW may be configured to include, but is not limited to, a wireless LAN (Local Area Network), a wired LAN, a WAN, or a mobile data communication network. The server 20-i and the management terminal 30-i are connected to each other so as to be able to communicate with each other. The server 20-i and the management terminal 30-i may be connected by an in-house LAN (not shown) that manages the server 20-i, or may be connected via the network NW. Note that, although FIG. 3 shows an example in which one management terminal 30-i is connected to one server 20-i, multiple management terminals 30-i may be connected to one server 20-i.

[0035] The server 20-i also stores the machine learning model ML-i. Instead of being stored in the server 20-i, the machine learning model ML-i may be communicably connected to the server 20-i. In this case, the machine learning model ML-i and the server 20-i may be connected by the above-mentioned in-house LAN or may be connected via the network NW.

[0036] Hereinafter, when there is no need to distinguish between the multiple servers 20-i, each will be simply referred to as a server 20. Furthermore, when there is no need to distinguish between the multiple management terminals 30-i, each will be simply referred to as a management terminal 30. Furthermore, when there is no need to distinguish between the multiple machine learning models ML-i, each will be simply referred to as a machine learning model ML.

[0037] (Machine learning model ML) The machine learning model ML is a model that generates an answer corresponding to a request. As an example, when a query generated based on a request is input, the machine learning model ML outputs an answer corresponding to the query. As an example, the machine learning model ML is configured by a language model. The language model may be, for example, a model called LLM (Large Language Models), but is not limited to this. The language model is a generative model that receives language as input and outputs language as output. The language model is a model that learns the relationship between words in a sentence, and generates a related string related to a target string from the target string. By using a language model that has been trained with sentences and sentences of various contexts, it is possible to generate a related string with appropriate content related to the target string.

[0038] For example, a case where a language model is used in question answering will be described. The language model receives an input of a question such as "What kind of country is Japan?" as a target string. The language model generates a string such as "Japan is an island country in the northern hemisphere..." as an answer to the question.

[0039] The learning method of the language model is not particularly limited, but as an example, it may be one that is learned to output at least one sentence including an input string. As a specific example, the language model is GPT-2 (Generative Pre-Training-2) or GPT-3, which outputs a sentence including an input string by predicting a string that is likely to follow the input string. In addition, 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.

[0040] Furthermore, the character string generated by the language model is not limited to a natural language. For example, the language model may output an artificial language (such as a program source code) for a character string input in a natural language. For example, the language model accepts an input question such as "How can I retrieve data containing a specific character string from a database?" as a target character string. The language model may output a program source code for performing database processing. Alternatively, the language model may output a natural language corresponding to a character string input in an artificial language. Furthermore, the content generated by the language model is not limited to a character string. For example, the language model may generate image data, video data, audio data, or other data formats corresponding to the input character string.

[0041] The machine learning model ML-i is managed by an organization different from the other machine learning models ML-j (i ≠ j), and is trained to generate answers to queries that fit the organization. For example, when the same query is input to the machine learning models ML-i and Mj, the answers that each model outputs may be different. For example, assume that the machine learning model ML-1 is managed by a company "Company A" and the machine learning model ML-2 is managed by a company "Company B". Also assume that Company A sells stuffed dog toys for 100 yen each, and Company B sells stuffed dog toys for 80 yen each. In this case, in response to a query such as "I want a stuffed dog toy," the machine learning model ML-1 outputs an answer of "100 yen each," and the machine learning model ML-2 outputs an answer of "80 yen each." Note that the organization that manages each machine learning model ML-i is hereinafter also referred to as the management organization. The management organization is, for example, a company, but is not limited to this.

[0042] Moreover, each of the multiple machine learning models ML is used in order by the user terminal 10 described later. Hereinafter, the machine learning model ML used kth will be referred to as the machine learning model ML[k] (k=1, 2, 3, ...). The order in which the multiple machine learning models ML are used may be changed. For this reason, the subscript "i" that identifies the machine learning model ML-i does not necessarily match the "k" that indicates the order in which the machine learning model ML-i is applied. For example, the machine learning model ML-1 may be the machine learning model ML[1] that is used first, or the machine learning model ML[N] that is used Nth. The server 20-i that stores the machine learning model ML[k] that is used kth may be referred to as the server 20[k].

[0043] The order in which the multiple machine learning models ML are used may be predetermined or may be dynamically determined. When dynamically determined, for example, the order may be a random order determined by the answer acquisition unit 112, or an order may be determined according to a user's instruction. The order may be changeable according to a user's instruction. For example, the order may be determined based on advertising fees from a management organization that manages the machine learning models ML. For example, a later order can refer to first answers from more other machine learning models ML, and a more useful first answer can be generated. Therefore, the order may be determined so that the management organization with a higher advertising fee is later.

[0044] (Configuration of user terminal 10) The configuration of the user terminal 10 will be described with reference to FIG. 4. The user terminal 10 is a computer used by a user. The user terminal 10 includes a control unit 110, a storage unit 120, a communication unit 130, an input device 140, and a display device 150. The control unit 110 controls each unit of the user terminal 10. The storage unit 120 stores various data and programs used by the control unit 110. The communication unit 130 communicates with other devices under the control of the control unit 110. The input device 140 may be configured by a mouse, a touch pad, a keyboard, or a combination of some or all of these. The display device 150 may be configured by a liquid crystal display or an organic EL (Electro Luminescence) display. The input device 140 and the display device 150 may be configured by a touch panel formed integrally. In addition, some or all of the storage unit 120, the communication unit 130, the input device 140, and the display device 150 may be built into the user terminal 10, or may be connected to the outside as peripheral devices. In addition, the user terminal 10 may further include a microphone, a speaker, etc., or these may be connected to the outside.

[0045] The control unit 110 includes an input acquisition unit 111, an answer acquisition unit 112, a summary generation unit 113, a display control unit 114, a history transmission unit 115, and an interactive user interface (hereinafter, UI) unit 116. The input acquisition unit 111 is an example of a configuration that realizes an input acquisition means. The answer acquisition unit 112 is an example of a configuration that realizes an answer acquisition means. The summary generation unit 113 is an example of a configuration that realizes a summary generation means. The display control unit 114 is an example of a configuration that realizes a display control means. The history transmission unit 115 is an example of a configuration that realizes a history transmission means. The interactive UI unit 116 is an example of a configuration that realizes an interactive request receiving means.

[0046] The input acquisition unit 111 acquires a first request from a user. Note that the "nth request from the user" (n is a natural number) may be information input by the user from the input device 140, or may be information previously stored in the storage unit 120. The first request may include an instruction to specify multiple machine learning models ML from which an answer is to be acquired among the multiple machine learning models ML (hereinafter, also referred to as a target specification instruction).

[0047] Furthermore, the input acquiring unit 111 may acquire a second request of the user in response to the display of the first answer set. The first answer set includes a first answer generated by each of the multiple machine learning models ML in response to the first request. Note that the input acquiring unit 111 may thereafter repeatedly acquire the (n+1)th request of the user in response to the display of the (n)th answer set. When n is 2 or more, the (n)th answer set includes the (n-1)th answer generated by each of the multiple machine learning models ML in response to the (n-1)th request.

[0048] Furthermore, the input acquisition unit 111 may acquire a user instruction (hereinafter also referred to as an order change instruction) indicating a change in the order in which the multiple machine learning models ML are used in response to the display of the first answer set. The "user instruction" may be an operation performed by the user using the input device 140. The input acquisition unit 111 may acquire the above-mentioned order change instruction in response to the display of an arbitrary n-th answer set where n is 2 or more, not limited to the display of the first answer set. Furthermore, the input acquisition unit 111 may acquire a user instruction (hereinafter also referred to as a dialogue response instruction) indicating a response to a dialogue request. The dialogue request is transmitted from at least one management terminal 30.

[0049] The answer acquisition unit 112 acquires an answer using each of the multiple machine learning models ML. In this exemplary embodiment, the "process of acquiring an answer using a machine learning model ML" includes (i) a process of generating a query to be input to the machine learning model ML, (ii) a process of transmitting the generated query to the server 20 that stores the machine learning model ML, and (iii) a process of receiving the answer generated by the machine learning model ML from the server 20.

[0050] For example, the answer acquisition unit 112 uses each of the multiple machine learning models ML in order to acquire a first answer corresponding to the first request from each machine learning model ML. The first answer obtained from the machine learning model ML[k] used kth is described as the first answer[k]. At this time, the answer acquisition unit 112 causes each of the second and subsequent machine learning models ML[k] to refer to at least one of the first answers[1] to [k-1] obtained from at least one of the machine learning models ML[1] to ML[k-1] used before the machine learning model ML[k]. In this exemplary embodiment, each of the second and subsequent machine learning models ML[k] is caused to refer to the first answers[1] to [k-1] obtained from the machine learning models ML[1] to ML[k-1] used before the machine learning model ML[k]. For example, the answer acquisition unit 112 generates a query to be input to the first machine learning model ML[1] based on the first request. Furthermore, the answer acquiring unit 112 generates queries to be input to the second and subsequent machine learning models ML[k] based on the first request and the first answers [1] to [k-1].

[0051] The answer acquiring unit 112 may also acquire a second answer set after displaying the first answer set. The second answer set may be acquired in response to a second request input by the user in response to the display of the first answer set. The second answer set may also be acquired based on the first answer set without the user inputting a second request.

[0052] A case where the second answer set is acquired in response to the second request will be described. For example, the answer acquisition unit 112 acquires a second answer corresponding to the second request from each machine learning model ML by using each of the multiple machine learning models ML in order. At this time, the answer acquisition unit 112 makes each of the second and subsequent machine learning models ML[k] refer to at least one of the second answers [1] to [k-1] obtained from at least one of the machine learning models ML[1] to [k-1] used before the machine learning model ML[k]. In this exemplary embodiment, each of the second and subsequent machine learning models ML[k] refers to the second answers [1] to [k-1] obtained from each of the machine learning models ML[1] to ML[k-1] used before the machine learning model ML[k].

[0053] For example, the answer acquiring unit 112 generates a query to be input to the first machine learning model ML[1] based on the first request, the first answer set, and the second request. The answer acquiring unit 112 generates queries to be input to the second and subsequent machine learning models ML[k] based on the first request, the first answer set, the second request, and the second answers [1] to [k-1]. The answer acquiring unit 112 may further acquire the nth answer set in response to the nth request.

[0054] A case will be described where the second answer set is acquired based on the first answer set without the user inputting the second request. For example, the answer acquisition unit 112 acquires a second answer based on the first answer set from each machine learning model ML by using each of the multiple machine learning models ML in order. At this time, the answer acquisition unit 112 further causes each of the second and subsequent machine learning models ML[k] to refer to at least one of the second answers [1] to [k-1] obtained from at least one of the machine learning models ML[1] to ML[k-1] used before the machine learning model ML[k]. In this exemplary embodiment, each of the second and subsequent machine learning models ML[k] is caused to refer to the second answers [1] to [k-1] obtained from the machine learning models ML[1] to ML[k-1] used before the machine learning model ML[k].

[0055] For example, the answer acquisition unit 112 generates a query to be input to the first machine learning model ML[1] based on the first request and the first answer set. The answer acquisition unit 112 also generates queries to be input to the second and subsequent machine learning models ML[k] based on the first request, the first answer set, and the second answers [1] to [k-1]. The answer acquisition unit 112 may further repeat the process of acquiring the n+1th answer set based on the nth answer set for each integer of n=2 or more. In this case, a termination condition for the repetition is determined. For example, the termination condition may be determined to terminate when the fourth answer set is obtained, but is not limited thereto.

[0056] The answer acquisition unit 112 may also acquire a new first answer based on an order change instruction for the first answer set. The changed order is described as k(new). For example, the answer acquisition unit 112 causes at least one machine learning model ML[k(new)] among the multiple machine learning models ML to refer to the first answers [1] to [k(new)-1] obtained from the machine learning models ML[1] to ML[k(new)-1] whose order is earlier than the machine learning model ML[k(new)] after the change. In this way, the answer acquisition unit 112 acquires a new first answer [k(new)] corresponding to the first request from the machine learning model ML[k(new)]. The answer acquisition unit 112 may acquire a new nth answer based on an order change instruction for the nth answer set.

[0057] The answer acquisition unit 112 may also control the machine learning model ML not to output an unfavorable answer. In this case, each of the multiple machine learning models ML[k] is controlled not to generate the first answer [k] that the machine learning model ML[k] can generate in response to the first request when the first answer [k] that the machine learning model ML[k] can generate is unfavorable compared to at least one of the first answers [l] generated by each of the other machine learning models ML[l] (k≠l) that the machine learning model ML[k] can generate. Note that the answer acquisition unit 112 may also control the machine learning model ML not to output an unfavorable answer for the n-th answer [k] that can be generated in response to the n-th request. A specific example of such a control method will be described later.

[0058] In addition, when a target designation instruction from the user is acquired, the answer acquisition unit 112 uses multiple machine learning models ML of the specified target in sequence when acquiring the above-mentioned nth answer set (n=1, 2, ...) instead of using multiple machine learning models ML in sequence.

[0059] The summary generation unit 113 generates a summary based on at least the first answer set. When not only the first answer set but also up to the mth answer set (m is an integer of 2 or more) are obtained, the summary generation unit 113 generates the summary based on the first answer set to the mth answer set.

[0060] The display control unit 114 displays a first answer set including a first answer obtained from each of the multiple machine learning models ML on the display device 150. The display control unit 114 may also display a second answer set including a second answer obtained from each of the multiple machine learning models ML on the display device 150. The display control unit 114 may also display an n-th answer set including an n-th answer obtained from each of the multiple machine learning models ML on the display device 150.

[0061] Furthermore, the display control unit 114 may display a new first answer [k(new)] acquired in response to an order change instruction corresponding to the first answer set on the display device 150. Here, the display control unit 114 may display the new first answer [k(new)] in addition to the first answer set. Furthermore, if the order before the change of the machine learning model ML that generated the new first answer [k(new)] was k(prev), the display control unit 114 may display the new first answer [k(new)] in place of the previous first answer [k(prev)].

[0062] Furthermore, the display control unit 114 may display the summary on the display device 150. Furthermore, in response to a dialogue response instruction, the display control unit 114 may display on the display device 150 a user interface for dialogue between the user and an operator of at least one management terminal 30. Such a user interface is generated by the dialogue UI unit 116, which will be described later.

[0063] The history transmission unit 115 transmits the history of the information displayed on the display device 150 to the management terminal 30 corresponding to each of the multiple machine learning models ML. The history transmission unit 115 transmits the history to each of the multiple servers 20, thereby transmitting the history to the management terminal 30 connected to each server 20. The history transmission unit 115 may also transmit the information displayed on the display device 150 to the multiple management terminals 30 at a predetermined interval or each time the information on the display device 150 is updated. The history transmission unit 115 may also transmit to the management terminal 30 the difference between the information transmitted up to the last time and the information currently displayed on the display device 150.

[0064] The dialogue UI unit 116 receives a dialogue request from at least one management terminal 30. The dialogue request is information indicating that an operator of the management terminal 30 desires to dialogue with a user. In addition, the dialogue UI unit 116 provides a user interface for direct dialogue between the user of the user terminal 10 and the operator using the management terminal 30.

[0065] (Server 20 configuration) The configuration of the server 20 will be described with reference to Fig. 5. Fig. 5 is a block diagram showing the configuration of the server 20. The server 20 is managed by a management organization that manages the machine learning model ML.

[0066] The server 20 includes a control unit 210, a storage unit 220, and a communication unit 230. The control unit 210 controls each unit of the server 20. The storage unit 220 stores various data and programs used by the control unit 210. The communication unit 230 communicates with other devices under the control of the control unit 210. Note that the storage unit 220 and a part or all of the communication unit 230 may be built into the user terminal 10, or may be connected to the outside as a peripheral device.

[0067] The storage unit 220 stores the machine learning model ML. The control unit 210 includes a query receiving unit 211, an answer generating unit 212, and an answer transmitting unit 213. The query receiving unit 211 receives a query from the user terminal 10. The answer generating unit 212 inputs a query to the machine learning model ML to obtain an answer to be output. The answer transmitting unit 213 transmits the obtained answer to the user terminal 10. The control unit 210 also relays information transmitted and received between the user terminal 10 and the management terminal 30.

[0068] (Configuration of management terminal 30) The configuration of the management terminal 30 will be described with reference to FIG. 6. FIG. 6 is a block diagram showing the configuration of the management terminal. The management terminal 30 is a terminal used by an operator belonging to an organization that manages the server 20. The management terminal 30 includes a control unit 310, a storage unit 320, a communication unit 330, an input device 340, and a display device 350. The control unit 310 controls each unit of the management terminal 30. The storage unit 320 stores various data and programs used by the control unit 310. The communication unit 330 communicates with other devices under the control of the control unit 310. The input device 340 may be configured by a mouse, a touch pad, a keyboard, or a combination of some or all of these. The display device 350 may be configured by a liquid crystal display or an organic EL (Electro Luminescence) display. The input device 340 and the display device 350 may be configured by a touch panel formed integrally. In addition, some or all of the storage unit 320, communication unit 330, input device 340, and display device 350 may be built into the management terminal 30, or may be connected to the outside as peripheral devices. Furthermore, the management terminal 30 may further include a microphone, a speaker, etc., or these may be connected to the outside.

[0069] The control unit 310 includes a history receiving unit 311 and a dialogue UI unit 312. The history receiving unit 311 receives a history of interactions between a user and each machine learning model ML from the user terminal 10, and displays the received history on the display device 350. The dialogue UI unit 312 also displays a user interface for dialogue between the user of the user terminal 10 and the operator using the management terminal 30 on the display device 350. For example, when the dialogue UI unit 312 receives an operation input by the operator to request the dialogue, it transmits the request for the dialogue to the user terminal 10. For example, the operator of the management terminal 30 can perform an operation to transmit a dialogue request when the operator judges that the management organization to which the operator belongs is less favorable than other management organizations while viewing the interactions between the user of the user terminal 10 and each machine learning model ML in real time.

[0070] (Basic function: flow of information processing method S10) The flow of the information processing method S10 executed by the information processing system 1A will be described with reference to Fig. 7. Fig. 7 is a flow diagram illustrating the flow of the information processing method S10. The information processing method S10 realizes a basic function of a reply service provided by the information processing system 1A. As shown in Fig. 7, the information processing method S10 includes steps S101 to S113.

[0071] Step S101 is an example of an input acquisition process. In step S101, the input acquisition unit 111 of the user terminal 10 acquires a first request input by the user to the input device 140. In addition, the display control unit 114 displays the acquired first request on the display device 150. An example of the first request displayed on the display device 150 in step S101 will be described with reference to FIG. 8.

[0072] FIG. 8 shows an example of a screen displayed on the display device 150 in the basic function. In FIG. 8, the screen example G1 includes the first request G11 acquired in step S101. The first request G11 is "I want a stuffed dog." In the example of FIG. 8, the number of servers 20 is three (i.e., N=3). The server 20-1 is managed by company A, and the machine learning model ML-1 stored in the server 20-1 is trained to generate answers suitable for company A. The server 20-2 is managed by company B, and the machine learning model ML-2 stored in the server 20-2 is trained to generate answers suitable for company B. The server 20-3 is managed by company C, and the machine learning model ML-3 stored in the server 20-3 is trained to generate answers suitable for company C. In this example, the order in which the machine learning models ML-1 to ML-3 are used is determined in this order at least in the basic function. That is, server 20-1 stores the first used machine learning model ML[1] and is therefore also referred to as server 20[1]. Similarly, servers 20-2 and 20-3 are also referred to as servers 20[2] and 20[3].

[0073] Steps S102 to S106 in FIG. 7 are an example of at least a part of the answer acquisition process, and are an example of a series of processes for acquiring a first answer [1] from a first machine learning model ML [1]. In step S102, the answer acquisition unit 112 of the user terminal 10 generates a query to be input to the first machine learning model ML [1] based on the first request. For example, the answer acquisition unit 112 may generate a query in a format suitable for the machine learning model ML [1] from the first request. The suitable format is a format for obtaining an appropriate answer from the machine learning model ML [1]. For example, the answer acquisition unit 112 may use the first request as a query as it is, or may convert the first request according to a predetermined conversion rule to generate a query. In addition, for example, the answer acquisition unit 112 may generate a query in the suitable format by inputting the first request into a query generation model (not shown) capable of generating a query in a format suitable for the machine learning model ML [1]. Such a query generation model may be, but is not limited to, a language model. As an example, a first request "I want a stuffed dog" may generate the query "The user is considering purchasing a stuffed dog."

[0074] In step S103, the answer acquisition unit 112 transmits the generated query to the server 20[1] that stores the machine learning model ML[1]. In step S104, the query reception unit 211 of the server 20[1] receives the query. In step S105, the answer generation unit 212 of the server 20[1] inputs the received query to the machine learning model ML[1] to obtain an answer to be output. The answer has content that is suitable for the management organization of the server 20[1] in response to the first request. The answer generation unit 212 transmits the output answer to the user terminal 10. In step S106, the answer acquisition unit 112 of the user terminal 10 acquires the answer received from the server 20[1] as the first answer[1].

[0075] Step S107 is an example of at least a part of the display control process. In step S107, the display control unit 114 displays the acquired first answer [1] on the display device 150. An example of the first answer [1] displayed on the display device 150 in step S107 will be described with reference to FIG. 8. In FIG. 8, the screen example G1 includes the first answer G121 "100 yen each." The first answer G121 is an example of the first answer [1]. The first answer G121 was generated by the machine learning model ML [1] of Company A in response to the first request G11 "I want a stuffed dog."

[0076] Next, steps S108 to S113 are executed for each of k=2, 3, ..., N. Steps S108 to S112 are an example of at least a part of an answer acquisition process, and an example of a series of processes for acquiring a first answer[k] from the second or subsequent machine learning model ML[k]. Step S113 is an example of at least a part of a display control process.

[0077] First, the case of k=2 for acquiring the first answer [2] will be described. In step S108, the answer acquisition unit 112 generates a query to be input to the second machine learning model ML [2] based on the first request and the first answer [1]. For example, the answer acquisition unit 112 may generate a query "I want a stuffed dog. Company A: 100 yen each" using information on "Company A", which is the management organization of the machine learning model ML [1], in addition to the first request and the first answer [1]. Also, for example, the answer acquisition unit 112 may generate a query in a format suitable for the machine learning model ML [2] from the first request and the first answer [1]. For example, the answer acquisition unit 112 may generate a query in the suitable format by inputting the first request and the first answer [1] to the above-mentioned query generation model. A specific example of the query generation model is as described above, and therefore detailed description will not be repeated. As an example, a query “The user is considering purchasing a stuffed dog, and company A appears to be offering them for 100 yen each” may be generated from the first request “I want a stuffed dog” and the first answer [1] “They are 100 yen each.”

[0078] In step S109, the answer acquisition unit 112 transmits the generated query to the server 20[2] that stores the machine learning model ML[2]. In step S110, the query reception unit 211 of the server 20[2] receives the query. In step S111, the answer generation unit 212 of the server 20[2] inputs the received query to the machine learning model ML[2] to obtain an answer to be output. The answer generation unit 212 transmits the output answer to the user terminal 10. In step S112, the answer acquisition unit 112 of the user terminal 10 acquires the answer received from the server 20[2] as a first answer[2].

[0079] In step S113, the display control unit 114 displays the acquired first answer [2] on the display device 150. The first answer [2] is a response to the first request and has contents conforming to the management organization of the server 20 [2], and may be different from the first answer [1] corresponding to the same first request. An example of the first answer [2] displayed on the display device 150 in step S113 will be described with reference to FIG. 8. In FIG. 8, the screen example G1 includes the first answer G122 "It's cheaper than Company A, 80 yen each". The first answer G122 is an example of the first answer [2]. The first answer G122 is a response from Company B corresponding to the first request G11 "I want a stuffed dog", and was generated by the machine learning model ML [2] of Company B by referring to Company A's first answer G121 "It's 100 yen each".

[0080] Next, the case of k=3 in which the first answer [3] is obtained will be described. In step S108, the answer acquisition unit 112 generates a query to be input to the third machine learning model ML [3] based on the first request and the first answer [1] to the first answer [2]. For example, the answer acquisition unit 112 may generate a query such as "I want a stuffed dog toy. Company A: 100 yen each. Company B: 80 yen each" using information on the management organizations of the machine learning models ML [1] to ML [2], in addition to the first request and the first answer [1] to the first answer [2]. Also, for example, the answer acquisition unit 112 may generate a query in a format suitable for the machine learning model ML [3] from the first request and the first answer [1] to the first answer [2]. For example, the answer acquisition unit 112 may generate a query in the suitable format by inputting the first request and the first answer [1] to the first answer [2] to the above-mentioned query generation model. A specific example of the query generation model is described above, and detailed description will not be repeated. As an example, a query "A user is considering purchasing a stuffed dog, and it appears that company A offers them for 100 yen each. Company B offers them even cheaper, at 80 yen each" may be generated from a first request "I want a stuffed dog," a first answer [1] "100 yen each," and a first answer [2] "80 yen each."

[0081] Steps S109 to S113 for k=3 can be explained in substantially the same manner as in the explanation for k=2, except that [2] is replaced with [3]. As a result, the first response [3] acquired in step S112 corresponds to the first request and has content that is appropriate for the management organization of server 20 [3], and may differ from the first response [1] and second response [2] corresponding to the same first request.

[0082] An example of the first answer [3] displayed on the display device 150 in step S113 will be described with reference to FIG. 8. In FIG. 8, the screen example G1 includes the first answer G123 "150 yen each, it can't beat Company A and Company B in price, but the fur is the best." The first answer G123 is an example of the first answer [3]. The first answer G123 is a response from Company C to the first request G11 "I want a stuffed dog," and was generated by the machine learning model ML [3] of Company C with reference to Company A's first answer G121 "100 yen each" and Company B's first answer G122 "80 yen each."

[0083] Thereafter, steps S108 to S113 are executed for each k while incrementing k by 1, and when this series of steps is completed for k=N, the information processing method S10 ends. The series of steps for k=4 or more are substantially similar to the cases of k=2 and 3, and therefore detailed description will not be repeated. In the example of FIG. 8, since N=3, first answer G121 to first answer G123 are displayed and the information processing method S10 ends. First answer G121 to first answer G123 configure an example of a first answer set G12.

[0084] (Intervention function: flow of information processing method S20) Next, the flow of the information processing method S20 executed by the information processing system 1A will be described with reference to Fig. 9. Fig. 9 is a flow diagram illustrating the flow of the information processing method S20. The information processing method S20 realizes an intervention function in the reply service provided by the information processing system 1A. The information processing method S20 is executed following the information processing method S10. For example, the information processing method S20 is started in a state where the screen example G1 illustrated in Fig. 8 is displayed on the display device 150. As shown in Fig. 9, the information processing method S20 includes steps S201 to S213.

[0085] Each of steps S201 to S213 can be described in substantially the same manner as in steps S101 to S113 of the information processing method S10, except that the first request is replaced with the second request and the first response is replaced with the second response.

[0086] In step S202, the answer acquisition unit 112 of the user terminal 10 may generate a query to be input to the first machine learning model ML[1] based not only on the second request but also on one or both of the first request and the first answer set obtained up to that point. Also, in step S208, the answer acquisition unit 112 of the user terminal 10 may generate a query to be input to the kth machine learning model ML[k] based not only on the second request and the second answers [1] to [k-1] but also on one or both of the first request and the first answer set obtained up to that point.

[0087] An example of the second request displayed on the display device 150 in step S201 will be described with reference to Fig. 10. Fig. 10 shows an example of a screen displayed on the display device 150 in the intervention function. Note that the first request G11 and first answer set G12 included in the screen example G2 in Fig. 10 have already been displayed by execution of the information processing method S10 (basic function) and have been described with reference to Fig. 8, and therefore detailed description will not be repeated. Here, the second request G21 and second answer set G22 displayed by execution of the subsequent information processing method S20 (intervention function) will be described.

[0088] The screen example G2 includes the second request G21 "I'm curious about the size" acquired in step S201. In addition, in the example of the screen example G2, the query generated in step S202 is generated based on the first request G11 and the first answer set G12 in addition to the second request G21. As an example, it is assumed that the query generated is "The user is considering purchasing a stuffed dog. Each company has responded with a price, but the user seems to be curious about the size of the stuffed dog."

[0089] The screen example G2 also includes the second answer G221 "It's 50 cm (centimeters) tall" displayed in step S207. The second answer G221 is an example of a second answer [1]. The second answer G221 is a response from Company A to the second request G21 "I'm concerned about the size" and is generated by Company A's machine learning model ML [1]. The screen example G2 also includes the second answer G222 "It's smaller than Company A and is 30 cm tall" displayed in step S213 (k = 2). The second answer G222 is an example of a second answer [2]. The second answer G222 is a response from Company B to the second request G21 "I'm concerned about the size" and is generated by Company B's machine learning model ML [2] with reference to Company A's second answer G221 "It's 50 cm tall." The screen example G2 also includes the second answer G223 "Smaller than Company A and Company B, 10 cm tall" displayed in step S213 (k=3). The second answer G223 is an example of the second answer [3]. The second answer G223 is Company C's answer to the second request G21 "I'm concerned about the size", and was generated by Company C's machine learning model ML [3] by referring to Company A's second answer G221 "50 cm tall" and Company B's second answer G222 "Smaller than Company A, 30 cm tall".

[0090] 10, since N=3, when the second answers G221 to G223 are displayed, the information processing method S20 ends. The second answers G221 to G223 configure an example of the second answer set G22.

[0091] As described above, in the user terminal 10, the input acquisition unit 111 acquires the second request input by the user in response to the display of the first answer set, and the answer acquisition unit 112 uses each of the multiple machine learning models ML in turn to acquire a second answer corresponding to the second request from each machine learning model ML, causing each of the second and subsequent machine learning models ML to refer to the second answer obtained from at least one machine learning model ML used before the machine learning model ML, and the display control unit 114 displays the second answer set including the second answers obtained from each of the multiple machine learning models ML on the display device 150.

[0092] Therefore, according to the user terminal 10, in addition to the effects of the information processing device 1 described above, the user can intervene and make a second request in response to answers obtained from the multiple machine learning models ML, and a second answer set can be obtained collectively from each machine learning model ML in response to the intervention. Also, like the first answer, the second answer is generated by referring to other second answers obtained up to that point from each of the other machine learning models ML, so that an effect is achieved in that a more useful second answer set can be obtained efficiently compared to the case where answers are simply obtained collectively from the multiple machine learning models ML.

[0093] (Give up function: Variations of information processing methods S10 and S20) The give-up function is realized by modifying step S108 in the information processing method S10 and step S208 in the information processing method S20 as follows.

[0094] In step S108, the answer acquisition unit 112 generates a query to be input to the k-th machine learning model ML[k] based on the first request and the first answer[1] to the first answer[k-1], as described above. Furthermore, the answer acquisition unit 112 may add information to the query that controls not to output the first answer[k] that can be generated when the first answer[k] that can be generated is less favorable than at least one of the first answers[1] to the first answer[k-1]. For example, such control information may be, but is not limited to, "If you determine that your company's answer is less favorable than the answers of other companies, you do not need to give an answer."

[0095] As a result, the machine learning model ML[k] used second or later is controlled not to output the first possible answer[k] if the first possible answer[k] is less favorable than at least any of the first answers[1] to [k-1]. The transformation in step S208 can be similarly explained by replacing the first request with the second request and the first answer with the second answer in the explanation of the transformation in step S108.

[0096] When steps S108 and S208 are modified as described above, the screen example G2 shown in FIG. 10 is modified to a screen example G2A shown in FIG. 11. FIG. 11 is a diagram showing an example of a screen to which the give-up function is applied. The screen example G2A is different from the screen example G2 in that the first answer set G12 does not include the first answer G123 of company C. This is because the query "The user is considering purchasing a stuffed dog toy. Company A seems to offer them for 100 yen each. Company B seems to offer them for an even cheaper price of 80 yen each. However, if you judge that your company's answer is less favorable than the answers of other companies, you do not need to answer" was generated in step S108. In response to the query, in step S111, the machine learning model ML[2] of company C does not output an answer because the price of the stuffed dog toy of company C is less favorable than the first answer G121 of company A and the first answer G122 of company B. As a result, as shown in screen example G2A, Company C's first answer [3], which is less favorable than Companies A and B, will not be displayed.

[0097] In addition, in the screen example G2A, the second answer set G22 includes the second answer G221 of Company A, the second answer G222 of Company B, and the second answer G223 of Company C, as in the screen example G2. This is because it is unclear whether the user wants a large stuffed animal or a small stuffed animal, and therefore the second answer G223 of Company C cannot necessarily be said to be disadvantageous compared to the second answer G221 of Company A and the second answer G222 of Company B. Therefore, in step S111, the machine learning model ML[2] of Company C determines that it is not necessarily disadvantageous that Company C's dog stuffed animal is smaller than those of Company A and Company B, and outputs the second answer G223.

[0098] As described above, in the user terminal 10, each of the multiple machine learning models ML is configured to be controlled by the answer acquisition unit 112 so as not to generate a first answer that the machine learning model ML can generate in response to a first request when the first answer that the machine learning model can generate is less favorable than at least one of the first answers generated by each of the other machine learning models ML that it references.

[0099] Therefore, in addition to the effects of the information processing device 1 described above, the user terminal 10 has the effect that the management organization of each machine learning model ML is able to maintain a situation in which unfavorable information is not presented to the user and no disadvantage is caused. Also, for the user, since the first answer set does not contain unfavorable information, visibility is improved and useful first answers can be found more easily.

[0100] (Flip function: flow of information processing method S30) Next, the flow of the information processing method S30 executed by the information processing system 1A will be described with reference to Fig. 12. Fig. 12 is a flow diagram illustrating the flow of the information processing method S30. The information processing method S30 realizes a flip function in the reply service provided by the information processing system 1A. The information processing method S30 is executed following the information processing method S10. For example, the information processing method S30 is started in a state where the screen example G1 illustrated in Fig. 8 is displayed on the display device 150. As shown in Fig. 12, the information processing method S30 includes steps S301 to S308.

[0101] In step S301, the input acquisition unit 111 acquires an order change instruction input by the user in response to the display of the first answer set. As described above, the order change instruction indicates an instruction to change the order in which a plurality of machine learning models ML are used. For example, the user may drag an area of ​​any first answer displayed on the display device 150 to an area indicating the changed order as an order change instruction. An example of an operation for inputting an order change instruction will be described with reference to FIG. 13. FIG. 13 is a diagram showing an example screen in the flip function. The example screen G1 shown in the upper half of FIG. 13 has the same configuration as the example screen G1 shown in FIG. 8. In this example, it is assumed that the user wants to obtain the first answer from Company A last. That is, the user wants an answer from Company A that has referred to the first answers from Company B and Company C. In this case, as shown in FIG. 10, the user drags the first answer G121 of Company A in the example screen G1 so as to overlap with the first answer G123 of Company C that has been obtained last at the present time. This inputs an order change instruction to change the order in which multiple machine learning models ML are used from machine learning models ML-1 (Company A), M-2 (Company B), and M-3 (Company C) to machine learning models ML-2 (Company B), M-3 (Company C), and M-1 (Company A). Note that the operation of inputting an order change instruction is not limited to the above example.

[0102] In step S302 of FIG. 12, the answer acquisition unit 112 specifies the first answer [k(new)] to be updated by applying the changed order. Here, all of the first answers [1] to [N] may be specified as the first answer [k(new)] to be updated. In this case, in other words, the changed order is applied and the information processing method S1, which is the basic function, is repeated once again. Also, a part of the first answers [1] to [N] may be specified as the first answer [k(new)] to be updated. For example, the first answer operated by the user to issue an order change instruction may be specified as the first answer [k(new)] to be updated. This is because it is highly likely that the user wants to know the new first answer to which the changed order is applied for the first answer operated by the user. Also, the user may be able to select whether the first answer [k(new)] to be updated is all or a part of the first answers [1] to [N]. Here, the explanation will be continued assuming that the first answer operated by the user is specified as the first answer [k(new)] to be updated. In the example of FIG. 13, the first answer G121 of company A operated by the user is identified as the first answer [k(new)] to be updated.

[0103] In step S303 of Fig. 12, the answer acquisition unit 112 generates a query based on the first request and the first answer [1(new)] to the first answer [k(new)-1]. "1(new)" indicates that the order after the change is number 1. The first answer [1(new)] to the first answer [k(new)-1] may be included in the previous first answer set.

[0104] For example, in the example of FIG. 13, the changed order of the machine learning model ML-1 of company A is third, and k(new)=3. The first answer [1(new)] is the first answer G122 already obtained from the machine learning model ML-2 whose changed order is the first. The first answer [k(new)-1]=the first answer [2(new)] is the first answer G123 already obtained from the machine learning model ML-3 whose changed order is the second. Then, the answer acquisition unit 112 generates a query based on the first request, the first answer G122, and the first answer G123. For example, the query "The user is considering purchasing a stuffed dog. Company B can provide one for 80 yen, and Company C can provide one for 150 yen, but they are promoting their good fur" is generated.

[0105] In step S304, the answer acquisition unit 112 transmits the generated query to the server 20[k(new)] that stores the machine learning model ML[k(new)]. In step S305, the query reception unit 211 of the server 20[k(new)] receives the query. In step S306, the answer generation unit 212 of the server 20[k(new)] inputs the received query to the machine learning model ML[k(new)] to obtain an answer to be output. The answer generation unit 212 transmits the output answer to the user terminal 10. In step S307, the answer acquisition unit 112 of the user terminal 10 acquires the answer received from the server 20[k(new)] as a new first answer[k(new)].

[0106] In step S308, the display control unit 114 displays the acquired first answer [k(new)] on the display device 150. For example, the display control unit 114 may display the new first answer [k(new)] by adding it to the first answer set already displayed. Furthermore, the display control unit 114 may replace the first answer [k(new)] in the first answer set already displayed with the new first answer [k(new)] and display it. Furthermore, in this case, the display control unit 114 may rearrange the display order of the first answers [1(new)] to [N(new)] including the updated new first answer [k(new)] in the first answer set already displayed according to the changed order.

[0107] An example of the new first answer [k(new)] displayed on the display device 150 in step S308 will be described with reference to FIG. 13. The screen example G3 shown in the lower half of FIG. 13 is different from the screen example G1 in the upper half in that it includes the first answer G321 "It's 100 yen each, but if you buy two, you get a 20% discount, so it's the same price as company B, and the fur is just as nice as company C's" instead of the first answer G121. The first answer G321 is a response from company A corresponding to the first request "I want a stuffed dog," and is an updated version of the first answer G121. The first answer G321 is generated by referring to the first answer G122 from company B and the first answer G123 from company C, which could not be referenced before the order was changed. In addition, in the screen example G3, the display order of the first answer G122 from company B, the first answer G123 from company C, and the first answer G321 from company A has been rearranged according to the changed order.

[0108] 13, if the user is interested in Company A, the user can change the order in which Company A is used to a later order (for example, last). This allows the user to obtain the more useful first answer G321 of Company A that references the first answers G122 and G123 of other companies, instead of the first answer G121 of Company A that did not reference the first answers G122 and G123 of other companies.

[0109] The user is not limited to issuing an order change instruction to move the order of the management organization of interest to a later order as in FIG. 13, but may issue an order change instruction to move the order of the management organization of no interest to an earlier order.

[0110] As described above, in the user terminal 10, the input acquisition unit 111 acquires an instruction input by the user in response to the display of the first answer set, which is an instruction to change the order in which the multiple machine learning models ML are used, the answer acquisition unit 112 acquires a new first answer corresponding to the first request from the machine learning model ML by having at least one of the multiple machine learning models ML refer to the first answer obtained from each machine learning model ML that is earlier in the changed order than the machine learning model ML in question, and the display control unit 114 displays the new first answer obtained from the machine learning model ML on the display device 150.

[0111] Therefore, in addition to the effects of the information processing device 1 described above, the user terminal 10 has the effect that the user can change the order in which the machine learning models ML are used, and for example, the user can obtain a more useful new first answer from a machine learning model ML in which the user is interested. For example, the user can change the order of the machine learning model ML in which the user is interested to a later order (for example, the very end), thereby making the machine learning model ML in which the user is interested refer to the first answers of more other machine learning models. As a result, the user can efficiently obtain a more useful first answer from the machine learning model ML in which the user is interested.

[0112] (Debate function: flow of information processing method S40) Next, the flow of the information processing method S40 executed by the information processing system 1A will be described with reference to Fig. 14. Fig. 14 is a flow diagram explaining the flow of the information processing method S40. The information processing method S40 realizes a debate function in the reply service provided by the information processing system 1A. As shown in Fig. 14, the information processing method S40 includes steps S401 to S414.

[0113] In step S401, the answer acquisition unit 112 of the user terminal 10 acquires a first request including a target designation instruction. As described above, the target designation instruction is an instruction to designate, among the multiple machine learning models ML, a plurality of machine learning models ML from which an answer is to be acquired. In addition, the display control unit 114 displays the acquired first request on the display device 150. An example of the first request displayed on the display device 150 in step S401 will be described with reference to FIG. 15.

[0114] Fig. 15 shows an example of a screen displayed on the display device 150 in the debate function. In Fig. 15, screen example G4 includes the first request G41 acquired in step S401. The first request G41 has the content "I would like to compare stuffed dog toys from company A and company B." This first request G41 includes a target designation instruction. The target designation instruction specifies the machine learning models ML of company A and company B as targets from which an answer can be acquired, among multiple machine learning models ML from which the information processing system 1A can acquire answers.

[0115] In step S401, the first request including the target designation instruction is not limited to being input in the reply service provided by the information processing system 1A as in the example of FIG. 15, but may be input in an external service linked to the information processing system 1A. For example, the information processing system 1A may cooperate with an external search service and acquire information input in a search result screen of the search service as the first request including the target designation instruction. Such a search result screen includes, for example, a search keyword input by the user (e.g., "stuffed dog doll") and a list of search results (e.g., a list of links to sales sites of companies selling stuffed dog dolls). In addition, the search service may arrange, in cooperation with the information processing system 1A, a button for calling up a debate function and a selection button (such as a check box) for selecting each search result on the search result screen. In this case, the user can input the first request including the target designation instruction by selecting multiple search results (e.g., a link of company A and a link of company B) to be compared and then pressing a button for calling up a debate function. In this manner, in step S401, the information processing system 1A may acquire the selected search results and search keywords from the linked search service as a first request including a target designation instruction.

[0116] Next, in step S402 of FIG. 14, the answer acquisition unit 112 acquires a first answer set corresponding to the first request by sequentially using the specified multiple machine learning models ML. Furthermore, the display control unit 114 displays the first answer set on the display device 150. Details of the processing of step S402 can be similarly explained by replacing "multiple machine learning models ML" with "specified multiple machine learning models ML" and replacing "N", the number of machine learning models ML, with "N1" in the explanation of steps S102 to S113 of the information processing method S10. N1 is the number of specified multiple machine learning models ML, and in the example of FIG. 15, N1=2.

[0117] An example of the first answer set displayed on the display device 150 in step S402 will be described with reference to FIG. 15. In FIG. 15, the screen example G4 includes the first answer set G42. The first answer set G42 includes the first answer G421 of company A and the first answer G422 of company B. The first answer G421 of company A, "100 yen each," is generated by inputting a query based on the first request, "I want to compare stuffed dog toys from company A and company B," into the machine learning model ML[1] of company A. The first answer G422 of company B, "It's cheaper than company A, 80 yen each," is generated by inputting a query based on the first request G41 and the first answer G421 of company A into the machine learning model ML[2] of company B.

[0118] Steps S403 to S414 in FIG. 14 are an example of a process for acquiring the (n+1)th answer set by referring to the nth answer set, and are repeated for n=1, 2, ..., E. E is the number of repetitions, and may be determined in advance or may be specified by the user. If specified by the user, the number of repetitions may be included in the first request acquired in step S401. First, a case where n=1, that is, a series of processes for acquiring the second answer set by referring to the first answer set, will be described.

[0119] First, in step S403, the answer acquisition unit 112 generates a query to be input to the first machine learning model ML[1] based on the first answer set. The answer acquisition unit 112 may generate a query based on the first request in addition to the first answer set. Examples of the query generation method include a method of directly concatenating the first request and the first answer set to generate a query, a method of converting the first request and the first answer set into a query according to a predetermined conversion rule, and a method of inputting the first request and the first answer set into a query generation model. The query generation model is as described in the explanation of the information processing method S10, and therefore detailed explanation will not be repeated. As an example, a query "Regarding dog stuffed toys, Company B can provide them for 80 yen each, cheaper than Company A's 100 yen. Please tell me the advantages of Company A's stuffed toys based on this information" may be generated based on the first request G41 and the first answer set G42 in FIG. 15.

[0120] In step S404, the answer acquisition unit 112 transmits the generated query to the server 20[1] that stores the machine learning model ML[1]. In step S405, the query reception unit 211 of the server 20[1] receives the query. In step S406, the answer generation unit 212 of the server 20[1] inputs the received query to the machine learning model ML[1] to obtain an answer to be output. The answer generation unit 212 transmits the output answer to the user terminal 10. In step S407, the answer acquisition unit 112 of the user terminal 10 acquires the answer received from the server 20[1] as a second answer[1]. In step S408, the display control unit 114 displays the acquired second answer[1] on the display device 150.

[0121] An example of the second answer [1] displayed on the display device 150 in step S408 will be described with reference to Fig. 15. In Fig. 15, the screen example G4 includes the second answer G431 "It is more expensive than Company B's, but our dog stuffed animal is 50 cm tall, and is a great deal for this size and price." The second answer G431 is an example of the second answer [1]. The second answer G431 was generated by the machine learning model ML [1] of Company A in response to the first request G41 and the first answer set G42.

[0122] Steps S409 to S414 in FIG. 14 are an example of a process for acquiring a second answer [k] from the second or subsequent machine learning model ML [k], and are repeated for k = 2, ..., N1. First, the case of acquiring a second answer [2] will be described. In step S409, the answer acquisition unit 112 generates a query to be input to the second machine learning model ML [2] based on the first answer set and the second answer [1]. The answer acquisition unit 112 may generate a query based on the first request in addition to the first answer set and the second answer [1]. The method of generating the query is almost the same as the method of generating a query for acquiring the second answer [1] (described in step S403), and therefore detailed description will not be repeated. As an example, based on the first request G41, first answer set G42, and second answer G431 in Figure 15, a query may be generated: "Regarding dog stuffed animals, Company A claims that although their prices are higher than those of Company B at 100 yen each, a stuffed animal that is 50 cm tall is a good deal for 100 yen. Taking this information into account, please tell us the advantages of Company B's stuffed animals."

[0123] In step S410, the answer acquisition unit 112 transmits the generated query to the server 20[2] that stores the machine learning model ML[2]. In step S411, the query reception unit 211 of the server 20[2] receives the query. In step S412, the answer generation unit 212 of the server 20[2] inputs the received query to the machine learning model ML[2] to obtain an answer to be output. The answer generation unit 212 transmits the output answer to the user terminal 10. In step S413, the answer acquisition unit 112 of the user terminal 10 acquires the answer received from the server 20[2] as a second answer[2]. In step S414, the display control unit 114 displays the acquired second answer[2] on the display device 150.

[0124] An example of the second answer [2] displayed on the display device 150 in step S414 will be described with reference to FIG. 15. In FIG. 15, the screen example G4 includes the second answer G432 "Our dog stuffed animal is 30 cm tall and smaller than that of Company A, but if you purchase two or more, you will receive a 20% discount." The second answer G432 is an example of the second answer [2]. The second answer G432 was generated by the machine learning model ML[2] of Company B in response to the first request G41, the first answer set G42, and Company A's second answer G431.

[0125] Thereafter, steps S409 to S414 are executed for each k while incrementing k by 1, and when this series of steps is completed for k=N1, the process of acquiring the second answer set ends. The series of steps for k=3 or more are substantially similar to the case of k=2, and therefore detailed description will not be repeated. In the example of FIG. 15, since N1=2, second answers G431 to G432 are displayed, and the process of acquiring the second answer set G43 ends.

[0126] Next, a case where n=2, that is, a series of processes for acquiring a third answer set by referring to the second answer set, will be described. The process for acquiring the second answer set can be described in a similar manner to the description for acquiring the second answer set, except that the second answer is read as the third answer. However, the information referenced when generating a query in steps S403 and S409 is different from the case for acquiring the second answer set. The following description will focus on the differences.

[0127] In step S403, the answer acquisition unit 112 generates a query to be input to the first machine learning model ML[1] based on the first answer set and the second answer set. The answer acquisition unit 112 may generate a query based on the first request in addition to the first answer set and the second answer set. The method of generating a query is as described in the explanation of step S403 in the case of n=2, and detailed explanation will not be repeated. As an example, a query "Regarding dog stuffed toys, Company B claims that it can be offered for 80 yen each, which is cheaper than Company A's 100 yen, and that although the size is smaller than Company A's 50 cm, it will be discounted by 20% if two or more are purchased. Please answer the advantages of Company A's stuffed toys based on this information" may be generated based on the first request G41, the first answer set G42, and the second answer set G43 in FIG. 15.

[0128] Next, steps S404 to S408 are executed in the same manner as in the case of n=2. As a result, the third answer [1] is displayed on the display device 150. An example of the third answer [1] displayed on the display device 150 in step S408 will be described with reference to FIG. 15. In FIG. 15, the screen example G4 includes the third answer G441 "We will give you a 30% discount if you purchase 5 or more items." The third answer G441 is an example of the third answer [1]. The third answer G441 is generated by the machine learning model ML [1] of company A in response to the first request G41, the first answer set G42, and the second answer set G43.

[0129] In step S409 when k=2, the answer acquisition unit 112 generates a query to be input to the second machine learning model ML[2] based on the first answer set, the second answer set, and the third answer[1]. The answer acquisition unit 112 may generate a query based on the first request in addition to the first answer set, the second answer set, and the third answer[1]. The query generation method is as described in the explanation of step S403 when n=2, so detailed explanation will not be repeated. As an example, a query "Regarding dog stuffed toys, Company A claims to offer a 30% discount on purchases of 5 or more, while Company B claims to offer a 20% discount on purchases of 2 or more. Please answer the merits of Company B's stuffed toys based on this information" may be generated based on the first request G41, the first answer set G42, the second answer set G43, and the third answer G441 in FIG. 15.

[0130] Next, steps S410 to S414 are executed in the same manner as in the case of n=2. As a result, the third answer [2] is displayed on the display device 150. An example of the third answer [2] displayed on the display device 150 in step S414 will be described with reference to FIG. 15. In FIG. 15, the screen example G4 includes the third answer G442 "If you purchase 5 or more items, we will give you a free cat stuffed toy." The third answer G441 is an example of the third answer [1]. The third answer G441 is generated by the machine learning model ML [1] of company A in response to the first request G41, the first answer set G42, and the second answer set G43.

[0131] Thereafter, steps S409 to S414 are executed for each k while incrementing k by 1, and when this series of steps is completed for k=N1, the process of acquiring the third answer set ends. The series of steps for k=3 or more are substantially the same as for k=2, and therefore detailed description will not be repeated. In the example of FIG. 15, since N1=2, the third answer G441 to the third answer G442 are displayed, and the process of acquiring the third answer set G44 ends.

[0132] Thereafter, n is incremented by 1 and steps S403 to S414 are executed for each n. When this series of steps is completed for n=E, information processing method S40 ends. The series of steps for n=4 or more are substantially similar to the cases of n=2 and 3, and therefore detailed description will not be repeated. In the example of Fig. 15, since E=3, third answer G441 to third answer G442 are displayed and information processing method S40 ends.

[0133] As described above, in the user terminal 10, the answer acquisition unit 112 uses each of the multiple machine learning models ML in turn to acquire a second answer based on the first answer set from each machine learning model ML, and causes each of the second and subsequent machine learning models ML to refer to the second answer obtained from at least one machine learning model ML used before the machine learning model ML in question, and the display control unit 114 displays the second answer set including the second answers obtained from each of the multiple machine learning models ML on the display device 150.

[0134] Therefore, according to the user terminal 10, in addition to the effects achieved by the information processing device 1 described above, by having the multiple machine learning models ML refer to the first answer set obtained so far to obtain a second answer set, it is possible to have each machine learning model ML debate each other's arguments, and to obtain useful information such as pointing out the opponent's weaknesses as in a debate or re-emphasizing one's own advantages.

[0135] (Thumb up function: flow of information processing method S50) Next, the flow of the information processing method S50 executed by the information processing system 1A will be described with reference to Fig. 16. Fig. 16 is a flow diagram illustrating the flow of the information processing method S50. The information processing method S50 realizes a thumbs-up function in the reply service provided by the information processing system 1A. As shown in Fig. 14, the information processing method S40 includes steps S501 to S503.

[0136] In step S501, the input acquisition unit 111 acquires an instruction to create a summary input by the user. The instruction to create a summary may be input by, for example, operating the Summary button G15 included in the screen examples G1 to G4 shown in Figs. 8, 10, 11, 13, and 15. Note that instead of executing step S501, the following steps S502 to S503 may be executed successively after execution of at least one of the information processing methods S10 to S40. In this case, a summary is added to the answer from each machine learning model and displayed without the user inputting an instruction to create a summary.

[0137] In step S502, summary generation unit 113 generates a summary based on at least the first answer set. Also, if the first through m-th answer sets (m is an integer equal to or greater than 2) have been acquired, summary generation unit 113 generates a summary based on the first through m-th answer sets. In step S503, display control unit 114 displays the generated summary on display device 150.

[0138] For example, the summary may be generated based on a keyword common to the answers from the multiple machine learning models ML in the first answer set (or from the first answer set to the m-th answer set). The keyword may be more than one. The summary may include information other than the information related to the above-mentioned keyword included in any of the multiple machine learning models ML. The summary may be formed in a form (e.g., a table format, bullet points, etc.) in which the answers from the multiple machine learning models ML can be compared. It should be noted that a known technique can be used for generating the summary. For example, the summary generating unit 113 may generate the summary by inputting the first answer set (or from the first answer set to the m-th answer set) into a summary generation model (not shown) capable of generating a summary. Such a summary generation model may be, but is not limited to, a language model.

[0139] An example of the generated summary will be described with reference to FIG. 17. FIG. 17 is a diagram showing an example of a screen including a summary. In FIG. 17, the screen example G5 includes an example of a summary displayed when the summary button G15 is operated on the screen example G2 by the intervention function shown in FIG. 10. In the screen example G5, the summary is shown in a table format that allows the first and second answers of Company A, Company B, and Company C to be compared. The columns of the table correspond to Company A, Company B, and Company C. The first and second rows of the table are generated based on the keywords "yen" (price) and "height" (size) that are common between the first and second answers from Company A, Company B, and Company C. The third row of the table is generated for the second answer from Company C, which is information other than information related to the above-mentioned keywords included in the second answer from Company C, that is, "fur coat".

[0140] Another example of the generated summary will be described with reference to FIG. 18. FIG. 18 is a diagram showing another example of a screen including a summary. In FIG. 18, a screen example G5A includes an example of a summary displayed when the summary button G15 is operated on the screen example G4 by the debate function shown in FIG. 15. In the screen example G5A, the summary is displayed in a table format that allows comparison of the first to third answers between Company A and Company B. The columns of the table correspond to Company A and Company B. Moreover, the first to third rows of the table are generated based on the keywords "yen" (price), "height" (size), and "discount" that are common between the first to third answers from Company A and Company B. Moreover, the fourth row of the table is generated with respect to "bonus," which is information other than information related to the above-mentioned keywords that is included in the third answer from Company B.

[0141] As described above, the user terminal 10 further includes the summary generating unit 113 that generates a summary based on at least the first answer set, and the display control unit 114 displays the summary on the display device 150. Therefore, the user terminal 10 has an effect that the user can more easily understand the characteristics and differences of the answers from each machine learning model ML, and as a result, can more easily make an optimal selection for himself / herself.

[0142] (Interactive function: flow of information processing method S60) Next, the flow of the information processing method S60 executed by the information processing system 1A will be described with reference to Fig. 19. Fig. 19 is a flow diagram illustrating the flow of the information processing method S60. The information processing method S60 realizes an interactive function in a reply service provided by the information processing system 1A. As shown in Fig. 19, the information processing method S60 includes steps S601 to S608.

[0143] In step S601, the history sending unit 115 of the user terminal 10 sends the history of the information displayed on the display device 150 to the management terminal 30 via the servers 20 corresponding to each of the multiple machine learning models ML. In step S602, the history receiving unit 311 of the management terminal 30 receives the history and displays it on the display device 350.

[0144] Here, the processes of steps S601 to S602 are repeatedly executed during execution of the information processing methods S10 to S40. For example, the processes of steps S601 to S602 may be executed in response to updates of information displayed on the display device 150 in the information processing methods S10 to S40. This allows an operator using the management terminal 30 to check in real time the interactions between the user of the user terminal 10 and each machine learning model ML.

[0145] In step S603, the dialogue UI unit 312 of the management terminal 30 acquires an instruction to transmit a dialogue request input by the operator. Furthermore, upon acquiring the instruction, the dialogue UI unit 312 transmits the dialogue request to the user terminal 10 via the server 20. For example, a dialogue request transmission button may be disposed on a screen on the management terminal 30 where an exchange is displayed in real time. In this case, the operator can issue an instruction to transmit a dialogue request by operating the dialogue request transmission button.

[0146] In step S604, when the dialogue UI unit 116 of the user terminal 10 receives the dialogue request, it displays information indicating the dialogue request on the display device 150. For example, the information indicating the dialogue request may include a button for instructing to respond to the dialogue request. An example of a screen displayed on the display device 150 in step S604 will be described with reference to FIG. 20. The screen example G2B shown in FIG. 20 includes a dialogue request G26 from the operator of company C in addition to the same configuration as the screen example G2 shown in FIG. 10. For example, in a state in which the screen example G2 is displayed, information equivalent to the screen example G2 is displayed on the display device 350 of each management terminal 30. Here, it is assumed that the operator of company C judges that the stuffed toys of company C are less advantageous in terms of price and size than those of company A and company B by viewing the information equivalent to the screen example G2, and instructs to send a dialogue request. The dialogue request G26 in the screen example G2B is displayed on the display device 150 of the user terminal 10 in response to an operation by the operator of company C. In addition, for example, if the operator of company C determines that the user is a royal customer, the operator can instruct the user to send the dialogue request. In addition, the dialogue request G26 includes a button for responding to the dialogue request.

[0147] In step S605, the input acquisition unit 111 of the user terminal 10 acquires an instruction to respond to the dialogue request input by the user. For example, the user inputs an instruction to respond to the dialogue request by operating the dialogue button G261 included in the dialogue request G26 of the screen example G2B. In addition, when the dialogue UI unit 116 acquires the instruction, it transmits information indicating that the dialogue will be responded to via the server 20 to the management terminal 30, which is the sender of the dialogue request.

[0148] In step S606, the dialogue UI unit 116 of the user terminal 10 generates a user interface for direct dialogue between the user and the operator. Furthermore, the display control unit 114 displays the user interface on the display device 150. The user interface may be realized in a form in which an operator of company C participates in a screen example G2B in which the user and a plurality of machine learning models ML are exchanged. Furthermore, the user interface may be displayed on a screen different from the screen example G2B. In this case, the user interface may be, for example, a text chat or a voice call.

[0149] In step S607, the dialogue UI unit 312 of the management terminal 30 receives information indicating a response to the dialogue request from the user terminal 10. In step S608, the dialogue UI unit 312 generates a user interface for direct dialogue between the user and the operator, and displays the user interface on the display device 350. Details of the user interface are as described in step S606, and therefore will not be described again in detail.

[0150] As described above, the user terminal 10 further includes a history sending unit 115 that sends the history of information displayed on the display device 150 to the management terminal 30 corresponding to at least one of the multiple machine learning models ML, and a dialogue UI unit 116 that receives a dialogue request from the management terminal 30. The input acquiring unit 111 acquires an instruction input by the user to respond to the dialogue request, and the display control unit 114 displays a user interface for dialogue between the user and the operator of the management terminal 30 on the display device 150 in response to the instruction. Therefore, according to the user terminal 10, the user can obtain not only answers from each machine learning model ML but also information from the operator, which increases the user's satisfaction. In addition, the user terminal 10 can put itself in a more advantageous position for the management organization.

[0151] (Application example) As described above, an example has been described in which the information processing system 1A provides an answer service that answers information about a product using multiple machine learning models ML that are each managed by a company that sells the product. However, the organization that manages the multiple machine learning models ML is not limited to the company that sells the product. For example, the organization that manages the multiple machine learning models ML may be a hospital. In this case, the user can obtain answers corresponding to the request from multiple hospitals in a lump sum. In addition, the answers obtained in order from each of the multiple hospitals are answers that have been obtained after referring to answers from other hospitals that are earlier in the order than the hospital in question. This makes it possible to efficiently obtain more useful information for comparing multiple hospitals than when answers from each hospital are simply obtained in a lump sum. Specifically, the user can use the information processing system A to perform actions such as (i) seeking a second opinion, (ii) making reservations at multiple hospitals in a lump sum, and (iii) completing a questionnaire in advance. In (i), the user can obtain an opinion from any hospital after referring to opinions from other hospitals. In (ii), the user can make a reservation at any hospital without overlapping with a reservation at another hospital that has been made earlier. In (iii), a user can include in the medical questionnaire he or she submits to any hospital information obtained from other hospitals to which the user has previously submitted a medical questionnaire.

[0152] (Modification of the device configuration) The device configuration of the information processing system 1A shown in Fig. 3 can be modified as shown in Fig. 21 to Fig. 23. Fig. 21 to Fig. 23 are block diagrams showing variations of the device configuration of the information processing system 1A.

[0153] In the information processing system 1B shown in FIG. 21, the general-purpose machine learning model ML-B behaves as a plurality of machine learning models ML by referring to a plurality of databases DB-i. The information processing system 1B includes a user terminal 10B, a plurality of servers 20B-i, a plurality of management terminals 30-i, and an answer generation server 40B. Each management terminal 30-i is described in the same manner as the information processing system 1A. Unlike the server 20-i, each server 20B-i includes a database DB-i instead of storing the machine learning model ML-i. The database DB-i stores information that conforms to the management organization of the server 20-i. Also, unlike the server 20-i, each server 20B-i does not include a query receiving unit 211, an answer generating unit 212, and an answer transmitting unit 213. Each of these functional blocks is stored in the answer generating server 40B. As the answer generating server 40B, a server that stores the general-purpose machine learning model ML-B can be applied. For example, the user terminal 10B is configured almost the same as the user terminal 10, but the query generated by the answer acquisition unit 112 is slightly different. For example, the answer acquisition unit 112 includes information such as "Please generate an answer by referring to the database DB-k" in addition to the above-mentioned query as a query to be input to the k-th machine learning model ML. This enables the user terminal 10B to perform processing by regarding the general-purpose machine learning model ML-B as multiple machine learning models ML. As a result, the information processing system 1B can provide a response service similar to that of the information processing system 1A.

[0154] In the information processing system 1C shown in FIG. 22, the machine learning model ML-C stored in the user terminal 10C behaves as multiple machine learning models ML by referring to multiple databases DB-i, respectively. The information processing system 1C includes a user terminal 10C, multiple servers 20B-i, and multiple management terminals 30-i. Each management terminal 30-i is described in the same manner as in the information processing system 1A. Each server 20B-i is as described with reference to FIG. 21. The user terminal 10C is configured almost the same as the user terminal 10, but differs in that the machine learning model ML-C is stored therein. The machine learning model ML-C may be a general-purpose machine learning model. In addition, the query generated by the answer acquisition unit 112 of the user terminal 10C is similar to the query generated by the answer acquisition unit 112 of the user terminal 10B, and therefore detailed description will not be repeated. Moreover, the answer acquisition unit 112 of the user terminal 10C obtains an answer by inputting a query to the machine learning model ML-C stored in the user terminal 10C, instead of transmitting and receiving a query and an answer to the server 20-i. This allows the user terminal 10C to process one machine learning model ML-C as multiple machine learning models ML. As a result, the information processing system 1C can provide an answer service similar to that of the information processing system 1A.

[0155] In the information processing system 1D shown in FIG. 23, the answer service server 50D realizes each function of the answer service and provides it to the user terminal 10D. The answer service server 50D is an example of an information processing device. Here, the above-mentioned user terminals 10, 10B, and 10C are realized by terminals in which dedicated applications that realize each function of the answer service are installed. In the information processing system D, it is not necessary for the dedicated application to be installed in the user terminal 10D. A user can use the answer service by accessing the answer service server 50D using the user terminal 10D.

[0156] The information processing system 1D includes a user terminal 10D, a plurality of servers 20-i, a plurality of management terminals 30-i, and a response service server 50D. Each server 20-i and each management terminal 30-i will be described in the same manner as in the information processing system 1A. Unlike the user terminal 10, the user terminal 10D does not include an input acquisition unit 111, a response acquisition unit 112, a summary generation unit 113, a display control unit 114, a history transmission unit 115, and a dialogue UI unit 116. These functional blocks are included in the response service server 50D. The user terminal 10D simply transmits an input from the input device 140 to the response service server 50D, and displays a screen received from the response service server 50D on the display device 150. This allows the information processing system 1D to provide a response service similar to that of the information processing system 1A.

[0157] [Software implementation example] Some or all of the functions of each device constituting the information processing device 1 and the information processing systems 1A, 1B, 1C, and 1D (hereinafter also referred to as "each of the above-mentioned devices") may be realized by hardware such as an integrated circuit (IC chip), or by software.

[0158] In the latter case, each of the above devices is realized, for example, by a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter, referred to as computer C) is shown in Fig. 24. Fig. 24 is a block diagram showing the hardware configuration of computer C that functions as each of the above devices.

[0159] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for causing the computer C to operate as each of the above-mentioned devices. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing the functions of each of the above-mentioned devices.

[0160] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0161] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may further include a communication interface for transmitting and receiving data to and from other devices. The computer C may further include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0162] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can obtain the program P via such a recording medium M. Furthermore, the program P can be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also obtain the program P via such a transmission medium.

[0163] [Appendix A] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0164] (Appendix A1) An input acquisition means for acquiring a first request from a user; an answer acquisition means for, when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, having the second machine learning model refer to the first answer obtained from the first machine learning model; and a display control means for displaying, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model. Information processing device.

[0165] (Appendix A2) the answer acquisition means, when sequentially using a plurality of machine learning models including the first machine learning model and the second machine learning model to acquire a first answer corresponding to the first request from each machine learning model, causes each of the second and subsequent machine learning models to refer to the first answers obtained from each machine learning model used before the machine learning model in question; 2. The information processing device according to claim 1 .

[0166] (Appendix A3) the input acquisition means acquires a second request of the user in response to the display of the first answer set; the answer acquisition means, when acquiring a second answer corresponding to the second request from each of the first machine learning model and the second machine learning model, causes the second machine learning model to refer to the second answer obtained from the first machine learning model; The display control means displays a second answer set on the display device, the second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. 13. The information processing device according to claim 12,

[0167] (Appendix A4) the input acquiring means, in response to the display of the first answer set, acquires an instruction from the user indicating a change in an order of using the first machine learning model and the second machine learning model; and the answer acquisition means acquires a new first answer corresponding to the first request from the first machine learning model by having the first machine learning model refer to a first answer obtained from the second machine learning model; and The display control means displays a new first answer obtained from the first machine learning model on the display device. An information processing device according to any one of appendices A1 to A3.

[0168] (Appendix A5) The second machine learning model is controlled by the answer acquisition means so as not to generate a first answer that the second machine learning model can generate in response to the first request when the first answer is less favorable than the first answer obtained from the first machine learning model. An information processing device according to any one of appendices A1 to A4.

[0169] (Appendix A6) a summary generating means for generating a summary based on at least the first answer set; The display control means displays the summary on the display device. An information processing device according to any one of appendices A1 to A5.

[0170] (Appendix A7) the answer acquisition means, when acquiring a second answer based on the first answer set from each of the first machine learning model and the second machine learning model, causes the second machine learning model to refer to the second answer obtained from the first machine learning model; The display control means displays a second answer set on the display device, the second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. An information processing device according to any one of appendices A1 to A6.

[0171] (Appendix A8) a history transmission means for transmitting a history of information displayed on the display device to a management terminal corresponding to at least one of the first machine learning model and the second machine learning model; A dialogue request receiving means for receiving a dialogue request from the management terminal; Further equipped with The input acquisition means acquires an instruction from the user indicating a response to the request; the display control means, in response to the instruction, displays on the display device a user interface for enabling a dialogue between the user and an operator of the management terminal. An information processing device according to any one of appendices A1 to A7.

[0172] [Appendix B] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0173] (Appendix B1) an input acquisition process in which at least one processor acquires a first request from a user; an answer acquisition process in which, when the at least one processor acquires a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, the first answer obtained from the first machine learning model is referred to by the second machine learning model; and a display control process in which the at least one processor displays, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model. Information processing methods.

[0174] (Appendix B2) In the answer acquisition process, when acquiring a first answer corresponding to the first request from each machine learning model by sequentially using a plurality of machine learning models including the first machine learning model and the second machine learning model, the at least one processor causes each of the second and subsequent machine learning models to refer to a first answer obtained from each machine learning model used before the machine learning model. 2. An information processing method as described in Appendix B1.

[0175] (Appendix B3) In the input acquisition process, the at least one processor acquires a second request of the user in response to displaying the first answer set; In the answer acquisition process, when acquiring a second answer corresponding to the second request from each of the first machine learning model and the second machine learning model, the at least one processor causes the second machine learning model to refer to the second answer obtained from the first machine learning model; In the display control process, the at least one processor displays, on the display device, a second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. 13. An information processing method according to claim 12, wherein the information processing method is a method for processing information according to claim 13.

[0176] (Appendix B4) In the input acquisition process, the at least one processor acquires, in response to displaying the first answer set, an instruction from the user indicating a change in an order of using the first machine learning model and the second machine learning model; and In the answer acquisition process, the at least one processor acquires a new first answer corresponding to the first request from the first machine learning model by having the first machine learning model refer to a first answer obtained from the second machine learning model; In the display control process, the at least one processor displays a new first answer obtained from the first machine learning model on the display device. 10. An information processing method according to any one of appendices B1 to B3.

[0177] (Appendix B5) The second machine learning model is controlled by the answer acquisition process so as not to generate a first answer that the second machine learning model can generate in response to the first request when the first answer is less favorable than the first answer obtained from the first machine learning model. 10. An information processing method according to any one of appendices B1 to B4.

[0178] (Appendix B6) the at least one processor further comprising a summary generation process for generating a summary based at least on the first answer set; In the display control process, the at least one processor displays the summary on the display device. 13. An information processing method according to any one of Appendices B1 to B5.

[0179] (Appendix B7) In the answer acquisition process, when acquiring a second answer based on the first answer set from each of the first machine learning model and the second machine learning model, the at least one processor causes the second machine learning model to refer to the second answer obtained from the first machine learning model; In the display control process, the at least one processor displays, on the display device, a second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. 13. An information processing method according to any one of appendices B1 to B6.

[0180] (Appendix B8) a history transmission process in which the at least one processor transmits a history of information displayed on the display device to a management terminal corresponding to at least one of the first machine learning model and the second machine learning model; an interaction request receiving process in which the at least one processor receives an interaction request from the management terminal; Further comprising: In the input acquisition process, the at least one processor acquires an instruction from the user indicating a response to the request; In the display control process, the at least one processor, in response to the instruction, displays on the display device a user interface for dialogue between the user and an operator of the management terminal. 13. An information processing method according to any one of appendices B1 to B7.

[0181] [Appendix C] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0182] (Appendix C1) A program for causing a computer to function as an information processing device, The computer, An input acquisition means for acquiring a first request from a user; an answer acquisition means for, when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, having the second machine learning model refer to the first answer obtained from the first machine learning model; and a display control means for displaying, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model. Information processing program.

[0183] (Appendix C2) the answer acquisition means, when sequentially using a plurality of machine learning models including the first machine learning model and the second machine learning model to acquire a first answer corresponding to the first request from each machine learning model, causes each of the second and subsequent machine learning models to refer to the first answers obtained from each machine learning model used before the machine learning model in question; 2. An information processing program according to claim 1.

[0184] (Appendix C3) the input acquisition means acquires a second request of the user in response to the display of the first answer set; the answer acquisition means, when acquiring a second answer corresponding to the second request from each of the first machine learning model and the second machine learning model, causes the second machine learning model to refer to the second answer obtained from the first machine learning model; The display control means displays a second answer set on the display device, the second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. 2. An information processing program according to claim C1 or C2.

[0185] (Appendix C4) the input acquiring means, in response to the display of the first answer set, acquires an instruction from the user indicating a change in an order of using the first machine learning model and the second machine learning model; and the answer acquisition means acquires a new first answer corresponding to the first request from the first machine learning model by having the first machine learning model refer to a first answer obtained from the second machine learning model; and The display control means displays a new first answer obtained from the first machine learning model on the display device. 13. An information processing program according to any one of appendices C1 to C3.

[0186] (Appendix C5) The second machine learning model is controlled by the answer acquisition means so as not to generate a first answer that the second machine learning model can generate in response to the first request when the first answer is less favorable than the first answer obtained from the first machine learning model. 13. An information processing program according to any one of appendices C1 to C4.

[0187] (Appendix C6) The computer, a summary generator that generates a summary based on at least the first answer set; The display control means displays the summary on the display device. 13. The information processing program according to any one of appendices C1 to C5.

[0188] (Appendix C7) the answer acquisition means, when acquiring a second answer based on the first answer set from each of the first machine learning model and the second machine learning model, causes the second machine learning model to refer to the second answer obtained from the first machine learning model; The display control means displays a second answer set on the display device, the second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. 13. The information processing program according to any one of appendices C1 to C6.

[0189] (Appendix C8) a history transmission means for transmitting a history of information displayed on the display device to a management terminal corresponding to at least one of the first machine learning model and the second machine learning model; The computer, A dialogue request receiving means for receiving a dialogue request from the management terminal; It also functions as The input acquisition means acquires an instruction from the user indicating a response to the request; the display control means, in response to the instruction, displays on the display device a user interface for enabling a dialogue between the user and an operator of the management terminal. 13. An information processing program according to any one of appendices C1 to C7.

[0190] [Appendix D] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0191] (Appendix D1) at least one processor, the at least one processor comprising: an input acquisition process for acquiring a first request from a user; an answer acquisition process for, when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, causing the second machine learning model to refer to the first answer obtained from the first machine learning model; and a display control process for displaying, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model. Information processing device.

[0192] The information processing device may further include a memory, and the memory may store a program for causing the at least one processor to execute each of the processes.

[0193] (Appendix D2) In the answer acquisition process, when acquiring a first answer corresponding to the first request from each machine learning model by sequentially using a plurality of machine learning models including the first machine learning model and the second machine learning model, the at least one processor causes each of the second and subsequent machine learning models to refer to a first answer obtained from each machine learning model used before the machine learning model. 2. The information processing device according to claim 1 .

[0194] (Appendix D3) In the input acquisition process, the at least one processor acquires a second request of the user in response to displaying the first answer set; In the answer acquisition process, when acquiring a second answer corresponding to the second request from each of the first machine learning model and the second machine learning model, the at least one processor causes the second machine learning model to refer to the second answer obtained from the first machine learning model; In the display control process, the at least one processor displays, on the display device, a second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. An information processing device according to appendix D1 or D2.

[0195] (Appendix D4) In the input acquisition process, the at least one processor acquires, in response to displaying the first answer set, an instruction from the user indicating a change in an order of using the first machine learning model and the second machine learning model; and In the answer acquisition process, the at least one processor acquires a new first answer corresponding to the first request from the first machine learning model by having the first machine learning model refer to a first answer obtained from the second machine learning model; In the display control process, the at least one processor displays a new first answer obtained from the first machine learning model on the display device. An information processing device according to any one of appendices D1 to D3.

[0196] (Appendix D5) The second machine learning model is controlled by the answer acquisition process so as not to generate a first answer that the second machine learning model can generate in response to the first request when the first answer is less favorable than the first answer obtained from the first machine learning model. An information processing device according to any one of appendices D1 to D4.

[0197] (Appendix D6) the at least one processor: further performing a summary generation process to generate a summary based on at least the first answer set; In the display control process, the at least one processor displays the summary on the display device. An information processing device according to any one of appendices D1 to D5.

[0198] (Appendix D7) In the answer acquisition process, when acquiring a second answer based on the first answer set from each of the first machine learning model and the second machine learning model, the at least one processor causes the second machine learning model to refer to the second answer obtained from the first machine learning model; In the display control process, the at least one processor displays, on the display device, a second answer set including second answers obtained from each of the first machine learning model and the second machine learning model. An information processing device according to any one of appendices D1 to D6.

[0199] (Appendix D8) the at least one processor: a history transmission process for transmitting a history of information displayed on the display device to a management terminal corresponding to at least one of the first machine learning model and the second machine learning model; A dialogue request receiving process for receiving a dialogue request from the management terminal; Further execute In the input acquisition process, the at least one processor acquires an instruction from the user indicating a response to the request; In the display control process, the at least one processor, in response to the instruction, displays on the display device a user interface for dialogue between the user and an operator of the management terminal. An information processing device according to any one of appendices D1 to D7.

[0200] [Appendix E] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0201] (Appendix E1) A program for causing a computer to function as an information processing device, The computer includes: an input acquisition process for acquiring a first request from a user; an answer acquisition process for, when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, causing the second machine learning model to refer to the first answer obtained from the first machine learning model; and a display control process for displaying, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model. A non-transitory recording medium on which an information processing program is recorded. [Explanation of symbols]

[0202] 1. Information processing device 1A, 1B, 1C, 1D Information Processing System 10, 10B, 10C, 10D User terminals 20, 20B Server 30 Management terminal 40B Answer Generation Server 50D Answer service server 110, 210, 310 Control unit 120, 220, 320 storage section 130, 230, 330 Communications Department 140, 340 Input device 150, 350 display device 11, 111 Input acquisition unit 12, 112 Answer acquisition part 13, 114 Display control unit 113 Summary generator 115 History transmission unit 116, 312 Dialogue UI section 211 Query Reception Unit 212 Answer generation part 213 Reply Sending Department 311 History Reception Unit C1 Processor C2 Memory

Claims

1. an input acquisition means for acquiring a first request from a user; an answer acquisition means for, when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, causing the second machine learning model to refer to the first answer obtained from the first machine learning model; and a display control means for displaying, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model. Information processing device.

2. the answer acquisition means, when sequentially using a plurality of machine learning models including the first machine learning model and the second machine learning model to acquire a first answer corresponding to the first request from each machine learning model, causes each of the second and subsequent machine learning models to refer to the first answers obtained from each machine learning model used before the machine learning model in question; The information processing device according to claim 1 .

3. the input acquisition means acquires a second request of the user in response to the display of the first answer set; the answer acquisition means, when acquiring a second answer corresponding to the second request from each of the first machine learning model and the second machine learning model, causes the second machine learning model to refer to the second answer obtained from the first machine learning model; The display control means displays a second answer set including second answers obtained from each of the first machine learning model and the second machine learning model on the display device.

3. The information processing device according to claim 1 or 2.

4. the input acquiring means acquires, in response to the display of the first answer set, an instruction from the user indicating a change in an order of using the first machine learning model and the second machine learning model; the answer acquisition means acquires a new first answer corresponding to the first request from the first machine learning model by having the first machine learning model refer to a first answer obtained from the second machine learning model; The display control means displays a new first answer obtained from the first machine learning model on the display device.

3. The information processing device according to claim 1 or 2.

5. The second machine learning model is controlled by the answer acquisition means so as not to generate a first answer that the second machine learning model can generate in response to the first request when the first answer is less favorable than the first answer obtained from the first machine learning model.

3. The information processing device according to claim 1 or 2.

6. a summary generating means for generating a summary based on at least the first answer set; The display control means displays the summary on the display device.

3. The information processing device according to claim 1 or 2.

7. When acquiring a second answer based on the first answer set from each of the first machine learning model and the second machine learning model, the answer acquisition means causes the second machine learning model to refer to the second answer obtained from the first machine learning model; The display control means displays a second answer set including second answers obtained from each of the first machine learning model and the second machine learning model on the display device.

3. The information processing device according to claim 1 or 2.

8. a history transmission means for transmitting a history of information displayed on the display device to a management terminal corresponding to at least one of the first machine learning model and the second machine learning model; A dialogue request receiving means for receiving a dialogue request from the management terminal; Further equipped with the input acquiring means acquires an instruction from the user indicating a response to the request; the display control means, in response to the instruction, displays on the display device a user interface for enabling a dialogue between the user and an operator of the management terminal.

3. The information processing device according to claim 1 or 2.

9. an input acquisition process in which at least one processor acquires a first request from a user; an answer acquisition process in which, when the at least one processor acquires a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, the first answer obtained from the first machine learning model is referred to by the second machine learning model; and a display control process in which the at least one processor displays, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model. Information processing methods.

10. A program for causing a computer to function as an information processing device, The computer, an input acquisition means for acquiring a first request from a user; an answer acquisition means for, when acquiring a first answer corresponding to the first request from each of a first machine learning model and a second machine learning model, causing the second machine learning model to refer to the first answer obtained from the first machine learning model; a display control means for displaying, on a display device, a first answer set including first answers obtained from each of the first machine learning model and the second machine learning model; Information processing program.

Images