Information processing device, response method, and response program

Abstract

In order to use a language model to present a response containing details corresponding to a target person, this information processing device comprises: a reception unit that receives input of a query; an attribute determination unit that determines an attribute of the target person who will be presented with a response to the query; and a presentation unit that presents, to the target person, the response generated through use of the language model and containing details corresponding to the attribute.

Description

Information processing device, response method, and response program

[0001] The present disclosure relates to an information processing device or the like that responds to a query.

[0002] There are known technologies for automatically responding to queries written in natural language. For example, Patent Document 1 below discloses a terminal device that uses an automatic response AI to generate a response to a user's comment and notify the user of the response. The automatic response AI used in Patent Document 1 also has setting items that determine the attributes (properties and characteristics) of the automatic response AI. The automatic response AI then creates a response specific to the attributes set in the setting items.

[0003] Japanese Patent Application Publication No. 2019-91387

[0004] In the terminal device described in Patent Document 1, attributes of the automatic response AI can be changed according to the user's attributes, thereby enabling a response that takes the user's attributes into consideration. However, the technology of Patent Document 1 is based on the premise that the attributes of the automatic response AI can be changed using setting items. Therefore, the technology of Patent Document 1 cannot be applied when using AI whose attributes cannot be changed using setting items. For example, recently, technology that generates responses to queries using language models has been attracting attention. However, when a language model is used, the technology of Patent Document 1 cannot be applied to provide a response that takes into consideration the attributes of the target person to whom the response to the query is to be presented.

[0005] The present disclosure has been made in consideration of the above-mentioned problems, and one exemplary purpose thereof is to provide a technology that enables a response with content appropriate to a target user to be presented when generating a response to a query using a language model.

[0006] An information processing device according to an exemplary aspect of the present disclosure includes a receiving means for receiving a query input, an attribute determination means for determining attributes of a subject to whom a response to the query is to be presented, and a presentation means for presenting a response to the subject, the response having content corresponding to the attributes and generated using a language model.

[0007] A response method according to an exemplary aspect of the present disclosure includes at least one processor receiving a query input, determining attributes of a subject to whom a response to the query is to be presented, and presenting a response to the subject, the response having content corresponding to the attributes, generated using a language model.

[0008] A response program according to an exemplary aspect of the present disclosure causes a computer to function as a receiving means for receiving a query input, an attribute determination means for determining the attributes of a subject to whom a response to the query is to be presented, and a presentation means for presenting a response to the subject, the response having content corresponding to the attributes and generated using a language model.

[0009] According to an exemplary aspect of the present disclosure, when a response to a query is generated using a language model, an exemplary effect is achieved in that it becomes possible to present a response whose content is appropriate for the target person.

[0010] 1. A block diagram showing a configuration of an information processing device according to the present disclosure. 2. A flow diagram showing the flow of a response method according to the present disclosure. 3. A diagram showing an overview of a response system according to the present disclosure. 4. A block diagram showing a configuration of another information processing device according to the present disclosure. 5. A flow diagram showing the flow of processing performed by the information processing device described in FIG. 4. 6. A block diagram showing the configuration of yet another information processing device according to the present disclosure. 7. A diagram showing an example of a UI (User Interface) screen displayed by the information processing device described in FIG. 6. 8. A flow diagram showing the flow of processing performed by the information processing device described in FIG. 6. 9. A block diagram showing the configuration of yet another information processing device according to the present disclosure. 10. A diagram showing an example of a UI screen displayed by the information processing device described in FIG. 9. 11. A flow diagram showing the flow of processing performed by the information processing device described in FIG. 11. 12. A block diagram showing the configuration of yet another information processing device according to the present disclosure. 13. A diagram showing an example of a UI screen displayed by the information processing device described in FIG. 12. 14. A flow diagram showing the flow of processing performed by the information processing device described in FIG. 13. 15. A block diagram showing the configuration of yet another information processing device according to the present disclosure.

[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, embodiments obtained by appropriately combining the technical means employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, embodiments obtained by appropriately omitting some of the technical means employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, the effects mentioned in 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, embodiments that do not exhibit the effects mentioned in 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 a basic form for each of the exemplary embodiments described below. Note that the scope of application of each technical means employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise. Furthermore, each technical means shown in the drawings referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise.

[0013] (Configuration of information processing device 1) 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 a reception unit 101, an attribute determination unit 102, and a presentation unit 103.

[0014] The reception unit 101 receives an input of a query. Note that a "query" refers to an instruction or command to generate data for a large-scale language model. Therefore, the term "query" in the following description can be replaced with a "generation instruction" or a "generation command." The format of the input query is not particularly limited. For example, the input query may be in text format, or in another format such as audio format, or may include data in multiple formats, such as a combination of text and images.

[0015] The attribute determination unit 102 determines the attributes of a target person to whom a response to a query received by the receiving unit 101 is to be presented. Note that the person who inputs the query and the target person to whom a response to the query is to be presented may be the same or different. Here, "attributes" refer to characteristics or properties possessed by the target person. These attributes can be set arbitrarily. For example, the target person's age, gender, personality, etc. may be the target person's attributes, or the organization, group, community, etc. to which the target person belongs may be the target person's attributes, or the target person's behavioral history, etc. may be the target person's attributes.

[0016] The presentation unit 103 presents to the subject a response generated using a large-scale language model, the response content corresponding to the attribute determined by the attribute determination unit 102. Here, the "large-scale language model" refers to a language model constructed by machine learning using a large amount of data to learn the arrangement of components (e.g., words) in a sentence expressed in natural language or the arrangement of sentences in a piece of writing. The difference between a large-scale language model and a non-large-scale language model is the amount of data used for machine learning and the number of parameters to be learned. In this exemplary embodiment and each of the exemplary embodiments described below, the "large-scale language model" can be rephrased as a "language model." For example, the large-scale language model may be a model trained by machine learning to generate a response sentence for a sentence expressed in natural language. The response sentence may be, for example, a summary or translation of a sentence or sentence input as a query. Other examples of large-scale language models that can be used include those that generate missing or subsequent parts of a sentence input as a query, and those that generate program code, tables, images, sounds, or music in response to a query. The content of the response presented to the subject depends on the large-scale language model used and the attribute determined by the attribute determination unit 102 .

[0017] The manner in which the response is presented is not particularly limited. For example, the presenting unit 103 may present the response by displaying and outputting an image showing the content of the response on a display device, by outputting a sound showing the content of the response on an audio output device, or by printing out the content of the response on a printer. Furthermore, the device that presents the response (for example, the display device, audio output device, or printer) may be included in the information processing device 1 or may be an external device to the information processing device 1.

[0018] (Effects of Information Processing Device 1) As described above, the information processing device 1 includes the receiving unit 101 that receives a query input, the attribute determination unit 102 that determines the attributes of a subject to whom a response to the query is to be presented, and the presentation unit 103 that presents to the subject a response whose content is appropriate for the attribute, generated using a large-scale language model. With this configuration, when a response to a query is generated using a large-scale language model, it is possible to present a response whose content is appropriate for the subject.

[0019] (Response Program) The functions of the information processing device 1 described above can also be realized by a program. The response program according to this exemplary embodiment causes a computer to function as a receiving means for receiving a query input, an attribute determination means for determining the attributes of a target person to whom a response to the query is to be presented, and a presentation means for presenting to the target person a response that has been generated using a large-scale language model and that corresponds to the attributes. Therefore, the response program according to this exemplary embodiment has the effect of making it possible to present a response that corresponds to the target person when a response to a query is generated using a large-scale language model.

[0020] (Flow of the response method) The flow of the response method will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the response method. Note that the execution entity of each step in this response method may be a processor provided in the information processing device 1, or a processor provided in another device, or the execution entity of each step may be a processor provided in each different device.

[0021] In S11, at least one processor receives a query input.

[0022] In S12, at least one processor determines attributes of a target person to whom a response to the query input received in S11 is to be presented.

[0023] In S13, at least one processor presents to the subject a response generated using the large-scale language model, the response content being dependent on the attributes determined in S12.

[0024] (Effects of Response Method) As described above, the response method according to this exemplary embodiment includes, by at least one processor, accepting an input of a query, determining attributes of a target person to whom a response to the query is to be presented, and presenting to the target person a response that is generated using a large-scale language model and that corresponds to the attributes. Therefore, when a response to a query is generated using a large-scale language model, it is possible to provide a response that corresponds to the target person.

[0025] 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 assigned the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technical means employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs. Furthermore, each technical means shown in each drawing referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs. These matters also apply to the third exemplary embodiment and subsequent exemplary embodiments described below.

[0026] (Configuration of response system 7) The configuration of the response system 7 will be described with reference to Fig. 3. Fig. 3 is a diagram showing an overview of the response system 7. The response system 7 is a system having a function of accepting an input of a query, generating an answer to the query using the large-scale language model 2, and presenting the generated answer to a user of the response system 7.

[0027] As shown in the figure, the response system 7 includes an information processing device 1A, a large-scale language model 2, and a terminal device 3. In the response system 7, the information processing device 1A can communicate with the terminal device 3 via any network. The large-scale language model 2 may be stored inside the information processing device 1A. In addition, in the response system 7, output data generated using the large-scale language model 2 may be generated by a device different from the information processing device 1A. In this case, the information processing device 1A may transfer a query to the device to generate output data, obtain the generated output data from the device, and use the output data to present an answer to the user.

[0028] A user of the response system 7 can input a query and view the response to the input query using a terminal device 3. FIG. 3 shows an example in which the terminal device 3 is a smartphone. The terminal device 3 is not limited to a smartphone as long as it is a device that allows input of a query and viewing of the response. Furthermore, while FIG. 3 shows only two terminal devices 3, the response system 7 can be used by three or more users using their own terminal devices.

[0029] The information processing device 1A acquires a query input to the terminal device 3 and determines the attributes of a user (hereinafter also referred to as a "target person") to whom a response to the query is to be presented. The information processing device 1A then presents to the target person a response whose content is generated using the large-scale language model 2 and whose content is appropriate for the target person's attributes. This makes it possible to present a response whose content is appropriate for the target person. The response is presented to the target person, for example, by displaying an image showing the response content on the terminal device 3.

[0030] For example, in the example of FIG. 3 , user U1, one of the users of the response system 7, inputs a query using his / her own terminal device 3 and receives a presentation of an answer to the query. The attribute of user U1 is "high school student." Specifically, a UI (User Interface) screen A1 is displayed on the display unit of user U1's terminal device 3. The UI screen A1 includes an input field A11 for accepting a query input and an answer field A12 for displaying an answer to the query. The query "Please explain about large-scale language models" has been input in the input field A11, and the answer text to the query is displayed in the answer field A12.

[0031] The answer shown in the answer column A12 is intended for high school students based on the user U1's attribute of "high school student." For example, the answer shown in the answer column A12 includes a sentence introducing a university laboratory. In this way, the response system 7 can present a response with content appropriate to the target person of the response.

[0032] Another user of the response system 7, user U2, also inputs a query using his / her own terminal device 3 and is presented with an answer to the query. User U2's attribute is "elementary school student." Specifically, a UI screen A2 including an input field A21 and an answer field A22 is displayed on the display unit of user U2's terminal device 3. The query "What is a large-scale language model?" has been input in the input field A21, and the answer text to this query is displayed in the answer field A22.

[0033] The answer shown in the answer column A22 is intended for elementary school students based on the attribute of user U2, "elementary school student." For example, the answer shown in the answer column A22 includes an explanatory text in simple language next to the term "natural language," as well as a sentence that illustrates in simple language a specific usage example of a large-scale language model. In this way, the response system 7 can present a response tailored to user U2 even if the query entered by user U2 is not significantly different in content from the query entered by user U1.

[0034] (Configuration of Information Processing Device 1A) The configuration of information processing device 1A will be described with reference to FIG. 4. FIG. 4 is a block diagram showing the configuration of information processing device 1A. As shown in the figure, information processing device 1A includes a control unit 10A that controls each unit of information processing device 1A and a storage unit 11A that stores various data used by information processing device 1A. Information processing device 1A also includes a communication unit 12A that enables information processing device 1A to communicate with other devices (e.g., terminal device 3), an input unit 13A that accepts various data input to information processing device 1A, and an output unit 14A that enables information processing device 1A to output various data. As shown in the figure, control unit 10A of information processing device 1A includes an acquisition unit 104A, an adjustment unit 105A, a training data generation unit 106A, and a learning unit 107A in addition to a reception unit 101, an attribute determination unit 102, and a presentation unit 103 that are included in information processing device 1A.

[0035] As described in the first exemplary embodiment, the attribute determination unit 102 determines the attributes of a target person to whom a response to a query is to be presented. For example, the attribute determination unit 102 may determine the attributes by having the target person input the attributes. The attributes may be input when the query is entered, or may be input in advance (e.g., during user registration in the response system 7). Furthermore, for example, if it is known that a target person with a predetermined attribute will use a predetermined terminal device 3, the attribute determination unit 102 may identify the terminal device 3 used to enter the query and determine the attributes corresponding to the terminal device 3 as the target person's attributes. For example, if a query is entered from a terminal device 3 exclusively for employees of a company, the attribute determination unit 102 may determine the target person's attributes as an employee of the company or a working member of the workforce. Furthermore, the attribute determination unit 102 may determine multiple attributes for a single target person.

[0036] In this exemplary embodiment, an example will be described in which the attribute determination unit 102 determines attributes using an attribute estimation model 111A. The attribute estimation model 111A is a model for estimating attributes of a target person to whom a response to a query is to be presented. The attribute estimation model 111A can be generated by machine learning the relationship between a query and the attributes of a target person to whom a response to the query is to be presented.

[0037] The attribute determination unit 102 inputs an input query or a feature quantity extracted from the query into the attribute estimation model 111A, thereby causing the attribute estimation model 111A to output an estimation result of the attribute of a subject who is to be presented with a response to the query. Note that the attribute determination unit 102 may directly adopt the estimation result by the attribute estimation model 111A as the determination result of the subject's attribute, or may determine the subject's attribute by presenting the estimation result to the subject and having the subject confirm whether it is appropriate.

[0038] The acquiring unit 104A acquires output data by inputting the query accepted by the accepting unit 101 into the large-scale language model 2 to generate the output data. The output data is data indicating the content of a response to the query. As described above, the large-scale language model 2 may be stored inside the information processing device 1A (for example, in the storage unit 11A). Alternatively, the acquiring unit 104A may cause a device other than the information processing device 1A to generate the output data using the large-scale language model 2, and acquire the generated output data from that device.

[0039] The adjustment unit 105A performs a process of adding information according to the attribute determined by the attribute determination unit 102 to the output data acquired by the acquisition unit 104A. The adjustment unit 105A also performs a process of changing or deleting a portion of the output data according to the attribute determined by the attribute determination unit 102. Note that the adjustment unit 105A does not necessarily need to perform all of adding information, changing a portion of the output data, and deleting a portion of the output data, but may perform at least one of these processes.

[0040] In this exemplary embodiment, the presentation unit 103 presents to the subject the output data that has been adjusted by the adjustment unit 105A as described above. As a result, a response that has content that corresponds to the attributes of the subject is presented to the subject. Details of the adjustment method used by the adjustment unit 105A will be described later in the section "Method for adjusting output data."

[0041] The training data generation unit 106A generates training data to be used for machine learning of the attribute estimation model 111A, using the query received by the reception unit 101. More specifically, the training data generation unit 106A generates training data by associating the query received by the reception unit 101 or the feature extracted from the query with attributes of a subject to whom a response to the query is to be presented, as correct answer data. For example, the training data generation unit 106A may set attributes that have been confirmed to be appropriate for the subject as correct answer data, or may acquire correct answer data by having an operator of the information processing device 1A or the like input correct answer data for each query.

[0042] The learning unit 107A generates the attribute estimation model 111A through machine learning. The learning unit 107A can also update the attribute estimation model 111A. Training data generated by the training data generation unit 106A is used to generate and update the attribute estimation model 111A. As described above, the training data is generated using the query received by the reception unit 101. Therefore, it can be said that the learning unit 107A generates or updates the attribute estimation model 111A through machine learning using the query received by the reception unit 101 as training data.

[0043] As described above, the information processing device 1A includes the adjustment unit 105A that performs at least one of a process of adding information according to attributes to output data obtained by inputting a query to the large-scale language model 2, and a process of changing or deleting part of the output data according to the attributes. The presentation unit 103 then presents the output data processed by the adjustment unit 105A to the subject.

[0044] According to the above configuration, the output data to which attribute-specific information has been added, the output data to which part of the output data has been changed or deleted depending on the attribute, or the output data to which attribute-specific information has been added and part of the output data to which attribute-specific information has been changed or deleted depending on the attribute is presented to the subject. This provides the effect of being able to present a response with content appropriate to the subject, in addition to the effect achieved by the information processing device 1.

[0045] The adjustment unit 105A may also add advertisements or notices to the output data as attribute-specific information. This provides the effect of being able to present advertisements or notices that correspond to the target user's attributes, in addition to the effects provided by the information processing device 1. For example, an advertisement for a university, such as that shown in response column A12 in FIG. 3, may be displayed to users whose attribute is high school student, effectively raising awareness of the university. The information processing device 1A may also perform processing to bill the advertiser or the entity that issued the notice for the presented advertisement. In this way, the information processing device 1A can also be used in advertising businesses, etc.

[0046] Furthermore, as described above, in the information processing device 1A, the attribute determination unit 102 may use the attribute estimation model 111A generated by machine learning the relationship between a query and the attributes of a person who is to submit a response to the query to determine the attributes of the subject person from the query received by the receiving unit 101. This provides the effect of being able to automatically determine the attributes of the subject person in addition to the effect provided by the information processing device 1.

[0047] As described above, the information processing device 1A also includes the learning unit 107A that generates or updates the attribute estimation model 111A through machine learning using, as training data, the query received by the receiving unit 101. This provides, in addition to the effects of the information processing device 1, an effect of generating the attribute estimation model 111A that matches the query received by the receiving unit 101, or of being able to match the attribute estimation model 111A to the query received by the receiving unit 101.

[0048] Furthermore, the learning unit 107A may use queries input by members of the same community as training data to generate or update the attribute estimation model 111A for determining attributes of members of the community. This provides the effect of enabling accurate attribute estimation according to the community, in addition to the effect provided by the information processing device 1.

[0049] Here, a "community" refers to any group of people having multiple members. For example, a "community" can be defined as a family, residents of the same area, people who work for or attend the same school, or people who share common hobbies or interests. For example, the learning unit 107A may generate or update the attribute estimation model 111A using queries input by each employee working for the same company as training data. This enables the learning unit 107A to generate an attribute estimation model 111A that can accurately determine the attributes of each employee of the company, or an attribute estimation model 111A that can determine attributes unique to the company.

[0050] When generating or updating the attribute estimation model 111A for each community, the training data generation unit 106A generates training data for that community using queries entered by users belonging to the same community among the queries received by the reception unit 101. When generating or updating the attribute estimation model 111A for each community, the attribute determination unit 102 first determines the community of the target user, and then determines the user's attributes using the attribute estimation model 111A according to the determination result.

[0051] The attribute determination unit 102 can determine the community to which the user belongs by applying a method similar to that for determining attributes. The community can also be one of the attributes. For example, the attribute determination unit 102 may determine the first attribute, i.e., the community, of the subject using a first attribute estimation model. Then, the attribute determination unit 102 may determine the second attribute of the subject belonging to the community using a second attribute estimation model for estimating a second attribute (e.g., a role in the community) of the subject.

[0052] (Method for adjusting output data) As described above, the adjustment unit 105A performs at least one of the following processes: adding information according to attributes to the output data obtained by inputting a query into the large-scale language model 2; and changing or deleting part of the output data according to the attributes.

[0053] If the information to be added is an advertisement, a notice, or the like, the information may be prepared in advance for each attribute. Furthermore, the adjustment unit 105A may obtain the information to be added from dictionary data that describes the meanings and explanations of various words, may obtain the information to be added by a search, or may generate the information to be added using the large-scale language model 2.

[0054] For example, in the answer displayed in the answer column A22 in Fig. 3, the term "natural language" is surrounded by an explanation in parentheses. The adjustment unit 105A can extract such an explanation from the "natural language" item in the dictionary data. The adjustment unit 105A can also search for the keyword "natural language" and extract the above explanation from the search results.

[0055] In addition, the adjustment unit 105A may acquire the information to be added from attributes indicating the subject's preferences, or from history data indicating the subject's behavioral history. For example, if the output data includes a dish name, the adjustment unit 105A may add a phrase indicating the subject's preference for that dish. As a result, if the output data includes the dish name "tuna sashimi" and the subject likes this dish, the adjustment unit 105A can add the phrase "your favorite" to the dish name "tuna sashimi" to change it to "tuna sashimi that you like."

[0056] The adjustment unit 105A can also generate an explanation by generating a query such as "Please explain natural language in simple terms" and inputting it to the large-scale language model 2. Similarly, the adjustment unit 105A can generate a query such as "Please explain a specific application example of the large-scale language model in simple terms" and input it to the large-scale language model 2. This allows the adjustment unit 105A to generate an explanation sentence with a specific example, such as the last sentence in the answer displayed in the answer column A22 in FIG. 3, and add this to the output data.

[0057] The adjustment unit 105A may also change part of the output data using the various types of information described above that can be used to add output data. For example, the adjustment unit 105A may replace the "natural language" portion of the output data with "language that people normally use." The manner of change is arbitrary. For example, the adjustment unit 105A may convert the tone of the output data to one that corresponds to the attributes of the target person. The adjustment unit 105A may also replace words included in the output data with words used by target people with specific attributes (e.g., technical terms, slang, etc.). Such conversion can be achieved by preparing conversion rules in advance for each attribute.

[0058] Furthermore, by specifying deletion targets for each attribute in advance, the adjustment unit 105A can delete the deletion targets when they are included in the output data. For example, it is possible to prevent a response containing content that should not be presented to a subject with a certain attribute from an ethical or age-related standpoint from being presented to that subject. In this case, prohibited words that should not be included in the response can be associated with the attribute and recorded in the storage unit 11A, etc. This allows the adjustment unit 105A to delete the prohibited words or sentences containing the prohibited words from the output data generated for a subject determined to have that attribute.

[0059] The adjustment unit 105A may also add information extracted from queries previously input by individuals with the same attributes as the target person (which may include the target person) to the output data. For example, the adjustment unit 105A may count the frequency of occurrence of each word included in queries previously input by individuals with the same attributes as the target person, and store words with a frequency of occurrence equal to or greater than a threshold in the storage unit 11A or the like in association with the attributes. This enables the adjustment unit 105A to, for example, add words with a frequency of occurrence equal to or greater than a threshold to the output data, or to replace words included in the output data with words that share the same meaning as the target person and have a frequency of occurrence equal to or greater than a threshold in past queries. In addition, the adjustment unit 105A may also add sentences related to words with a frequency of occurrence equal to or greater than a threshold to the output data. This enables the presentation unit 103 to present responses adjusted based on queries previously input by the target person or individuals with the same attributes as the target person.

[0060] Previously input queries and information generated based on those queries can be used to adjust the output data as described above, as well as to estimate the attributes of a subject. For example, as described above, among words included in queries previously input by a subject with a certain attribute, words with an appearance frequency equal to or greater than a threshold may be recorded in association with the attribute. This allows the attribute determination unit 102 to determine that the attribute of a subject who inputs a query including the recorded word is the attribute associated with the word. In this case, too, the presentation unit 103 can present a response generated based on a query previously input by the subject or a person with the same attribute as the subject.

[0061] In this way, the presentation unit 103 may present to the subject a response generated based on a query previously input by the subject or a person with the same attribute as the subject. This provides the effect of being able to present a response that reflects a query previously input, in addition to the effect achieved by the information processing device 1. For example, with the above configuration, it is possible to present a response that is familiar and easy to understand for subjects with a specific attribute, using words or phrases that are commonly used among subjects with that attribute.

[0062] (Processing Flow) The processing flow executed by the information processing device 1A will be described with reference to Fig. 5. Fig. 5 is a flow diagram showing the processing flow executed by the information processing device 1A. Note that, before starting the processing of S21, the presentation unit 103 may display a UI screen for accepting query input (for example, a screen including an input field A11 and an answer field A12 such as the UI screen A1 shown in Fig. 3) on a device (for example, the terminal device 3) used by the subject to input the query.

[0063] In S21, the reception unit 101 receives a query input. As described above, the reception unit 101 may receive a query input via another device such as the terminal device 3. Alternatively, the reception unit 101 may receive a query input via the input unit 13A.

[0064] In S22, the attribute determination unit 102 determines the attribute of a subject to be presented with a response to the query received in S21. For example, the attribute determination unit 102 may determine the attribute of the subject based on an estimation result obtained by inputting the query received in S21 into the attribute estimation model 111A.

[0065] In S23, the acquisition unit 104A inputs the query received in S21 into the large-scale language model 2 to generate output data, thereby acquiring the output data. This output data is data that will be the basis for a response to the query.

[0066] In S24, the adjustment unit 105A adds information according to the attribute determined in S22 to the output data acquired in S23. For example, the adjustment unit 105A may add an advertisement or notice according to the attribute.

[0067] In S25, the adjustment unit 105A applies changes to the output data acquired in S23 according to the attribute determined in S22. For example, the adjustment unit 105A may apply a conversion rule for each attribute to convert a word, sentence, or paragraph included in the output data.

[0068] In S26, the adjustment unit 105A deletes part of the output data acquired in S23 in accordance with the attribute determined in S22. For example, the adjustment unit 105A may delete prohibited words defined for each attribute or sentences containing such prohibited words from the output data. Note that the adjustment unit 105A does not necessarily need to perform all of the processes in S24 to S26, but may perform at least one of these processes.

[0069] In S27, the presentation unit 103 presents the output data after the processes in S24 to S26 to the subject, and the process in Fig. 5 is then completed. As described above, the presentation unit 103 may present the content to the user by displaying the processed output data on the terminal device 3. The presentation unit 103 may also output the processed output data via the output unit 14A.

[0070] 5, the training data generation unit 106A may generate training data by associating the query received in S21 or the feature extracted from the query with attributes of a subject to whom a response to the query is to be presented as correct answer data. In this case, it is preferable that the attributes included in the training data be attributes determined by a person (e.g., attributes entered by the subject or attributes recognized as correct by the subject). The learning unit 107A may then regenerate or update the attribute estimation model 111A using the generated training data.

[0071] [Third Exemplary Embodiment] (Configuration of Information Processing Device 1B) The configuration of information processing device 1B according to this exemplary embodiment will be described with reference to Fig. 6. Fig. 6 is a block diagram showing the configuration of information processing device 1B. As shown in the figure, information processing device 1B includes a control unit 10B that controls each unit of information processing device 1B. Control unit 10B includes an evaluation unit 105B instead of adjustment unit 105A that was included in information processing device 1A.

[0072] The evaluation unit 105B evaluates the output data acquired by the acquisition unit 104A by inputting the query accepted by the acceptance unit 101 into the large-scale language model 2, based on the attributes determined by the attribute determination unit 102. In other words, the evaluation unit 105B evaluates the compatibility between the output data and the attributes. Then, the presentation unit 103 included in the information processing device 1B presents the output data to the subject in a format according to the result of the evaluation by the evaluation unit 105B.

[0073] The evaluation method of the output data by the evaluation unit 105B may be any method that results in a higher evaluation of the output data the more it conforms to the attribute (or a lower evaluation of the output data the less it conforms to the attribute). For example, evaluation criteria may be set in advance for each attribute, such as +5 points if the output data contains a word frequently used by people with a certain attribute, or -5 points if the output data contains a word that is undesirable for people with that attribute. In this case, the evaluation unit 105B can evaluate the output data according to the evaluation criteria. The evaluation criteria are not limited to whether the output data contains or does not contain a certain word. For example, the evaluation criteria may be based on whether the overall volume of the output data (e.g., the number of characters or the number of words) or the complexity of the words contained in the output data conforms to the attribute.

[0074] The evaluation unit 105B may also evaluate the output data using, for example, a fitness estimation model that estimates the fitness of the output data with respect to an attribute of the output data. Such a fitness estimation model can be generated by machine learning using training data in which the fitness of the output data with respect to a predetermined attribute of the output data is associated as ground truth data with respect to the output data or features extracted from the output data.

[0075] As described above, the information processing device 1B includes the evaluation unit 105B that evaluates, based on attributes, the output data obtained by inputting a query into the large-scale language model 2. The presentation unit 103 then presents the output data acquired by the acquisition unit 104A to the subject in a format according to the result of the evaluation by the evaluation unit 105B. Therefore, the information processing device 1B has the effect of being able to present a response whose content is appropriate for the subject.

[0076] The "manner according to the result of the evaluation by the evaluation unit 105B" may be any manner in which the result of the evaluation by the evaluation unit 105B can be recognized. For example, the presentation unit 103 may display characters or images indicating the result of the evaluation by the evaluation unit 105B together with the output data. Furthermore, when the output data is text data, the presentation unit 103 may represent the result of the evaluation by the evaluation unit 105B by the typeface, font size, display color, background color, etc. of the output data.

[0077] (UI Screen Example) An example of a display screen displayed by information processing device 1B will be described with reference to Fig. 7. Fig. 7 is a diagram showing an example of a UI screen displayed by information processing device 1B. More specifically, Fig. 7 shows an example of a UI screen A3 displayed on terminal device 3 owned by user U3. The attribute of user U3 is age "70 or older."

[0078] 3, the UI screen A3 includes a query input field A31 and an answer field A32 in which a response to the query is displayed. The query input in the input field A31 is transmitted from the terminal device 3 to the information processing device 1B and received by the receiving unit 101. The acquisition unit 104A of the information processing device 1B then inputs the query to the large-scale language model 2 to generate output data, and acquires the generated output data. The presentation unit 103 displays the acquired output data in the answer field A32. Unlike the information processing device 1A of the exemplary embodiment 2, the information processing device 1B does not include the adjustment unit 105A, and therefore the output data is displayed as is in the answer field A32.

[0079] The information processing device 1B may also be provided with an adjustment unit 105A. This enables the presentation unit 103 to present output data to which information (e.g., advertisements or notices) according to the attributes of the subject has been added, or to present output data that has been processed so as not to include words or sentences that do not match the attributes of the subject. Also in the information processing device 1B, the presentation unit 103 may present to the subject a response generated based on a query previously input by the subject or a person with the same attributes as the subject.

[0080] Additionally, a message A33 indicating the evaluation result of the evaluation unit 105B is displayed on the UI screen A3. The message A33 indicates that the answer displayed in the answer field A32 may contain important information for the user U3. The presentation unit 103 presents such a message when the evaluation result of the evaluation unit 105B is a high evaluation. The content of the message may be predetermined according to the evaluation result. For example, if the evaluation unit 105B evaluates the output data on a three-level scale, a total of three types of messages corresponding to the evaluation levels may be predetermined, allowing the presentation unit 103 to identify and present a message according to the evaluation result.

[0081] Specifically, a query regarding the subjective symptoms of user U3, such as "I've been forgetting things a lot lately, and it's bothering me," is entered in the input field A31 of the UI screen A3. A message recommending dementia testing or a simple check is displayed in the response field A32. In this manner, the information processing device 1B can also be used for healthcare. When the information processing device 1B is used for healthcare, it is preferable that the evaluation unit 105B highly evaluate output data including healthcare-related information according to the attributes of the subject. For example, when users are classified according to attributes such as age, gender, and medical history, diseases that users in each category should be careful about for their health are identified. Then, words related to the identified diseases may be associated with the attributes and stored in the storage unit 11A or the like. This allows the evaluation unit 105B to highly evaluate output data including words related to diseases according to the attributes.

[0082] (Processing Flow) The processing flow executed by information processing device 1B will be described with reference to Fig. 8. Fig. 8 is a flowchart showing the processing flow executed by information processing device 1B. Note that the processing of S31 to S33 is similar to the processing of S21 to S23 in Fig. 5, respectively, and therefore the description of these processing will not be repeated here.

[0083] In S34, the evaluation unit 105B evaluates the output data acquired in S33 based on the attributes determined in S32. Then, in S35, the presentation unit 103 presents the output data to the subject in a format according to the evaluation result of S34, and the process in FIG. 8 is then terminated.

[0084] [Fourth Exemplary Embodiment] (Configuration of Information Processing Device 1C) The configuration of an information processing device 1C according to this exemplary embodiment will be described with reference to Fig. 9. Fig. 9 is a block diagram showing the configuration of the information processing device 1C. As shown in the figure, the information processing device 1C includes a control unit 10C that controls each unit of the information processing device 1C. The control unit 10C includes a query modification unit 105C instead of the adjustment unit 105A that was included in the information processing device 1A.

[0085] The query modification unit 105C adds attribute information indicating the attributes determined by the attribute determination unit 102 to the query received by the receiving unit 101. The acquisition unit 104A inputs the query with the added attribute information into the large-scale language model 2 to generate output data, and acquires the output data. The presentation unit 103 then presents the output data acquired by the acquisition unit 104A to the subject.

[0086] The "attribute information" may be any information indicating the attribute determined by the attribute determination unit 102. For example, the attribute information may be the name of the attribute determined by the attribute determination unit 102, or a sentence including the name of the attribute. When a sentence including the name of the attribute is used as the attribute information, the query modification unit 105C may generate the attribute information by inputting the name of the attribute into a predetermined template sentence. For example, the query modification unit 105C may input the name of the attribute "engineer" into the "X" part of the template sentence "I am X," thereby generating the sentence "I am an engineer," which may be used as the attribute information. By adding this sentence to the input query and inputting it into the large-scale language model 2, output data indicating an answer for an engineer can be output.

[0087] Note that the information processing device 1C may also be provided with an adjustment unit 105A, similar to the information processing device 1A. This enables the presentation unit 103 in the information processing device 1C to present output data to which information corresponding to the attributes of the subject (e.g., advertisements or notices) has been added, or to present output data that has been processed so as not to include words or sentences that do not match the attributes of the subject. Also in the information processing device 1C, the presentation unit 103 may present to the subject a response generated based on a query previously input by the subject or a person with the same attributes as the subject.

[0088] As described above, the information processing device 1C includes the query modification unit 105C that adds attribute information indicating the attribute determined by the attribute determination unit 102 to a query received by the receiving unit 101. The presentation unit 103 then inputs the query with the added attribute information into the large-scale language model 2, and presents the output data obtained to the subject. Therefore, the information processing device 1C has the effect of being able to present a response whose content is appropriate for the subject.

[0089] (UI Screen Example) An example of a display screen displayed by the information processing device 1C will be described with reference to FIG. 10. FIG. 10 is a diagram showing an example of a UI screen displayed by the information processing device 1C. More specifically, FIG. 10 shows an example of a UI screen A4 displayed on a terminal device 3 owned by user U4. User U4 has multiple attributes, one of which is "living in Y city" and another of which is "has been diagnosed with allergic rhinitis." In this way, the attributes determined by the attribute determination unit 102 may be attributes based on medical information such as a diagnosis history at a medical institution or the results of a health check or various tests. This makes it possible to present answers that reflect the medical information.

[0090] 3, the UI screen A4 includes a query input field A41 and an answer field A42 in which a response to the query is displayed. The input field A41 of the UI screen A4 has entered therein a query for consultation about the subjective symptoms of user U4, such as "My rhinitis is severe. Please tell me what to do."

[0091] In the information processing device 1C, the query modification unit 105C adds attribute information indicating the attributes of user U4, "living in Y City" and "having been diagnosed with allergic rhinitis," to the input query, "My rhinitis is severe. Please tell me how to deal with it." The acquisition unit 104A then inputs the query with the added attribute information into the large-scale language model 2 and acquires output data. As a result, output data reflecting the attribute information is acquired, and the presentation unit 103 displays the acquired output data in the answer field A42. Specifically, the answer field A42 displays a message indicating that the pollen level in the subject's residential area is high and a message recommending that the subject avoid going outdoors. These messages correspond to the attributes of user U4, "living in Y City" and "having been diagnosed with allergic rhinitis."

[0092] (Processing Flow) The processing flow executed by information processing device 1C will be described with reference to Fig. 11. Fig. 11 is a flowchart showing the processing flow executed by information processing device 1C. Note that the processing of S41 and S42 is similar to the processing of S21 and S22 in Fig. 5, respectively, and therefore description of these processing will not be repeated here.

[0093] In S43, the query modification unit 105C adds attribute information indicating the attribute determined in S42 to the query input in S41. Next, in S44, the acquisition unit 104A inputs the query with the added attribute information to the large-scale language model 2 and acquires output data. Then, in S45, the presentation unit 103 presents the output data acquired in S44, i.e., the output data obtained by inputting the query with the added attribute information to the large-scale language model 2, to the subject, and the processing in FIG. 11 is then terminated.

[0094] Fifth Exemplary Embodiment (Configuration of Information Processing Device 1D) The configuration of an information processing device 1D according to this exemplary embodiment will be described with reference to Fig. 12. Fig. 12 is a block diagram showing the configuration of the information processing device 1D. As shown in the figure, the information processing device 1D includes a control unit 10D that controls each unit of the information processing device 1D. The control unit 10D includes an output data selection unit 105D instead of the adjustment unit 105A that was included in the information processing device 1A.

[0095] The acquisition unit 104A included in the information processing device 1D acquires multiple pieces of output data by repeatedly inputting the query accepted by the acceptance unit 101 into the large-scale language model 2 and generating output data. Because the output of the large-scale language model 2 changes probabilistically, the output data acquired by such processing may be different from one another.

[0096] The output data selection unit 105D selects output data to be presented to the subject from the plurality of output data acquired by the acquisition unit 104A. More specifically, the output data selection unit 105D selects one or more output data from the plurality of output data that match the attribute determined by the attribute determination unit 102. Then, the presentation unit 103 presents the output data selected by the output data selection unit 105D to the subject.

[0097] The method for selecting output data based on attributes is arbitrary. For example, the output data selection unit 105D may perform a process for evaluating the compatibility between attributes and output data for each output data, similar to the evaluation unit 105B of exemplary embodiment 3. In this case, the output data selection unit 105D selects output data to be presented to the subject based on each evaluation result. For example, the output data selection unit 105D may select a predetermined number of output data with the highest compatibility, or may select output data with a compatibility equal to or greater than a predetermined threshold. In the latter case, it is assumed that no output data with a compatibility equal to or greater than the predetermined threshold does not exist. In such a case, the acquisition unit 104A may repeat the process of inputting a query into the large-scale language model 2 and generating output data until output data with a compatibility equal to or greater than the predetermined threshold is generated.

[0098] The information processing device 1D may also be provided with an adjustment unit 105A, similar to the information processing device 1A. This enables the presentation unit 103 in the information processing device 1D to present output data to which information corresponding to the attributes of the subject (e.g., advertisements or notices) has been added, or to present output data that has been processed so as not to include words or sentences that do not match the attributes of the subject. Also in the information processing device 1D, the presentation unit 103 may present to the subject a response generated based on a query previously input by the subject or a person with the same attributes as the subject.

[0099] As described above, the information processing device 1D includes the acquisition unit 104A that acquires multiple pieces of output data by repeatedly inputting the query accepted by the acceptance unit 101 into the large-scale language model 2 and generating output data. The presentation unit 103 then presents to the subject one or more pieces of output data selected from the acquired multiple pieces of output data according to the attribute determined by the attribute determination unit 102. Therefore, the information processing device 1D has the effect of being able to present a response whose content is appropriate for the subject.

[0100] (UI Screen Example) An example of a display screen displayed by information processing device 1D will be described with reference to Fig. 13. Fig. 13 is a diagram showing an example of a UI screen displayed by information processing device 1D. More specifically, Fig. 13 shows an example of a UI screen A5 displayed on terminal device 3 owned by user U5. User U5's attribute is "I like Japanese food."

[0101] 3, the UI screen A5 includes a query input field A51 and an answer field A52 in which a response to the query is displayed. The UI screen A5 also includes an object A53 for displaying other answers. The query "Can you recommend some restaurants?" has been entered in the input field A51 of the UI screen A5.

[0102] In the information processing device 1D, the acquisition unit 104A inputs an input query into the large-scale language model 2 and repeats the process of generating output data multiple times to acquire multiple pieces of output data. In the example of FIG. 13 , the acquisition unit 104A repeats the above process twice to generate two pieces of output data. In FIG. 13 , these pieces of output data are shown as "Answer 1" and "Answer 2." Answer 1 recommends an Italian restaurant, and Answer 2 recommends a Japanese restaurant. From these answers, i.e., output data, the output data selection unit 105D selects the output data for Answer 2 that matches the attribute of user U5, "I like Japanese food." As a result, the output data for Answer 2 is displayed in the answer column A52.

[0103] Furthermore, when the subject selects object A53, the presentation unit 103 presents to the subject the output data that was not selected by the output data selection unit 105D. In this way, the presentation unit 103 may also present the output data that was not selected by the output data selection unit 105D in response to the subject's operation, etc.

[0104] (Processing Flow) The processing flow executed by information processing device 1D will be described with reference to Fig. 14. Fig. 14 is a flowchart showing the processing flow executed by information processing device 1D. Note that the processing of S51 and S52 is similar to the processing of S21 and S22 in Fig. 5, respectively, and therefore description of these processing will not be repeated here.

[0105] In S53, the acquisition unit 104A repeats the process of inputting the query input in S51 into the large-scale language model 2 and generating output data multiple times to acquire multiple pieces of output data. Subsequently, in S54, the output data selection unit 105D selects one or more pieces of output data that match the attribute determined in S52 from the multiple pieces of output data acquired in S53. Then, in S55, the presentation unit 103 presents to the subject the output data selected in S54, i.e., one or more pieces of output data selected from the multiple pieces of output data acquired in S53 according to the attribute determined by the attribute determination unit 102, and the processing of FIG. 14 is then terminated.

[0106] Sixth Exemplary Embodiment (Configuration of Information Processing Device 1E) The configuration of an information processing device 1E according to this exemplary embodiment will be described with reference to FIG. 15. FIG. 15 is a block diagram showing the configuration of the information processing device 1E. As shown in the figure, the information processing device 1E includes a control unit 10E that controls each unit of the information processing device 1E in an integrated manner, and a storage unit 11E that stores various data used by the information processing device 1E. The control unit 10E includes a model selection unit 105E instead of the adjustment unit 105A included in the information processing device 1A. Furthermore, a plurality of large-scale language models 2E are stored in the storage unit 11E. Note that, like the large-scale language models 2, the large-scale language models 2E may also be stored in a device external to the information processing device 1E.

[0107] Like the large-scale language model 2, the large-scale language model 2E is a language model constructed by machine learning using a large amount of data on the arrangement of components (such as words) in a sentence and the arrangement of sentences in a text. The multiple large-scale language models 2E stored in the storage unit 11E are models optimized for each of multiple attributes. Note that "optimized" here means adjusted to improve compatibility with the attribute. Therefore, the adjusted model does not necessarily have to be optimal for that attribute.

[0108] The large-scale language model 2E can also be generated by updating the large-scale language model 2 through machine learning using training data that associates queries entered by people with predetermined attributes with appropriate output data for those queries. Note that such updating can also be referred to as fine-tuning.

[0109] In this way, by using a large-scale language model 2E optimized for each of multiple attributes, it is possible to vary the information disclosure level for each attribute. For example, when the learning unit 107A generates a large-scale language model 2E for a certain attribute, it is possible to prevent a response containing information that should not be presented to a person with that attribute from being generated by using training data that does not include that information. Furthermore, the acquisition unit 104A may limit the data referenced when generating a response using the large-scale language model 2E depending on the attribute of the target person. This makes it possible to generate a response with a disclosure level that corresponds to the attribute.

[0110] The model selection unit 105E selects one of the multiple large-scale language models 2E described above according to the attribute determined by the attribute determination unit 102. For example, if the attribute determination unit 102 determines that the attribute of the subject is "employee of a specific company," the model selection unit 105E selects one of the multiple large-scale language models 2E that is optimized for employees of that company. The acquisition unit 104A then inputs the query received by the reception unit 101 into the large-scale language model 2E selected by the model selection unit 105E to generate output data, and the presentation unit 103 presents the output data to the subject.

[0111] Note that the information processing device 1E may also be provided with an adjustment unit 105A, similar to the information processing device 1A. This enables the presentation unit 103 in the information processing device 1E to present output data to which information corresponding to the attributes of the subject (e.g., advertisements or notices) has been added, or to present output data that has been processed so as not to include words or sentences that do not match the attributes of the subject. Also in the information processing device 1E, the presentation unit 103 may present to the subject a response generated based on a query previously input by the subject or a person with the same attributes as the subject.

[0112] As described above, the presentation unit 103 of the information processing device 1E presents to the subject a response generated using the large-scale language model 2E, among the large-scale language models 2E optimized for each of the multiple attributes, that corresponds to the attribute determined by the attribute determination unit 102. Therefore, the information processing device 1E has the effect of being able to present a response whose content is appropriate for the subject.

[0113] (UI Screen Example) An example of a display screen displayed by the information processing device 1E will be described with reference to Fig. 16. Fig. 16 is a diagram showing an example of a UI screen displayed by the information processing device 1E. More specifically, Fig. 16 shows an example of a UI screen A6 displayed on a terminal device 3 owned by a user U6. The attribute of user U6 is "healthcare worker."

[0114] The UI screen A6 includes a query input field A61 and a response field A62 in which a response to the query is displayed, similar to the UI screens A1 and A2 shown in Fig. 3. The query "Tell me about first aid for heat stroke" is entered in the input field A61 of the UI screen A6.

[0115] In the information processing device 1E, the model selection unit 105E selects from multiple large-scale language models 2E one that corresponds to the attribute of user U6 determined by the attribute determination unit 102. In the example of FIG. 16 , since the attribute of user U6 is "medical worker," the model selection unit 105E selects a large-scale language model 2E optimized for medical workers. Then, the acquisition unit 104A inputs a query into the large-scale language model 2E for medical workers selected by the model selection unit 105E to generate output data, and the presentation unit 103 displays this output data in the answer column A62. As a result, output data indicating first aid for heatstroke for "medical workers" is displayed in the answer column A62.

[0116] (Processing Flow) The processing flow executed by information processing device 1E will be described with reference to Fig. 17. Fig. 17 is a flowchart showing the processing flow executed by information processing device 1E. Note that the processing of S61 and S62 is similar to the processing of S21 and S22 in Fig. 5, respectively, and therefore description of these processing will not be repeated here.

[0117] In S63, the model selection unit 105E selects one of the multiple large-scale language models 2E stored in the storage unit 11E that corresponds to the attribute determined in S62. Subsequently, in S64, the acquisition unit 104A inputs the query input in S61 to the large-scale language model 2E selected in S63 to generate output data, and acquires the output data. Then, in S65, the presentation unit 103 presents to the subject the output data acquired in S64, i.e., a response generated using the large-scale language model 2E that corresponds to the attribute determined in S62, out of the large-scale language models 2E optimized for each of the multiple attributes, and the processing of FIG. 17 is then terminated.

[0118] [Modifications] The execution entity of each process described in each of the exemplary embodiments above is arbitrary and is not limited to the above examples. In other words, the functions of the information processing devices 1, 1A, 1B, 1C, 1D, and 1E can be realized by multiple devices (which can also be called processors) that can communicate with each other. For example, each process described in the flow charts of Figures 2, 5, 8, 11, 14, and 17 can be shared and executed by multiple processors. In other words, the execution entity of the response method in each of the above embodiments may be one processor or multiple processors.

[0119] [Software Implementation Example] Some or all of the functions of the information processing devices 1, 1A, 1B, 1C, 1D, and 1E may be implemented by hardware such as an integrated circuit (IC chip), or by software.

[0120] In the latter case, the information processing devices 1, 1A, 1B, 1C, 1D, and 1E are realized by, for example, 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. 18. Fig. 18 is a block diagram showing the hardware configuration of computer C that functions as information processing device 1, 1A, 1B, 1C, 1D, or 1E.

[0121] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P (response program) for causing the computer C to operate as the information processing device 1, 1A, 1B, 1C, 1D, or 1E. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing each function of the information processing device 1, 1A, 1B, 1C, 1D, or 1E.

[0122] The processor C1 may be, for example, a central processing unit (CPU), a graphics 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.

[0123] 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 also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0124] The program P can also 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 acquire the program P via such a recording medium M. The program P can also 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 acquire the program P via such a transmission medium.

[0125] [Appendix 1] 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.

[0126] (Appendix A1) An information processing device comprising: a receiving means for receiving a query input; an attribute determination means for determining attributes of a subject to whom a response to the query is to be presented; and a presentation means for presenting to the subject a response whose content is generated using a language model and corresponds to the attributes.

[0127] (Appendix A2) The information processing device according to Appendix A1, further comprising an adjustment means that performs at least one of a process of adding information according to the attribute to output data obtained by inputting the query into the language model, and a process of changing or deleting a part of the output data according to the attribute, and the presentation means presents the output data after processing by the adjustment means to the subject.

[0128] (Supplementary Note A3) The information processing device according to Supplementary Note A2, wherein the adjustment means adds an advertisement or a notice to the output data as information according to the attribute.

[0129] (Appendix A4) The information processing device according to Appendix A1, further comprising: an evaluation means for evaluating output data obtained by inputting the query into the language model based on the attribute; and the presentation means for presenting the output data to the subject in a manner according to a result of the evaluation.

[0130] (Appendix A5) The information processing device according to Appendix A1, further comprising a query modification means for adding attribute information indicating the attribute to the query, wherein the presentation means presents to the subject output data obtained by inputting the query with the added attribute information into the language model.

[0131] (Appendix A6) The information processing device according to Appendix A1, further comprising an acquisition means for acquiring a plurality of output data by repeating a process of inputting the query into the language model and generating output data a plurality of times, and the presentation means for presenting one or a plurality of output data selected according to the attribute from the acquired plurality of output data to the subject.

[0132] (Appendix A7) The information processing device according to Appendix A1, wherein the presentation means presents to the subject a response generated using a language model corresponding to the attribute determined by the attribute determination means, out of the language models optimized for each of the plurality of attributes.

[0133] (Appendix A8) The information processing device according to any one of Appendices A1 to A7, wherein the attribute determination means determines the attributes of the target person from the query received by the reception means using an attribute estimation model generated by machine learning the relationship between a query and the attributes of the target person to whom a response to the query is to be presented.

[0134] (Supplementary Note A9) The information processing device according to Supplementary Note A8, further comprising: a learning unit that generates or updates the attribute estimation model by machine learning using the query accepted by the accepting unit as training data.

[0135] (Supplementary Note A10) The information processing device according to Supplementary Note A9, wherein the learning means uses the queries input by persons belonging to the same community as training data to generate or update the attribute estimation model for determining attributes of persons belonging to the community.

[0136] (Appendix A11) The information processing device according to any one of Appendices A1 to A10, wherein the presentation means presents to the subject the response generated based on a query previously input by the subject or a person with the same attributes as the subject.

[0137] (Appendix B1) A response method including: at least one processor accepting input of a query; determining attributes of a subject to whom a response to the query is to be presented; and presenting a response to the subject, the response having content corresponding to the attributes, generated using a language model.

[0138] (Appendix B2) The information processing method according to Appendix B1, wherein the at least one processor performs at least one of a process of adding information according to the attribute to output data obtained by inputting the query into the language model, and a process of modifying or deleting a part of the output data according to the attribute, and in presenting the response, the at least one processor presents the output data after the processing to the subject.

[0139] (Supplementary Note B3) The information processing method according to Supplementary Note B2, wherein in the process of adding information according to attributes, the at least one processor adds an advertisement or a notice to the output data.

[0140] (Appendix B4) The information processing method described in Appendix B1, further comprising: the at least one processor evaluating output data obtained by inputting the query into the language model based on the attributes; and in presenting the response, the at least one processor presents the output data to the subject in a manner according to the result of the evaluation.

[0141] (Appendix B5) The information processing method described in Appendix B1, further comprising: adding attribute information indicating the attribute to the query; and, in presenting the response, presenting to the subject output data obtained by inputting the query with the added attribute information into the language model.

[0142] (Appendix B6) The information processing method described in Appendix B1, further comprising: the at least one processor acquiring a plurality of output data by repeating a process of inputting the query into the language model and generating output data a plurality of times; and in presenting the response, the at least one processor presents to the subject one or more output data selected from the acquired plurality of output data according to the attribute.

[0143] (Appendix B7) The information processing method described in Appendix B1, wherein in presenting the response, the at least one processor presents to the subject a response generated using a language model corresponding to the determined attribute from among the language models optimized for each of the multiple attributes.

[0144] (Appendix B8) An information processing method described in any of Appendices B1 to B7, wherein in determining the attributes, the at least one processor determines the attributes of the target person from the query that has been input using an attribute estimation model generated by machine learning the relationship between the query and the attributes of the target person who will present a response to the query.

[0145] (Supplementary Note B9) The information processing method according to Supplementary Note B8, including the at least one processor generating or updating the attribute estimation model by machine learning using the received input query as training data.

[0146] (Appendix B10) The information processing method according to Appendix B9, wherein in the machine learning, the at least one processor uses the queries input by persons belonging to the same community as training data to generate or update the attribute estimation model for determining attributes of persons belonging to the community.

[0147] (Appendix B11) An information processing method described in any of Appendices B1 to B10, wherein in presenting the response, the at least one processor presents to the subject the response generated based on a query previously entered by the subject or a person with the same attributes as the subject.

[0148] (Appendix C1) A response program that causes a computer to function as a receiving means for receiving a query input, an attribute determination means for determining the attributes of a subject to whom a response to the query is to be presented, and a presentation means for presenting to the subject a response whose content is generated using a language model and corresponds to the attributes.

[0149] (Appendix C2) An information processing program according to Appendix C1, which causes the computer to function as an adjustment means that performs at least one of a process of adding information according to the attribute to output data obtained by inputting the query into the language model, and a process of changing or deleting a part of the output data according to the attribute, and wherein the presentation means presents the output data after processing by the adjustment means to the subject.

[0150] (Supplementary Note C3) The information processing program according to Supplementary Note C2, wherein the adjusting means adds an advertisement or a notice to the output data as information according to the attribute.

[0151] (Appendix C4) An information processing program described in Appendix C1, which causes the computer to function as an evaluation means that evaluates output data obtained by inputting the query into the language model based on the attributes, and the presentation means presents the output data to the subject in a manner corresponding to the result of the evaluation.

[0152] (Appendix C5) An information processing program according to Appendix C1, which causes the computer to function as a query modification means that adds attribute information indicating the attribute to the query, and the presentation means presents to the subject output data obtained by inputting the query with the added attribute information into the language model.

[0153] (Appendix C6) An information processing program as described in Appendix C1, which causes the computer to function as an acquisition means that acquires multiple output data by repeating the process of inputting the query into the language model and generating output data multiple times, and the presentation means presents one or more output data selected from the multiple acquired output data according to the attribute to the subject.

[0154] (Appendix C7) The information processing program described in Appendix C1, wherein the presentation means presents to the subject a response generated using a language model corresponding to the attribute determined by the attribute determination means, out of the language models optimized for each of the multiple attributes.

[0155] (Appendix C8) An information processing program described in any of Appendices C1 to C7, wherein the attribute determination means determines the attributes of the target person from the query accepted by the acceptance means using an attribute estimation model generated by machine learning the relationship between a query and the attributes of the target person to whom a response to the query is to be presented.

[0156] (Supplementary Note C9) The information processing program according to Supplementary Note C8, which causes the computer to function as a learning unit that generates or updates the attribute estimation model by machine learning using the query accepted by the accepting unit as training data.

[0157] (Appendix C10) The information processing program according to Appendix C9, wherein the learning means uses the queries input by persons belonging to the same community as training data to generate or update the attribute estimation model for determining attributes of persons belonging to the community.

[0158] (Appendix C11) The information processing program according to any one of Appendices C1 to C10, wherein the presentation means presents to the subject the response generated based on a query previously entered by the subject or a person with the same attributes as the subject.

[0159] [Appendix 2] 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.

[0160] (Appendix D1) An information processing device comprising at least one processor, the at least one processor executing a reception process for receiving a query input, an attribute determination process for determining attributes of a subject to whom a response to the query is to be presented, and a presentation process for presenting a response to the subject, the response having content corresponding to the attributes, generated using a language model.

[0161] (Appendix D2) The information processing device described in Appendix D1, wherein the at least one processor performs an adjustment process that performs at least one of a process of adding information according to the attribute to output data obtained by inputting the query into the language model, and a process of changing or deleting a part of the output data according to the attribute, and in the presentation process, the at least one processor presents the output data after processing by the adjustment process to the subject.

[0162] (Supplementary Note D3) The information processing device according to Supplementary Note D2, wherein in the adjustment process, the at least one processor adds an advertisement or a notice to the output data as information according to the attribute.

[0163] (Appendix D4) The information processing device described in Appendix D1, wherein the at least one processor performs an evaluation process to evaluate output data obtained by inputting the query into the language model based on the attributes, and in the presentation process, the at least one processor presents the output data to the subject in a manner corresponding to the result of the evaluation.

[0164] (Appendix D5) The information processing device described in Appendix D1, wherein the at least one processor executes a query modification process that adds attribute information indicating the attribute to the query, and in the presentation process, the at least one processor presents to the subject output data obtained by inputting the query with the added attribute information into the language model.

[0165] (Appendix D6) The information processing device described in Appendix D1, wherein the at least one processor executes an acquisition process to acquire multiple output data by repeating a process of inputting the query into the language model and generating output data multiple times, and in the presentation process, the at least one processor presents to the subject one or multiple output data selected according to the attribute from the acquired multiple output data.

[0166] (Appendix D7) The information processing device described in Appendix D1, wherein in the presentation process, the at least one processor presents to the subject a response generated using a language model corresponding to the attribute determined by the attribute determination process, from among the language models optimized for each of the multiple attributes.

[0167] (Appendix D8) An information processing device described in any of Appendices D1 to D7, wherein in the attribute determination process, the at least one processor determines the attributes of the target person from the query accepted by the reception process using an attribute estimation model generated by machine learning the relationship between the query and the attributes of the target person to whom a response to the query is to be presented.

[0168] (Supplementary Note D9) The information processing device according to Supplementary Note D8, wherein the at least one processor executes a learning process that generates or updates the attribute estimation model using the query accepted by the acceptance process as training data.

[0169] (Appendix D10) The information processing device described in Appendix D9, wherein in the learning process, the at least one processor uses the queries input by persons belonging to the same community as training data to generate or update the attribute estimation model for determining attributes of persons belonging to the community.

[0170] (Appendix D11) An information processing device described in any of Appendices D1 to D10, wherein in the presentation process, the at least one processor presents to the subject the response generated based on a query previously entered by the subject or a person with the same attributes as the subject.

[0171] (Appendix E1) A non-transient recording medium having recorded thereon a program that causes a computer to function as an information processing device, the non-transient recording medium having recorded thereon an information processing program that causes the computer to execute a reception process for receiving a query input, an attribute determination process for determining the attributes of a subject to whom a response to the query is to be presented, and a presentation process for presenting to the subject a response whose content is generated using a language model and corresponds to the attributes.

[0172] REFERENCE SIGNS LIST 1, 1A, 1B, 1C, 1D, 1E INFORMATION PROCESSING DEVICE 101 RECEIVING UNIT (RECEIVING MEANS) 102 ATTRIBUTE DETERMINATION UNIT (ATTRIBUTE DETERMINATION MEANS) 103 PRESENTING UNIT (PRESENTING MEANS) 104A ACQUIRING UNIT (ACQUIRING MEANS) 105A ADJUSTING UNIT (ADJUSTING MEANS) 105B EVALUATING UNIT (EVALUATING MEANS) 105C QUERY CHANGE UNIT (QUERY CHANGE UNIT) 107A LEARNING UNIT (LEARNING MEANS) 111A ATTRIBUTE ESTIMATION MODEL 2, 2E LARGE-SCALE LANGUAGE MODEL

Claims

1. A means of receiving query input, An attribute determination means for determining the attributes of the target person for whom a response to the aforementioned query will be presented, An information processing apparatus comprising: presentation means for presenting to the subject a response with content corresponding to the attribute, generated using a language model.

2. The system includes adjustment means that performs at least one of the following: adding information corresponding to the attribute to the output data obtained by inputting the query into the language model; and modifying or deleting a part of the output data according to the attribute. The information processing apparatus according to claim 1, wherein the presentation means presents the output data after processing by the adjustment means to the target person.

3. The information processing apparatus according to claim 2, wherein the adjustment means adds an advertisement or notice to the output data as information corresponding to the attribute.

4. The system includes an evaluation means for evaluating the output data obtained by inputting the query into the language model based on the attributes, The information processing apparatus according to claim 1, wherein the presentation means presents the output data to the subject in a manner corresponding to the result of the evaluation.

5. The query is further provided with a query modification means for adding attribute information indicating the aforementioned attribute to the query, The information processing apparatus according to claim 1, wherein the presentation means presents output data obtained by inputting the query to the language model to which the attribute information has been added to the subject.

6. The system includes an acquisition means for obtaining multiple output data by repeatedly inputting the aforementioned query into the language model and generating output data multiple times. The information processing apparatus according to claim 1, wherein the presentation means presents to the subject one or more output data selected from among the acquired multiple output data according to the attribute.

7. The information processing apparatus according to claim 1, wherein the presentation means presents to the target person a response generated using a language model that corresponds to the attribute determined by the attribute determination means, from among the language models that have been optimized for each of the plurality of attributes.

8. The information processing apparatus according to any one of claims 1 to 7, wherein the attribute determination means determines the attributes of the target person from the query received by the receiving means using an attribute estimation model generated by machine learning the relationship between the query and the attributes of the person to whom a response to the query is to be presented.

9. At least one processor, Accepting query input, Determining the attributes of the person to whom the response to the aforementioned query will be presented, A response method comprising presenting to the subject a response whose content corresponds to the attribute, generated using a language model.

10. Computers, A means of receiving query input, Attribute determination means for determining the attributes of the target person for which a response to the aforementioned query will be presented, and A response program that functions as a presentation means for presenting a response to the target person, the response having content corresponding to the attribute, generated using a language model.