Information processing device, investigation system, adjustment method, and adjustment program

Abstract

To generate an answer suitable for business use of a predetermined user with a language model.SOLUTION: An information processing device includes a specification unit and an adjustment unit. The specification unit specifies data relating to business of a predetermined user as business-related data. The adjustment unit adjusts for the predetermined user a language model that has learned to output an answer to a query by using the business-related data. The thereby adjusted language model can also be used for supporting a decision making of the predetermined user.SELECTED DRAWING: Figure 1

Description

[Technical field]

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

[0002] There are known technologies related to language models. One example of the technology related to language models is a system for generating a custom language model, which is described in Patent Document 1. In this system, documents are clustered to generate cluster vectors for each cluster, and a target vector associated with a target profile is generated. Then, each of the cluster vectors is compared with the target vector, and a language model is generated using documents of a cluster selected based on the comparison. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2012-507809 Summary of the Invention [Problem to be solved by the invention]

[0004] Recently, application of language models to various tasks by various users has been considered. However, in the system of Patent Document 1, it may be difficult to generate a language model suitable for the task use of a specific user. This is because it is difficult to prepare clusters corresponding to various users and their tasks in advance. Therefore, in the language model generated by the system of Patent Document 1, it may be difficult to generate an answer suitable for the task use of a specific user.

[0005] The present disclosure has been made in consideration of the above problems, and an exemplary objective thereof is to provide a technology that enables generating answers that are suited to a specific user's business usage using a language model. [Means for solving the problem]

[0006] An information processing device according to an exemplary aspect of the present disclosure includes an identification means for identifying data related to a specific user's work as work-related data, and an adjustment means for adjusting a language model that has been machine-learned to output answers to queries for the specific user using the work-related data.

[0007] An information processing device according to an exemplary aspect of the present disclosure includes a reception means for receiving a query input from a specified user, and a response means for generating a response to the query using a language model that is machine-learned to output an answer to the query, the language model being adjusted for the specified user using business-related data related to the business of the specified user, or the response means for generating a response to the query received by the reception means using the business-related data and the language model.

[0008] In an adjustment method according to an exemplary aspect of the present disclosure, at least one processor executes an identification process for identifying data related to a specific user's work as work-related data, and an adjustment process for adjusting a language model trained by machine learning to output answers to queries for the specific user using the work-related data.

[0009] An adjustment program according to an exemplary aspect of the present disclosure causes a computer to function as an identification means for identifying data related to a specific user's work as work-related data, and an adjustment means for adjusting a language model machine-learned to output answers to queries for the specific user using the work-related data. Effect of the Invention

[0010] According to an exemplary aspect of the present disclosure, an exemplary effect is provided in that a technology is provided that enables the generation of answers that are suited to the business use of a specified user using a language model. [Brief description of the drawings]

[0011] [Figure 1] 1 is a block diagram showing a configuration of an information processing device according to the present disclosure. [Diagram 2] FIG. 1 is a flow diagram illustrating the coordination and response method of the present disclosure. [Diagram 3] FIG. 13 is a block diagram showing a configuration of another information processing device according to the present disclosure. [Figure 4] FIG. 2 is a flow diagram illustrating the steps of an intermediation method according to the present disclosure. [Diagram 5] FIG. 13 is a block diagram showing a configuration of still another information processing device according to the present disclosure. [Figure 6] 6 is a diagram illustrating an example of the operation of the information processing device illustrated in FIG. 5. [Figure 7] 6 is a flowchart showing a process flow for adjusting a language model by the information processing device shown in FIG. 5. [Figure 8] 6 is a flow diagram showing a flow of processes executed by the information processing device shown in FIG. 5 when generating and presenting an answer to a query. [Figure 9] FIG. 1 is a diagram illustrating a configuration of a survey system according to the present disclosure. [Figure 10] FIG. 13 is a block diagram showing a configuration of still another information processing device according to the present disclosure. [Figure 11] FIG. 13 is a block diagram showing a configuration of still another information processing device according to the present disclosure. [Figure 12] FIG. 10 is a flow diagram showing a process flow in the investigation system shown in FIG. [Figure 13] 10 is a diagram showing a flow of a question and answer session carried out between the information processing devices in the investigation system shown in FIG. [Figure 14] 11 is a flow chart showing an example of processing executed by the information processing device shown in FIG. [Figure 15] 12 is a flow chart showing an example of processing executed by the information processing device shown in FIG. [Figure 16] 12 is a flow diagram showing an example of a process in which the information processing device shown in FIG. 11 accepts a follow-up question. [Figure 17] 12 is a flow diagram showing an example of a process in which the information processing device shown in FIG. 11 evaluates an answer and renegotiates. FIG. [Figure 18] FIG. 1 is a block diagram showing a configuration of a computer that functions as an information processing device according to the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, an embodiment obtained by appropriately combining the technical means employed in each of the exemplary embodiments shown below may also be included in the scope of the present invention. In addition, an embodiment obtained by appropriately omitting a part of the technical means employed in each of the exemplary embodiments shown below may also be included in the scope of the present invention. In addition, the effects mentioned in each of the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, an embodiment that does not exhibit the effects mentioned in each of the exemplary embodiments shown below may also be included in the scope of the present invention.

[0013] First Exemplary Embodiment A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is the basic form of each exemplary embodiment described later. The scope of application of each technical means adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means adopted in this exemplary embodiment can be adopted in other exemplary embodiments included in this disclosure to the extent that no particular technical obstacle occurs. In addition, each technical means shown in the drawings referred to for explaining this exemplary embodiment can also be adopted in other exemplary embodiments included in this disclosure to the extent that no particular technical obstacle occurs.

[0014] (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 configurations of the information processing devices 1 and 2. As shown in Fig. 1, the information processing device 1 includes an identifying unit 101 and an adjusting unit 102.

[0015] The identification unit 101 identifies data related to the business of a predetermined user as business-related data. The predetermined user is a user who is the target of adjustment of a language model. There is no particular limitation on the type of user that is the predetermined user. In addition, "business" means work that is continuously performed on a daily basis in relation to business, commerce, etc. The type of work that is defined as "business" is arbitrary.

[0016] The business-related data may be data related to the business of a specific user. For example, the business-related data may be a daily report or monthly report that describes the business content of the specific user, a document created by the specific user for business, an email or message sent or received by the specific user for business, or a questionnaire answered by the specific user about the business. In addition to such documents filled out by the specific user for business, for example, various data related to the company to which the specific user belongs or the department to which the specific user belongs may be the business-related data. For example, documents explaining the company or department, documents created by the company or department, etc. may be the business-related data. Note that the business-related data is not limited to documents (text), and may be data such as graphs, figures, tables, audio, or images. Business-related data in a format other than text can be converted into text using, for example, a well-known text conversion technology and used.

[0017] The adjustment unit 102 adjusts the language model that has been machine-trained to output an answer to a query for a specific user by using the business-related data. A "language model" is a model that has been machine-trained to determine the arrangement of components (such as words) in a sentence or the arrangement of sentences in a piece of writing, and has been machine-trained to output an answer to a query as described above.

[0018] Note that the term "query" refers to an instruction or command to generate an answer to a language model. Therefore, in the following description, "query" can be replaced with "generation instruction" or "generation command." The format of the input query is not particularly limited. The input query may be in text format, or in other formats such as audio format, or may include data in multiple formats, such as a combination of text and image. Similarly, the format of the answer output to the query is not particularly limited.

[0019] The method of adjusting the language model using the work-related data is not particularly limited as long as it is a method capable of generating a language model for a predetermined user. For example, the adjustment unit 102 may re-learn the language model using the work-related data identified by the identification unit 101. Specifically, the adjustment unit 102 re-learns the language model using the text included in the work-related data as training data so that a part of the data can be inferred from the other parts. For example, assume that the work-related data includes the text "Sales department's goal for the first half of the year is to improve sales performance by 10%." In this case, the adjustment unit 102 re-learns the language model using this text so that the answer "Sales department's goal for the first half of the year" is output as "Improve sales performance by 10%." in response to the query "What is the sales department's goal for the first half of the year?".

[0020] Also, for example, the adjustment unit 102 may retrain the language model using training data in which correct answer data is associated with text included in the business-related data identified by the identification unit 101. For example, assume that the business-related data includes a text "business process of the human resources department." In this case, the adjustment unit 102 retrains the language model using training data in which text explaining the business process of the human resources department is associated with this text as correct answer data. Note that the process of associating the correct answer data may be performed by the information processing device 1 or may be performed by another device. Also, training data associated with the correct answer data may be manually input to the information processing device 1. Alternatively, the adjustment unit 102 may generate multiple answers from the language model for the same query, and use the result of the user selecting a preferred answer as learning data.

[0021] The adjustment unit 102 may adjust the language model for a specific user by registering the work-related data of the specific user as data to be referred to when generating an answer using the language model. In this way, when generating an answer to a query using the language model, the work-related data of the specific user is referred to, and the query is rewritten using the work-related data related to the query, thereby generating an answer suitable for the specific user. The work-related data related to the query can be detected, for example, by searching the work-related data with a character string extracted from the query. For example, assume that a query of "Please explain points to note regarding the current job title" is input. In this case, the "job title" of the specific user is searched for from the work-related data, and the job title is added to the query to make it "Please explain points to note regarding the current job title. The job title is X.", thereby allowing the language model to generate an answer suitable for the job title of the specific user. The adjustment unit 102 inputs the detected job title into the "X" portion of the rewritten query.

[0022] (Effects of information processing device 1) As described above, the information processing device 1 includes the identification unit 101 that identifies data related to a task of a specific user as task-related data, and the adjustment unit 102 that adjusts a language model that has been machine-learned to output an answer to a query, for the specific user, using the task-related data. Therefore, the information processing device 1 has the effect of being able to generate an answer suited to the task of the specific user, using the language model.

[0023] The language model adjusted in this way can also be used to assist the predetermined user in making decisions. For example, the user can cause the language model to generate an answer to a question given to the user. The answer generated in this way has a content corresponding to the user's past work-related data. The user can then decide how to answer the question by referring to the answer generated by the language model according to the user's past work-related data.

[0024] The identification unit 101 can also identify data related to the work of each of the multiple users as work-related data for each of the multiple users. The adjustment unit 102 can then adjust the language model for each user using each of the identified work-related data. In this case, multiple language models suited to each of the users are generated. The adjustment unit 102 can also register the work-related data corresponding to each of the multiple users as data to be referenced when generating an answer using the language model for each of the multiple users.

[0025] (Configuration of information processing device 2) The configuration of the information processing device 2 will be described with continued reference to Fig. 1. The information processing device 2 includes a receiving unit 201 and a responding unit 202, as shown in Fig. 1.

[0026] The reception unit 201 receives an input of a query. This query instructs the generation of an answer to the query using a language model. For example, the reception unit 201 may receive a question about a task of a specific user as a query.

[0027] The response unit 202 is a language model that has been machine-learned to output a response to a query, and generates a response to the query received by the reception unit 201 using a language model that has been adjusted for a specific user using task-related data related to the task of the specific user. This language model may be adjusted by the adjustment unit 102 of the information processing device 1, for example.

[0028] Furthermore, instead of using the language model adjusted by machine learning as described above, the response unit 202 may generate a response to the query received by the reception unit 201 by using the work-related data and the language model. For example, the response unit 202 may search for work-related data related to the query, and add the detected work-related data to the query, or rewrite the query based on the detected work-related data. Then, the response unit 202 may generate a response by inputting the added or rewritten query to the language model. Note that the reference destination of the work-related data for generating the response may be, for example, that registered by the adjustment unit 102 of the information processing device 1.

[0029] (Effects of information processing device 2) As described above, the information processing device 2 includes the reception unit 201 that receives a query input, and the response unit 202 that generates a response to the query received by the reception unit 201 using a language model that is machine-learned to output a response to the query and is adjusted for a specific user using task-related data related to the task of the specific user, or generates a response to the query received by the reception unit 201 using the task-related data and the language model. Therefore, the information processing device 2 has an effect of enabling a response suited to the task of a specific user to be generated using the language model.

[0030] Note that multiple language models suitable for each of the multiple users may be prepared. In this case, the response unit 202 generates a response using a language model corresponding to the user for whom the response is to be generated, among the multiple language models. Also, for each of the multiple users, a reference destination of the business-related data corresponding to the user may be registered. In this case, the response unit 202 generates a response using the business-related data and language model of the reference destination corresponding to the user for whom the response is to be generated, among the registered reference destinations.

[0031] The information processing devices 1 and 2 can also be applied to the fields of medicine and healthcare. For example, medical data from a certain medical institution can be used as business-related data. This can adjust a language model that can generate answers to questions about medical treatment similar to those given by medical professionals belonging to the medical institution. The answers generated in this way can also be used as, for example, a second opinion.

[0032] (Adjustment Program) The above-mentioned functions of the information processing device 1 can also be realized by a program. The adjustment program according to this exemplary embodiment causes a computer to function as a specifying means for specifying data related to a task of a specific user as task-related data, and as an adjustment means for adjusting a language model machine-learned to output an answer to a query for the specific user using the task-related data. Therefore, according to the adjustment program according to this exemplary embodiment, it is possible to obtain an effect that it is possible to generate an answer suitable for the task of the specific user using the language model.

[0033] (Response Program) The above-mentioned functions of the information processing device 2 can also be realized by a program. The response program according to this exemplary embodiment causes a computer to function as a reception means for receiving a query input, and a response means for generating a response to a query received by the reception means using a language model that is machine-learned to output a response to the query and is adjusted for a specific user using task-related data related to the task of the specific user, or generating a response to a query received by the reception means using task-related data and a language model. Therefore, according to the response program according to this exemplary embodiment, it is possible to obtain an effect that a response suited to the task of a specific user can be generated using a language model.

[0034] (Adjustment method flow) The flow of the adjustment method will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the adjustment method and the response method. The execution subject of each step in the adjustment method shown in Fig. 2 may be a processor provided in the information processing device 1, or may be a processor provided in another device, or each step may be a processor provided in a different device. Similarly, the execution subject of each step in the response method shown in Fig. 2 may be a processor provided in the information processing device 2, or may be a processor provided in another device, or each step may be a processor provided in a different device.

[0035] As shown in flow F1 of Figure 2, this adjustment method includes an identification process S11 in which at least one processor identifies data related to a specific user's business as business-related data, and an adjustment process S12 in which at least one processor uses the business-related data to adjust a language model that has been machine-learned to output an answer to a query, for the specific user.

[0036] (Effect of adjustment method) As described above, the present adjustment method includes a process in which at least one processor identifies data related to a task of a specific user as task-related data, and a process in which a language model machine-learned to output an answer to a query is adjusted for the specific user using the task-related data. Therefore, the present adjustment method has the effect of enabling the generation of an answer suited to the task of the specific user using the language model.

[0037] (Response method flow) The flow of the present response method will be described with reference to Fig. 2. As shown in flow F2 of Fig. 2, the present response method includes a reception process S21 in which at least one processor receives an input of a query, and a response process S22 in which at least one processor generates a response to the query received in the reception process S21 using a language model that is machine-learned to output a response to the query and that is adjusted for a specific user using task-related data related to the task of the specific user, or generates a response to the query received in the reception process S21 using the task-related data and the language model, as shown in flow F2 of Fig. 2.

[0038] (Effect of response method) As described above, the response method includes a reception process in which at least one processor receives a query input, and a response process in which a language model that is machine-learned to output a response to the query is used to generate a response to the query received in the reception process S21 using a language model adjusted for a specific user using task-related data related to the task of the specific user, or a response to the query received in the reception process is generated using the task-related data and the language model. Therefore, the response method has the effect of enabling a response suited to the task of the specific user to be generated using the language model.

[0039] (Configuration of information processing device 3) The configuration of the information processing device 3 will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the information processing device 3. As shown in Fig. 3, the information processing device 3 includes a receiving unit 301 and a negotiation unit 302.

[0040] The reception unit 301 receives a request for a survey to collect answers to a predetermined question. The contents of the survey and the questions are not particularly limited. For example, the reception unit 301 may receive a request for a marketing survey. Also, for example, the reception unit 301 may receive a request for a questionnaire survey targeting people who belong to a predetermined company or people in a predetermined occupation such as medical workers. The reception unit 301 may receive one or more requests from one requester, or may receive one or more survey requests from each of multiple requesters. The survey request may include at least one question to be answered. In the following, a marketing survey will be described as an example, but the request contents received by the reception unit 301 are not limited to a survey, and any request can be received as long as it corresponds to information that the language model can output.

[0041] The negotiation unit 302 negotiates with a predetermined negotiation partner to have a proxy answer language model that has been machine-learned to generate an answer on behalf of a predetermined user generate an answer to an input question. The predetermined negotiation partner may be a person (e.g., the above-mentioned predetermined user), a device (e.g., the above-mentioned information processing device 2), or the proxy answer language model. The "negotiation" performed by the negotiation unit 302 may include, for example, at least notifying the negotiation partner of at least one of the question for which an answer is desired to be generated and the conditions for the answer to the question, and receiving an answer to the notification from the negotiation partner.

[0042] The proxy answer language model may be one that has been machine-learned so as to be capable of generating an answer on behalf of a specific user. For example, a language model adjusted by the above-mentioned adjustment unit 102 may be used as the proxy answer language model. In addition, a language model in which the work-related data of a specific user is registered as data to be referred to when generating an answer by the above-mentioned adjustment unit 102 may be used as the proxy answer language model.

[0043] (Effects of information processing device 3) As described above, the information processing device 3 includes a receiving unit 301 that receives a request for a survey to collect answers to a predetermined question, and a negotiation unit 302 that negotiates with a predetermined negotiating party to have a proxy answer language model that has been machine-learned to generate answers to an input question on behalf of a predetermined user generate an answer to the question. Therefore, the information processing device 3 has an effect of making it possible to easily carry out a survey to collect answers to a predetermined question. In addition, the information processing device 3 also makes it possible to optimize the answerers to questions. For example, the information processing device 3 makes it possible to negotiate with a plurality of answerers, i.e., proxy answer language models, to have an optimal proxy answer language model generate an answer.

[0044] (Intermediation Program) The functions of the information processing device 3 described above can also be realized by a program. The intermediation program according to this exemplary embodiment causes a computer to function as a receiving means for receiving a request for a survey to collect answers to a predetermined question, and as a negotiating means for negotiating with a predetermined negotiating party to have a proxy answer language model that has been machine-learned so as to be able to generate answers to input questions on behalf of a predetermined user generate answers to the questions. Therefore, according to the intermediation program according to this exemplary embodiment, it is possible to obtain an effect that a survey to collect answers to a predetermined question can be easily performed.

[0045] (Flow of mediation method) The flow of the intermediation method will be described with reference to Fig. 4. As shown in flow F3 in Fig. 4, this intermediation method includes a reception process S31 in which at least one processor receives a request for a survey to collect answers to a predetermined question, and a negotiation process S32 in which at least one processor negotiates with a predetermined negotiation partner to have a proxy answer language model that has been machine-learned so as to be able to generate answers to input questions on behalf of a predetermined user generate answers to the questions.

[0046] (Effect of mediation method) As described above, the present intermediation method includes a reception process in which at least one processor receives a request for a survey to collect answers to a predetermined question, and a negotiation process in which at least one processor negotiates with a predetermined negotiating party to have a proxy answer language model that has been machine-learned to be able to generate answers to the input question on behalf of the predetermined user generate an answer to the question. Therefore, according to the present intermediation method, it is possible to obtain an effect that a survey to collect answers to the predetermined question can be easily performed.

[0047] Second Exemplary Embodiment A second exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be given the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technical means employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means employed in this exemplary embodiment can be employed in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs. In addition, each technical means shown in each drawing referred to for explaining this exemplary embodiment can be employed in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs. The above matters are also applicable to the third exemplary embodiment described later.

[0048] (Configuration of information processing device 1A) The configuration of the information processing device 1A will be described with reference to FIG. 5. FIG. 5 is a block diagram showing the configuration of the information processing device 1A. As shown in the figure, the information processing device 1A includes a control unit 10A that controls each unit of the information processing device 1A, and a storage unit 11A that stores various data used by the information processing device 1A. The information processing device 1A also includes a communication unit 12A that allows the information processing device 1A to communicate with other devices, an input unit 13A that accepts input of various data to the information processing device 1A, and an output unit 14A that allows the information processing device 1A to output various data. As shown in the figure, the control unit 10A also includes an identification unit 101A, an adjustment unit 102, a reception unit 201, a response unit 202A, and a presentation unit 203A.

[0049] The identification unit 101A identifies data related to a task of a predetermined user as task-related data, similar to the identification unit 101 of the information processing device 1. The details will be described based on Fig. 6, but the identification unit 101A differs from the identification unit 101 in that the identification unit 101A identifies task-related data by using hierarchical information indicating the hierarchical structure of an organization to which the predetermined user belongs.

[0050] The response unit 202A generates a response to a query input by a predetermined user using a language model, similar to the response unit 202 of the information processing device 2. The response unit 202A differs from the response unit 202 in that the response unit 202A uses a language model 4A that is generated or updated using the task-related data identified using the hierarchical information.

[0051] The language model 4A is a language model machine-learned to output a response to a query. The language model 4A is adapted to a specific user through adjustment by the adjustment unit 102 using task-related data. A language model machine-learned to output a response to a query without such adjustment may be applied as the language model 4A. In this case, the adjustment unit 102 registers task-related data of a specific user in association with the language model 4A as data to be referred to when generating a response using the language model 4A. Then, the response unit 202A generates a response to a query using the task-related data and the language model 4A. Although the language model 4A is shown outside the information processing device 1A in FIG. 5, the language model 4A may be stored inside the information processing device 1A (for example, the storage unit 11A).

[0052] The presentation unit 203A presents the answer generated by the response unit 202A to a predetermined user. The presentation manner of the answer is not particularly limited. For example, the presentation unit 203A may present the answer by displaying and outputting it on a display device, by outputting it as sound on a sound output device, or by printing it on a printing device. In addition, the device that presents the answer (for example, the above-mentioned display device, sound output device, or printing device) may be included in the information processing device 1A, or may be an external device to the information processing device 1A.

[0053] The information processing device 1A can perform both generating an answer using the language model 4A and adjusting the language model 4A (updating the language model 4A or adding business-related data to be referenced). Therefore, the information processing device 1A can adjust the language model using a query input by a user. As in the first exemplary embodiment, the device that adjusts the language model (device corresponding to the information processing device 1) and the device that generates an answer using the language model (device corresponding to the information processing device 2) may be separate, independent devices.

[0054] (Example of operation) An operation example of the information processing device 1A will be described with reference to Fig. 6. Fig. 6 is a diagram showing an operation example of the information processing device 1A. Fig. 6 shows an example in which the information processing device 1A is a smartphone. The information processing device 1A may be any device capable of realizing the functions of each functional block shown in Fig. 5, and is not limited to a smartphone.

[0055] The information processing device 1A accepts an input of a query, generates an answer to the query, and presents the generated answer. For example, as shown in Fig. 6, assume that a specific user using the information processing device 1A inputs a query such as "Please tell me the characteristics of the X division of our company." In this case, the reception unit 201 of the information processing device 1A accepts the input of the query, the response unit 202A generates an answer to the query using the language model 4A, and the presentation unit 203A displays the answer on the display unit of the information processing device 1A.

[0056] 6 is a language model that is machine-learned to output an answer to a query, and is a language model that is adjusted for a specific user using task-related data related to the task of the specific user. This language model 4A differs from the language model described in the exemplary embodiment 1 in that it is a model that is generated or updated using task-related data identified by the identification unit 101A using hierarchical information.

[0057] As described above, the identification unit 101A differs from the information processing device 1 in that, in identifying business-related data, the identification unit 101A uses hierarchical information indicating the hierarchical structure of an organization to which a specific user belongs. More specifically, the identification unit 101A uses the hierarchical information to identify a hierarchical level having a specific relationship with the hierarchical level to which the specific user belongs, and identifies business-related data related to the identified hierarchical level.

[0058] 6 shows a schematic diagram of the hierarchical structure of each department belonging to the X division of a company to which a specific user belongs. As shown in the figure, the X division includes departments X1 to X4. Department X1 belongs to the highest level, and departments X2 and X3 belong to the level immediately below X1. Department X4 belongs to the level below department X2. The hierarchical information may be information indicating such a hierarchical structure of each department, that is, the level to which each department belongs and the relationship between the levels.

[0059] In FIG. 6, task-related data d1, d2, d31, d32, d4 related to departments X1 to X4 are shown as examples of task-related data that can be used for learning the language model 4A. Although all of these data can be broadly said to be related to the task of a specific user, there is a possibility that the data may include data that is less related to the task of the specific user. Therefore, the identification unit 101A uses hierarchical information to identify, among these task-related data, data that is highly related to the task of the specific user. This can improve the suitability of the language model 4A to the task of the specific user.

[0060] For example, assume that a specific user belongs to department X2. In this case, the identification unit 101A may identify, in addition to the business-related data d2 related to department X2, business-related data d4 related to department X4 belonging to the hierarchy immediately below department X2 as business-related data used for generating or updating the language model 4A. In addition to the business-related data d4, or instead of the business-related data d4, the identification unit 101A may identify, as business-related data used for generating or updating the language model 4A, business-related data d1 related to department X1 belonging to the hierarchy immediately above department X2. In addition to the business-related data d4, or instead of the business-related data d4, the identification unit 101A may identify, as business-related data used for generating or updating the language model 4A, business-related data d31 and d32 related to department X3 in the same hierarchy as department X2.

[0061] Here, the business-related data related to one department may include data suitable for sharing and data not suitable for sharing. For this reason, a confidentiality level may be set for each business-related data, and the identifying unit 101A may identify the business-related data according to the set confidentiality level as the business-related data to be used for generating or updating the language model 4A.

[0062] For example, assume that the confidentiality level of the business-related data d31 shown in FIG. 6 is set to "1" and the confidentiality level of the business-related data d32 is set to "0." Here, "1" indicates that confidentiality is required, and "0" indicates that confidentiality is not required. In this case, the identification unit 101A identifies the business-related data d32 as business-related data used to generate or update the language model 4A, but does not identify the business-related data d31 as business-related data used to generate or update the language model 4A. This makes it possible to prevent the business-related data to be confidential from being used for learning the language model 4A and leaking the business-related data to be confidential via the output of the language model 4A.

[0063] The confidentiality level may indicate whether or not data should be kept confidential, or may indicate the degree to which data should be kept confidential. In the latter case, the confidentiality level to which business-related data can be used may be determined for each user. In this case, the information processing device 1A can generate or update the language model 4A using business-related data with a confidentiality level according to the user.

[0064] The adjustment unit 102 of the information processing device 1A generates or updates the language model 4A using the task-related data identified by the identification unit 101A in the above manner, thereby making it possible to obtain a language model 4A that is suited to the hierarchical structure of the organization to which a given user belongs.

[0065] Instead of generating or updating the language model 4A, the adjustment unit 102 may perform adjustment to register the task-related data identified by the identification unit 101A as data to be referred to when the response unit 202A generates a response using the language model. In this case, the same language model as in the exemplary embodiment 1 may be applied as the language model 4A.

[0066] As described above, the identification unit 101A identifies a hierarchy having a predetermined relationship with the hierarchy to which a specific user belongs, using hierarchy information indicating the hierarchical structure of the organization to which the specific user belongs, and identifies business-related data related to the identified hierarchy. Therefore, according to the information processing device 1A, in addition to the effects of the information processing device 1, the effect of being able to identify business-related data useful for adjusting the language model 4A by taking into account the hierarchical structure of the organization to which the user belongs can be obtained.

[0067] As described above, the identifying unit 101A may identify the business-related data according to the confidentiality level set for each piece of data related to the business of a specific user. This provides an effect that the business-related data can be used in stages according to the confidentiality level in addition to the effect provided by the information processing device 1.

[0068] (Process flow when adjusting language model 4A) The flow of processing when the information processing device 1A adjusts the language model 4A will be described with reference to Fig. 7. Fig. 7 is a flow diagram showing the flow of processing when the information processing device 1A adjusts the language model 4A.

[0069] In S11a in the flow F1a shown in FIG. 7, the identification unit 101A identifies a predetermined user as a target user. The method of identifying the predetermined user is not particularly limited. For example, the operator of the information processing device 1A may input identification information of the predetermined user via the communication unit 12A or the input unit 13A. In this case, the identification unit 101A identifies the user identified by the input identification information as the predetermined user.

[0070] In S12a, the identification unit 101A identifies the department to which the user identified in S11a belongs. The method of identifying the department is not particularly limited. For example, the user of the information processing device 1A may input information indicating the department to which the user belongs via the communication unit 12A or the input unit 13A. In this case, the identification unit 101A identifies the department indicated in the input information as the department to which the specified user belongs.

[0071] In S13a, the identification unit 101A identifies a hierarchical level having a predetermined relationship with the hierarchical level to which the department identified in S12a belongs. Then, the identification unit 101A identifies the departments that belong to the identified hierarchical level, i.e., related departments. For example, the identification unit 101A may identify as related departments any of a department in the same hierarchical level as the belonging department, a department in a higher hierarchical level than the belonging department, and a department in a lower hierarchical level than the belonging department. Note that, when business-related data is managed on a hierarchical level basis, the identification unit 101A does not need to identify related departments, and only needs to identify related hierarchical levels.

[0072] In S14a, the identification unit 101A identifies data related to the business of a specific user as business-related data. More specifically, the identification unit 101A identifies data related to the related department identified in S13a from among the data related to the business of the specific user as business-related data. Note that the data related to the related department can also be said to be data related to a hierarchy related to the hierarchy to which the user belongs.

[0073] In S15a, the adjustment unit 102 adjusts the language model machine-learned to output a response to a query for a specific user using the task-related data identified in S14a. For example, the adjustment unit 102 may generate a language model 4A using the task-related data identified in S14a. Also, for example, the adjustment unit 102 may generate the language model 4A by updating the language model machine-learned to output a response to a query using the task-related data identified in S14a. The generated language model 4A is stored in a specific storage device, and the process of flow F1a ends. Also, the adjustment unit 102 may adjust the language model for a specific user by registering the task-related data identified in S14a as data to be referred to when generating a response by the language model.

[0074] (Process flow when responding to a query) The flow of processing when information processing device 1A generates and presents a response to a query will be described with reference to Fig. 8. Fig. 8 is a flow diagram showing the flow of processing executed when information processing device 1A generates and presents a response to a query.

[0075] In S21a in the flow F2a shown in Fig. 8, the reception unit 201 receives a query input. At this time, the presentation unit 203A may display a UI screen for receiving the query input on, for example, a display unit of a terminal device owned by a predetermined user. In this case, the reception unit 201 receives the query input via the terminal device. Note that the reception unit 201 may receive the query input via the input unit 13A.

[0076] In S22a, the response unit 202A generates a response to the query input in S21a by using a language model 4A generated by using the task-related data of a specific user. The response unit 202A may generate a response by using the task-related data of the specific user who input the query in S21a and a language model machine-learned to output a response to the query.

[0077] In S23a, the presenting unit 203A presents the answer generated in S22a. For example, when a query input is received via a terminal device, the presenting unit 203A may cause the terminal device to output the answer. The presenting unit 203A may also present the answer via the output unit 14A. This ends the processing of the flow F2a.

[0078] Note that language models 4A suitable for each of the multiple users may be prepared in advance. In this case, in S21a, the reception unit 201 identifies the target user by having the user input identification information, etc. Then, in S22a, the response unit 202A generates a response using the language model 4A suitable for the identified user. Also, instead of preparing language models 4A suitable for each of the multiple users, a reference destination for referencing the work-related data of each of the multiple users may be registered. In this case, in S22a, the response unit 202A generates a response using the work-related data identified by referring to the reference destination corresponding to the identified user.

[0079] Third Exemplary Embodiment (Configuration of Survey System 7B) The configuration of the investigation system 7B will be described with reference to FIG. 9. FIG. 9 is a diagram showing the configuration of the investigation system 7B. The investigation system 7B is a system that accepts a request to conduct an investigation and outputs the investigation results. As shown in the figure, the investigation system 7B includes an information processing device 1B and an information processing device 3B. Note that the number of information processing devices 1B and information processing devices 3B included in the investigation system 7B is arbitrary and is not limited to the example shown in the figure.

[0080] In the investigation system 7B, the information processing device 3B accepts a request for an investigation to collect answers to a given question. Then, the information processing device 3B negotiates with the information processing device 1B regarding conditions for the proxy answer language model 4B to generate an answer to the question. The negotiation language model 5B is used. In this way, the information processing device 3B functions as an intermediary device that acts as an intermediary between the information processing device 1B and the source of the investigation request.

[0081] The proxy answer language model 4B is a language model that has been machine-learned to generate an answer to an input question on behalf of a specific user. The information processing device 1B plays a role as a proxy or personal secretary for the user of the device by using the proxy answer language model 4B. For example, the information processing device 1B accepts an input of a question for the user of the device, and determines whether or not to cause the proxy answer language model 4B to generate an answer to the question. Each information processing device 1B is associated with a proxy answer language model 4B that generates an answer on behalf of the user.

[0082] The proxy answer language model 4B may be one that has been machine-learned so as to be able to generate an answer on behalf of a specific user. For example, a language model adjusted by the above-mentioned adjustment unit 102 may be used as the proxy answer language model 4B. In addition, a language model in which the work-related data of a specific user is registered as data to be referred to when generating an answer by the above-mentioned adjustment unit 102 may be used as the proxy answer language model 4B.

[0083] It is preferable that access to the information processing device 1B is restricted to those other than the user and the information processing device 3B, thereby making it possible to prevent information about the user from being leaked via the proxy answer language model 4B.

[0084] As described above, the survey system 7B includes an information processing device 1B that receives an input of a query and generates a response to the query using the proxy answer language model 4B, which is a language model that has been machine-learned to output a response to the query and is adjusted for a specific user using task-related data related to the task of the specific user, and an information processing device 3B that is an intermediary device that receives a request for a survey to collect answers to a specific question and negotiates with the information processing device 1B to have the proxy answer language model 4B generate the answer to the question. This survey system 7B has the effect of making it possible to easily conduct a survey to collect answers to a specific question.

[0085] The information processing device 1B may generate an answer to a query using the task-related data and the proxy answer language model 4B. In this case, the proxy answer language model 4B may be a language model machine-learned to output an answer to a query, and does not need to be adjusted using task-related data related to a specific user's task. It is sufficient to register the reference destination of the task-related data used when generating an answer in association with the proxy answer language model 4B.

[0086] (Configuration of information processing device 1B) The configuration of the information processing device 1B will be described with reference to Fig. 10. Fig. 10 is a block diagram showing the configuration of the information processing device 1B. As shown in the figure, the information processing device 1B includes a control unit 10B that controls each unit of the information processing device 1B. The control unit 10B includes a receiving unit 201B, a responding unit 202B, a presenting unit 203B, an authenticating unit 204B, a satisfaction determining unit 205B, an answer possibility determining unit 206B, an alternative condition generating unit 207B, a negotiating unit 208B, a reliability determining unit 209B, a correcting unit 210B, and a proofreading unit 211B.

[0087] The information processing device 1B may include the specification unit 101 or 101A and the adjustment unit 102. Details of the answer possibility determination unit 206B and the proofreading unit 211B will be described later with reference to Fig. 12, and details of the alternative condition generation unit 207B, the reliability determination unit 209B, and the correction unit 210B will be described later with reference to Fig. 14. The remaining components will be described below.

[0088] The receiving unit 201B receives an input of a query, similar to the receiving unit 201 of the information processing device 2. For example, the receiving unit 201B receives an input of a question for a predetermined user. Here, the predetermined user is a user corresponding to the proxy answer language model 4B used by the information processing device 1B.

[0089] The receiving unit 201B may also receive conditions for the answer to the question together with the question. The conditions are set as appropriate by the requester. For example, the conditions may include at least any of the following: a reward for providing the answer, the answering method (e.g., multiple choice or written answer, etc.), whether additional questions can be asked, the attributes of the answerer (e.g., field of expertise, work history, qualifications, etc.), and the range of information to be included in the answer. The range of information to be included in the answer may be determined based on, for example, the confidentiality level set for each piece of information.

[0090] The response unit 202B, like the response unit 202 of the information processing device 2, generates a response to a query received by the reception unit 201B using a language model (specifically, the proxy answer language model 4B) that is machine-learned to output a response to a query and is adjusted for a specific user using task-related data related to the task of the specific user. Also, instead of using the proxy answer language model 4B that is machine-learned to suit the specific user, the response unit 202B may generate a response using the task-related data of the specific user and the proxy answer language model 4B that is not adjusted for the specific user.

[0091] The presentation unit 203B presents the answer generated by the response unit 202B to the user. The presentation manner of the answer is not particularly limited. For example, the presentation unit 203B may present the answer by displaying and outputting an image showing the answer content on a display device, may present the answer by outputting a sound showing the answer content on a sound output device, or may present the answer by printing out the answer content on a printing device. In addition, the device that presents the answer (for example, the above-mentioned display device, sound output device, or printing device) may be included in the information processing device 1B, or may be an external device to the information processing device 1B.

[0092] When the receiving unit 201B receives an input of a question, the authentication unit 204B judges whether the sender of the question is legitimate or not. Specifically, if the sender of the question received by the receiving unit 201B is the information processing device 3B, the authentication unit 204B judges that the sender is a legitimate sender, and if the sender is not the information processing device 3B, the authentication unit 204B judges that the sender is not a legitimate sender. The method of judging whether the sender of the question is the information processing device 3B is not particularly limited. For example, the information processing device 3B may be made to transmit its identification information, and in this case, the authentication unit 204B may use the received identification information to judge whether the sender of the question is the information processing device 3B or not. In addition to this, for example, the authentication unit 204B may apply a technology such as a block chain to judge whether the sender of the question is the information processing device 3B or not.

[0093] The satisfaction determination unit 205B determines whether or not the user of the information processing device 1B satisfies the condition accepted by the acceptance unit 201B. For example, if the attribute of the respondent is specified in the condition accepted by the acceptance unit 201B, the satisfaction determination unit 205B refers to attribute information indicating the attribute of the user of the information processing device 1B and determines whether or not the condition is satisfied. For example, it is assumed that the attribute of the respondent is specified to have work experience in a specific department for a predetermined number of years or more. In this case, the satisfaction determination unit 205B refers to attribute information indicating the work history of the user of the information processing device 1B and determines whether or not the condition is satisfied. Similarly, the range in which a question can be answered (for example, the confidentiality level of information that can be included in the answer, etc.) may also be registered in advance as attribute information.

[0094] Furthermore, the satisfaction determination unit 205B may determine the attributes of the user from the business history data of the user of the information processing device 1B, and determine whether or not the condition is satisfied based on the determination result. For example, it is assumed that the condition specifies that the respondent belongs to a specific specialty. In this case, the satisfaction determination unit 205B may determine the degree of agreement between the specialty of the user of the information processing device 1B and the specialty indicated in the condition, from the business history data of the user. In this way, the satisfaction determination unit 205B may determine to what extent the condition is satisfied, rather than whether or not the condition is satisfied.

[0095] Also, if a reward for answering a question is specified as a condition, the satisfaction determination unit 205B may determine that the condition is satisfied if a reward desired by the user of the information processing device 1B is specified, and may determine that the condition is not satisfied if the reward is not specified. The user's desire may be stored in advance in the storage unit 11A or the like. The satisfaction determination unit 205B may also determine the degree to which the condition regarding the reward is satisfied. The satisfaction determination unit 205B may also determine whether or not each of a plurality of conditions is satisfied, or the degree of satisfaction. Then, the satisfaction determination unit 205B may determine the overall degree of satisfaction of the conditions based on the determination results for each condition.

[0096] The negotiation unit 208B negotiates with the information processing device 3B as to whether or not to cause the proxy answer language model 4B to generate an answer to the question accepted by the acceptance unit 201B. For example, the negotiation unit 208B performs a process of inquiring of the information processing device 3B about the question and the conditions, notifying the information processing device 3B of alternative conditions and inquiring of whether or not to accept the alternative conditions, and the like.

[0097] (Configuration of information processing device 3B) The configuration of the information processing device 3B will be described with reference to FIG. 11. FIG. 11 is a block diagram showing the configuration of the information processing device 3B. As shown in the figure, the information processing device 3B includes a control unit 30B that controls each unit of the information processing device 3B and a storage unit 31B that stores various data used by the information processing device 3B. The information processing device 3B also includes a communication unit 32B for the information processing device 3B to communicate with other devices, an input unit 33B that accepts input of various data to the information processing device 3B, and an output unit 34B for the information processing device 3B to output various data. As shown in the figure, the control unit 30B includes a reception unit 301B, a negotiation unit 302B, a question classification unit 303B, a request destination determination unit 304B, an alternative condition generation unit 305B, an answer evaluation unit 306B, a question addition unit 307B, and a report unit 308B. The answer evaluation unit 306B will be described later with reference to FIG. 17, and the question addition unit 307B will be described later with reference to FIG. 16.

[0098] The receiving unit 301B receives a request for a survey to collect answers to a predetermined question, similar to the receiving unit 301 of the information processing device 3. The receiving unit 301B may also receive conditions for the answers to the questions together with the questions.

[0099] The negotiation unit 302B, like the negotiation unit 302 of the information processing device 3, negotiates with a predetermined negotiation partner to have the proxy answer language model 4B, which has been machine-learned to be able to generate an answer to an input question on behalf of a predetermined user, generate an answer to the question. The predetermined negotiation partner may be a person (e.g., the above-mentioned predetermined user), a device, or the proxy answer language model 4B. In this exemplary embodiment, an example in which the predetermined negotiation partner is the information processing device 1B will be mainly described.

[0100] For example, the negotiation unit 302B may notify a predetermined negotiation partner of the conditions for the answer to the question associated with the question in the survey accepted by the acceptance unit 301B, and determine whether or not to request the negotiation partner to answer the question based on the answer to the notification received from the negotiation partner. This provides an effect that, in addition to the effect provided by the information processing device 3, it is possible to automatically perform negotiations that take into account the conditions for the answer.

[0101] As described above, the conditions are appropriately set by the requester. For example, the conditions may include at least any of the following: reward for providing an answer, answering method, whether additional questions can be asked, attributes of the answerer, and the range of information to be included in the answer. In this way, in addition to the effects of the information processing device 3, it is possible to automatically perform negotiations that take into account the above-mentioned conditions, which are important points in negotiations.

[0102] Also, for example, the negotiation unit 302B may negotiate using the negotiation language model 5B generated by machine learning using words exchanged in negotiations between people, in other words, negotiation history expressed in natural language, as training data. This provides an effect that, in addition to the effect of the information processing device 3, it is possible to realize negotiations similar to negotiations between people using the negotiation language model 5B. Note that the training data for the negotiation language model 5B may include words and sentences other than the words exchanged in negotiations between people. It is also possible to negotiate using the negotiation language model 5B generated without using words exchanged in negotiations between people as training data.

[0103] The question classification unit 303B classifies questions included in the multiple requests received by the reception unit 301B according to the target of the question. As a result, the questions included in the multiple requests are aggregated for each target. The target of the question may be specified, for example, as a condition for answering the question. For example, if there are multiple questions for which the condition for the answerer is that the answerer has a predetermined number of years or more of work experience in a specific department, the question classification unit 303B may classify those questions into the same category. When the question classification unit 303B classifies the questions, the negotiation unit 302B negotiates with the negotiation partner corresponding to the category for the questions classified into the same category.

[0104] The request destination determination unit 304B determines a destination to be requested to answer the question. The request destination is selected from among a plurality of information processing devices 1B (which can also be called proxy answer language models 4B). In addition, when the question classification unit 303B classifies the questions, the request destination determination unit 304B determines a destination to be requested to answer for each question classification.

[0105] The request destination determination unit 304B may determine the request destination based on the result of the negotiation by the negotiation unit 302B. For example, the request destination determination unit 304B may determine, as the request destination, some or all of the information processing devices 1B that have agreed to answer the question among the information processing devices 1B with which the negotiation unit 302B has negotiated. The request destination determination unit 304B may also determine the request destination (which in this case can also be called the negotiation destination) prior to the negotiation by the negotiation unit 302B.

[0106] When the answer to the question is not accepted by the negotiation partner, the alternative condition generating unit 305B generates an alternative condition related to the answer to the question. A method for generating alternative conditions by the alternative condition generating unit 305B will be described later with reference to FIG.

[0107] The report unit 308B reports to the requester the answer that the negotiation unit 302B has acquired from the information processing device 1B as a result of the above-mentioned negotiation. The format of the report is not particularly limited. For example, the report unit 308B may directly notify the requester of the answer acquired from the information processing device 1B. The report unit 308B may also generate a report by tallying up the answers acquired from the information processing device 1B, and transmit the generated report to the requester. The report may include statistics based on the results of the above-mentioned tallying, a graph generated using the statistics, keywords extracted from the answers, a summary of the tallying results, and the like. The report unit 308B may also generate such a report by using, for example, a language model.

[0108] (Processing flow in the investigation system 7B) The process flow in the investigation system 7B will be described with reference to Fig. 12. Fig. 12 is a flow diagram showing the process flow in the investigation system 7B. Fig. 12 shows a flow F3a executed by the information processing device 3B and a flow F1b executed by the information processing device 1B.

[0109] In S31a of flow F3a, the reception unit 301 of the information processing device 3B receives a request for a survey to collect answers to a predetermined question. For example, the reception unit 301 may receive a survey request from one requester as in the example of Fig. 9, or may receive a survey request from each of multiple requesters. The reception unit 301 may also receive multiple survey requests from one requester.

[0110] The survey request received by the receiving unit 301B includes conditions regarding the answer in addition to the question to be answered. The conditions can be set arbitrarily. For example, the conditions may include at least one of the following: whether or not there is a reward for providing the answer, the content of the reward (such as the amount of the reward if the reward is monetary), the answering method (multiple choice or written answer, etc.), whether or not additional questions can be asked, the attributes of the respondent (such as gender, age, department, field of expertise, etc.), and the confidentiality level of the information to be included in the answer.

[0111] In S32a, the question classification unit 303B classifies the questions included in the request accepted in S31a according to the target of the question. If multiple requests are accepted in S31a, the question classification unit 303B classifies each question included in each request. As a result, multiple questions are aggregated for each target.

[0112] In S33a, the request destination determination unit 304B determines a request destination to be requested to answer the question for each classification in S32a. The request destination is selected from among a plurality of information processing devices 1B (which may also be referred to as proxy answer language models 4B).

[0113] The method of determining the request destination in S33a may be determined in advance. For example, when the request destination determination unit 304B determines the request destination for each question classification, it may determine the information processing device 1B of a user who satisfies the condition corresponding to the classification (for example, those who have a certain number of years or more of work experience in a specific department). Also, for example, when a budget limit is set for the investigation request, the request destination determination unit 304B may determine the information processing device 1B to be the request destination according to the budget limit. In this case, a standard remuneration amount may be set in advance for each information processing device 1B. Also, the request destination determination unit 304B may determine the information processing device 1B to be the request destination after inquiring about the remuneration amount from each information processing device 1B.

[0114] In S34a and S35a, negotiation by the negotiation unit 302B, that is, negotiation processing, is performed. Specifically, in S34a, the negotiation unit 302B transmits to the request destination determined in S33a a question for which the request destination is requested to answer, and conditions corresponding to the question. Then, in S35a, the negotiation unit 302B receives a response to the question transmitted in S34a from the request destination. The response received here is either a response from the proxy answer language model 4B to the transmitted question, or a response notifying that the proxy answer language model 4B cannot answer the transmitted question.

[0115] In addition, when negotiating, the negotiation unit 302B may determine whether the negotiation partner is a legitimate information processing device 1B. This makes it possible to prevent fraud of rewards by answering questions while pretending to be the information processing device 1B. The method of determining the legitimacy is arbitrary. For example, the negotiation unit 302B may determine the legitimacy using identification information, or may determine the legitimacy by applying a technology such as a block chain.

[0116] In S36a, the negotiation unit 302B judges whether or not answers to each question in the requested survey have been collected. If it is judged in S36a that the answers have been collected (YES in S36a), the process proceeds to S37a, and if it is judged that the answers have not been collected (NO in S36a), the process returns to S33a. In S33a, which is the transition from S36a, the request destination determination unit 304B determines a new request destination.

[0117] In S37a, the reporting unit 308B reports the answer acquired in the processes of S33a to S36a to the request source, whereby the process of the flow F3a ends.

[0118] On the other hand, in S11b in flow F1b, the receiving unit 201B of the information processing device 1B receives the question and the condition transmitted in S34a by the information processing device 3B. In other words, the execution subject of flow F1b is the information processing device 1B determined as the request destination in S33a of flow F3a among the multiple information processing devices 1B included in the investigation system 7B.

[0119] In S12b, authentication unit 204B judges whether the sender of the question and condition received in S11b is legitimate or not. If it is judged to be legitimate in S12b (YES in S12b), the process proceeds to S13b, and if it is judged to be invalid (NO in S12b), the process of flow F1b ends.

[0120] In S13b, the satisfaction determination unit 205B determines whether or not the user of the information processing device 1B satisfies the condition received in S11b. As described above, the satisfaction determination unit 205B may determine the degree to which the condition is satisfied.

[0121] In S14b, the answer admissibility determination unit 206B determines whether or not to have the proxy answer language model 4B generate an answer to the question based on the determination result of S13b. In this way, the answer admissibility determination unit 206B determines whether or not to have the proxy answer language model 4B generate an answer to the question for a predetermined user. This determination may be made based on the determination result by the satisfaction determination unit 205B. If it is determined in S14b that an answer is to be generated (YES in S14b), the process proceeds to S15b, and if it is determined that an answer is not to be generated (NO in S14b), the process proceeds to S17b. In S17b, which is transitioned from S14b, the answer admissibility determination unit 206B transmits an answer to the information processing device 3B notifying that an answer cannot be given.

[0122] In S14b, the criteria for determining whether or not to generate an answer may be set appropriately. For example, the answer possibility determination unit 206B may determine to generate an answer when it is determined in S13b that the user satisfies the conditions and the remuneration amount is equal to or greater than a lower limit set by the user. The presentation unit 203B may present both or one of the received question and the conditions to the user, and allow the user to select whether or not to answer. In this case, the answer possibility determination unit 206B may determine to generate an answer when it is determined in S13b that the user satisfies the conditions and the user selects that an answer is possible.

[0123] In S15b, the response unit 202B generates an answer to the question received in S11b using the proxy answer language model 4B. Then, in S16b, the proofreading unit 211B judges whether the answer generated in S15b has transmittable content or not. In this way, the proofreading unit 211B judges whether the generated answer has transmittable content or not, in other words, whether the answer does not include content that should not be transmitted. For example, a list of information that should not be transmitted may be stored in the storage unit 11A or the like. In this case, the proofreading unit 211B judges that the generated answer cannot be transmitted if the generated answer includes information included in the list.

[0124] If the answer is YES in S16b, the process proceeds to S17b. In S17b, the presenter 203B transmits the answer generated in S15b to the information processing device 3B. This ends the process of the flow F1b.

[0125] On the other hand, if the result of the determination in S16b is NO, the process returns to S15b, and the response unit 202B uses the proxy response language model 4B to generate a response to the question received in S11b. In general, the output of a language model changes probabilistically. Therefore, even if the same query is input to the proxy response language model 4B, the output response may be different. For this reason, in S15b, which is a transition from S16b, the response unit 202B may input the same question as a query to the proxy response language model 4B. In addition, the response unit 202B may generate a query that instructs the proxy response language model 4B to generate a response that does not include the information that should not be transmitted that is detected by the proofreading unit 211B, and input the query to the proxy response language model 4B. This makes it possible to cause the proxy response language model 4B to generate a response that is more likely to be determined as transmittable in S16b.

[0126] As described above, the language model used by the information processing device 1B may be the proxy answer language model 4B that has been machine-learned to generate an answer to an input question on behalf of a specific user. The receiving unit 201B receives an input of a question for the specific user. The information processing device 1B includes an answer possibility determination unit 206B that determines whether or not to cause the proxy answer language model 4B to generate an answer to the question. This provides the effect of not causing the proxy answer language model 4B to generate an answer as necessary, in addition to the effect provided by the information processing device 1.

[0127] As described above, the answer possibility determination unit 206B may determine to generate an answer to a question in the proxy answer language model 4B corresponding to a specific user when the specific user satisfies the condition associated with the question. This provides an effect that an answer that meets the condition associated with the question can be generated, in addition to the effect provided by the information processing device 1.

[0128] As described above, the information processing device 3B includes a question classification unit 303B that classifies questions included in multiple requests received by the reception unit 301B according to the target person. The negotiation unit 302B then negotiates with the negotiation partner corresponding to the same classification for questions that are classified into the same classification. This provides the effect of collecting questions by target person and efficiently negotiating, in addition to the effect provided by the information processing device 3.

[0129] (Question and Answer Session) In the investigation system 7B, a question and answer session regarding questions and conditions can be conducted in the negotiation between the information processing devices 1B and 3B. This will be described with reference to FIG. 13. FIG. 13 is a diagram showing a flow of a question and answer session conducted between the information processing devices 1B and 3B. FIG. 13 shows a flow F3b executed by the information processing device 3B and a flow F1c executed by the information processing device 1B.

[0130] In S31b of flow F3b, negotiation unit 302B of information processing device 3B transmits a question and a condition corresponding to the question to information processing device 1B. Note that before the process of S31b, the processes of S31a to S33a in FIG.

[0131] In S32b, the reception unit 301B receives an inquiry from the information processing device 1B. This inquiry is an inquiry about at least one of the question and the condition transmitted in S31b, and is transmitted by the process of S14c described later.

[0132] In S33b, the negotiation unit 302B generates a response to the inquiry received in S32b. For example, the negotiation unit 302B may use the negotiation language model 5B to generate a response to the inquiry received in S32b. Specifically, the negotiation unit 302B generates a response to the inquiry (for example, "The remuneration cannot be increased" or "The remuneration can be increased if you can answer additional questions") by inputting the sentence of the inquiry received in S32b (for example, "Can the remuneration be increased?", etc.) into the negotiation language model 5B. In addition, for example, the negotiation unit 302B may notify the requester of the investigation of the received inquiry. In this case, the negotiation unit 302B may use the response from the requester of the investigation to the notification as the response to the inquiry received in S32b.

[0133] In S34b, the negotiation unit 302B transmits the answer generated in S33b to the information processing device 1B. In the following S35b, the negotiation unit 302B judges whether or not to end the question and answer session. If it is judged in S35b that the question and answer session should be ended (YES in S35b), the processing of the flow F3b ends, and if it is judged that the question and answer session should be continued (NO in S35b), the processing returns to S32b. The criterion for judging whether or not to end the question and answer session may be set appropriately. For example, the negotiation unit 302B may judge to end the question and answer session if it cannot confirm receipt of a re-inquiry within a predetermined time after transmitting the answer in S34b.

[0134] On the other hand, in S11c in flow F1c, the receiving unit 201B of the information processing device 1B receives the question and the condition transmitted by the information processing device 3B in S31b. As in the flow F1b in Fig. 12, after S11c, authentication may be performed by the authentication unit 204B to determine whether the source of the question and the condition is valid. Also, the following process may be performed after S13b in flow F1b (determining whether the user satisfies the condition).

[0135] In S12c, the negotiation unit 208B judges whether or not to make an inquiry regarding the question and the condition received in S11c. If the judgment in S12c is YES, the process proceeds to S13c, and if the judgment in S12c is NO, the process of the flow F1c ends.

[0136] The criteria for determining whether to make an inquiry in S12c may be set appropriately. For example, the negotiation unit 208B may generate a query asking whether there is an inquiry regarding the question and conditions received in S11c, input the query to the proxy answer language model 4B, and determine whether to make an inquiry based on the output answer. For example, if the answer output by the proxy answer language model 4B is something like "Both the question and the conditions are clear, so no inquiry is necessary," the negotiation unit 208B may determine not to make an inquiry. On the other hand, if the answer output by the proxy answer language model 4B is something like "Since the reward is not specified, we recommend making an inquiry," the negotiation unit 208B may determine to make an inquiry.

[0137] Also, for example, the negotiation unit 208B may determine to make an inquiry when the amount of the reward is less than a predetermined threshold value or when the user does not satisfy some of the conditions. In addition, for example, the presentation unit 203B may present the received question and conditions to the user and allow the user to select whether or not to make an inquiry. In this case, the negotiation unit 208B determines to make an inquiry when the user selects to make an inquiry.

[0138] In S13c, the negotiation unit 208B generates an inquiry. In addition, in S14c, the negotiation unit 208B transmits the generated inquiry to the information processing device 3B. The method of generating the inquiry may be determined in advance. For example, the negotiation unit 208B may cause the proxy answer language model 4B to generate an inquiry. In this case, the negotiation unit 208B may generate a query that instructs the generation of an inquiry regarding the question and the condition received in S11c and input it to the proxy answer language model 4B. In addition, for example, fixed inquiry sentences such as "Please tell me more detailed information about the condition" and "Is it possible to increase the remuneration amount" may be prepared in advance. In this case, the negotiation unit 208B may select an inquiry sentence according to the question and the condition received in S11c.

[0139] In S15c, the negotiation unit 208B receives the response transmitted by the information processing device 3B in S34b. Next, in S16c, the negotiation unit 208B judges whether or not to make a re-inquiry for the response received in S15c. As in S12c, the criterion for judging whether or not to make a re-inquiry may be set appropriately. If it is judged as YES in S16c, the process returns to S13c, and if it is judged as NO in S16c, the process of the flow F1c ends. After the process of the flow F1c ends, the process may proceed to the process of S14b in the flow F1b of FIG. 12, for example.

[0140] (Calculating confidence, generating alternative conditions, accepting modifications by users) The reliability determination unit 209B of the information processing device 1B determines the reliability of the answer generated by the proxy answer language model 4B. The reliability is an index showing the reliability of the answer. Furthermore, the alternative condition generation unit 207B generates an alternative condition to replace the condition associated with the question to which an answer is sought. Furthermore, the correction unit 210B accepts a user correction to the answer generated by the proxy answer language model 4B. These processes will be described with reference to FIG. 14. FIG. 14 is a flow diagram showing an example of the process executed by the information processing device 1B.

[0141] In S11d of flow F1d shown in Fig. 14, the reception unit 201B receives the question and the condition transmitted by the information processing device 3B. As in the flow F1b of Fig. 12, after S11d, authentication may be performed by the authentication unit 204B to determine whether the sender of the question and the condition is valid. Also, the following process may be performed after S13b in flow F1b (determining whether the user satisfies the condition).

[0142] In S12d, the answer acceptability determination unit 206B determines whether or not to generate an answer to the question in the proxy answer language model 4B. If the determination in S12d is YES, the process proceeds to S13d, and if the determination in S12d is NO, the process proceeds to S18d.

[0143] In S13d, the response unit 202B generates an answer to the question received in S11d using the proxy answer language model 4B. Then, in S14d, the reliability determination unit 209B determines the reliability of the answer generated in S13d.

[0144] The method of determining the reliability is not particularly limited. For example, the reliability determination unit 209B may determine the reliability of an answer to be high when an answer generated using the proxy answer language model 4B of a certain user includes the user's business-related data or matches an answer previously input by the user. On the other hand, when the answer does not include the user's business-related data or there is a discrepancy between the answer and an answer previously input by the user, the reliability of the answer may be determined to be low. The reliability determination unit 209B may determine which of a plurality of stages such as high, medium, or low the reliability corresponds to, or may calculate a numerical value indicating the reliability.

[0145] In S15d, the presenting unit 203B presents the answer generated in S13d to the user. At this time, the presenting unit 203B may also present the reliability determined in S14d to the user.

[0146] In S16d, the correction unit 210B accepts a correction by the user to the answer presented in S15d. Here, the reliability determination unit 209B may update the reliability of the answer corrected by the user and the answer not corrected after the user confirmed the validity of the content. For example, the reliability determination unit 209B may update the reliability of the answer as described above by increasing the reliability by one step or a predetermined value, or may update the reliability to its maximum value.

[0147] In S17d, the presenting unit 203B transmits the answer generated in S13d or the answer modified in S16d to the information processing device 3B. This ends the processing of flow F1d. Note that in S17d, the presenting unit 203B may notify the information processing device 3B of the reliability of the answer. Alternatively, the processing of S15d and S16d may be omitted, and in S17d, the presenting unit 203B may not transmit an answer whose reliability determined in S14d is less than a threshold value, or may transmit an answer indicating that the question cannot be answered.

[0148] Here, as described above, if the determination in S12d is NO, the process proceeds to S18d. In S18d, the alternative condition generating unit 207B generates an alternative condition to replace the condition received in S11d.

[0149] The method of generating the alternative condition may be determined in advance. For example, the alternative condition generating unit 207B may cause the proxy answer language model 4B to generate the alternative condition. In this case, the alternative condition generating unit 207B may generate a query for the question and condition received in S11d, instructing the generation of the alternative condition, and input the query to the proxy answer language model 4B. Also, for a question for which confirmation of the answer by the user is not a mandatory condition, a generation rule for the alternative condition may be set in advance, such as asking the user to confirm the answer, instead of asking the user to confirm the answer, to increase the presented reward by 10%. In this case, the alternative condition generating unit 207B can generate the alternative condition according to the set generation rule.

[0150] In S19d, the presentation unit 203B presents the alternative conditions generated in S18d to the user. Then, in S20d, the alternative condition generation unit 207B accepts the user's modification of the alternative conditions. Note that in S18d, the user may be prompted to input the alternative conditions. In this case, the processes of S19d and S20d are omitted.

[0151] In S21d, the negotiation unit 208B notifies the information processing device 3B of the alternative conditions generated in S18d or the alternative conditions modified in S20d, and inquires whether or not to accept the alternative conditions. In the information processing device 3B that has received the alternative conditions, the negotiation unit 302B determines whether or not to accept the alternative conditions, and notifies the information processing device 1B of the determination result.

[0152] It should be noted that the method of determining whether or not to accept the alternative conditions may be determined in advance. For example, the negotiation unit 302B may generate a query asking whether or not the received alternative conditions should be accepted, and input the query to the negotiation language model 5B, and based on the answer obtained, determine whether or not to accept the alternative conditions. Also, for example, an acceptable range may be determined in advance for the conditions such as remuneration. In this case, the negotiation unit 302B may determine that the alternative conditions are accepted if they are within an acceptable range, and may determine that the alternative conditions are not accepted if they are beyond the acceptable range. Also, the negotiation unit 302B may present the alternative conditions to the requester of the investigation, and allow the requester to select whether or not to accept the alternative conditions.

[0153] In S22d, the negotiation unit 208B judges whether the alternative condition is accepted. If the answer is YES in S22d, the process proceeds to S13d, and if the answer is NO in S22d, the process of the flow F1d ends. If the answer is NO in S22d, the process may return to S18d and new alternative conditions may be generated.

[0154] As described above, the information processing device 1B includes the alternative condition generating unit 207B that generates alternative conditions to replace the conditions associated with a question, and the negotiation unit 208B that notifies the sender of the question of the alternative conditions and asks whether or not to accept the alternative conditions. This provides the effect of enabling flexible negotiation while changing conditions, in addition to the effect provided by the information processing device 1.

[0155] As described above, the information processing device 1B may include the identification unit 101 or 101A and the adjustment unit 102. In this case, the identification unit 101 or 101A may identify a combination of the answer corrected in S16d and a question corresponding to the answer as business-related data. Then, the adjustment unit 102 may re-learn the proxy answer language model 4B using the business-related data as training data. Alternatively, instead of re-learning, the adjustment unit 102 may register the business-related data as data to be referred to when generating an answer using the proxy answer language model 4B. This can improve the accuracy of answers using the proxy answer language model 4B.

[0156] (Generation of alternative conditions on the information processing device 3B side) The generation of the substitution conditions can also be performed on the information processing device 3B side. The generation of the substitution conditions by the information processing device 3B will be described with reference to Fig. 15. Fig. 15 is a flow chart showing an example of processing executed by the information processing device 3B.

[0157] In S31c of flow F3c shown in Fig. 15, the negotiation unit 302B receives a response notifying that the question cannot be answered from the information processing device 1B. This response is transmitted in S17a when NO is determined in S14a in flow F1b of Fig. 12.

[0158] In S32c, the alternative condition generating unit 305B generates an alternative condition to replace the condition previously notified to the information processing device 1B. The method of generating the alternative condition may be determined in advance. For example, the alternative condition generating unit 305B may cause the negotiation language model 5B to generate an alternative condition. In this case, the alternative condition generating unit 305B may generate a query for the question and condition previously transmitted, instructing the generation of an alternative condition, and input the query to the negotiation language model 5B. Also, for example, when consent is not obtained for answering a question for which confirmation by the user is not essential, a rule for generating an alternative condition may be set in advance, such as having the user confirm the answer in exchange for increasing the reward by 5%. In this case, the alternative condition generating unit 305B can generate the alternative condition according to the set generation rule. The alternative condition generating unit 305B may notify the requester of the survey of the generated alternative condition and inquire whether or not the alternative condition is acceptable. Also, the range of conditions (for example, the range of the reward amount) that the requester accepts may be set in advance. In this case, the alternative condition generating unit 305B generates alternative conditions within the specified range. The alternative condition generating unit 305B may also notify the requester that the answer to the question was not accepted, and request the requester to set alternative conditions.

[0159] In S33c, the negotiation unit 302B transmits the alternative conditions generated in S32c to the information processing device 1B, which is the negotiation partner, and inquires whether or not to accept the alternative conditions. In the information processing device 1B that has received the alternative conditions, the negotiation unit 208B determines whether or not to accept the alternative conditions, and notifies the information processing device 3B of the determination result.

[0160] It should be noted that the method of determining whether or not to accept the alternative condition may be determined in advance. For example, the negotiation unit 208B may determine whether or not to accept the alternative condition based on a response obtained by generating a query asking whether or not to accept the received alternative condition and inputting the query into the proxy response language model 4B. Also, for example, an acceptable range may be determined in advance for the condition such as remuneration. In this case, the negotiation unit 208B may determine that the alternative condition is accepted if it is within an acceptable range, and may determine that the alternative condition is not accepted if it is beyond the acceptable range. Also, the negotiation unit 208B may present the alternative condition to the user of the information processing device 1B and allow the user to select whether or not to accept the alternative condition.

[0161] In S34c, the negotiation unit 302B judges whether the alternative conditions have been accepted. If the answer is YES in S34c, the process proceeds to S35c, and if the answer is NO in S34c, the process of flow F3c ends. If the answer is NO in S34c, the negotiation unit 302B may return to S32c to generate new alternative conditions. If the answer is NO in S34c, the negotiation unit 302B may request the information processing device 1B, which is the negotiation target, to present alternative conditions. In this case, the information processing device 1B, which is the negotiation target, performs the processes of S18d and the following steps in FIG. 14.

[0162] In S35c, the negotiation unit 302B requests the information processing device 1B, which is the negotiation partner, to answer the question. This ends the processing of the flow F3c. After this, the processing of S35a and subsequent steps in the flow F3a of FIG. 12 is performed.

[0163] As described above, the information processing device 3B includes an alternative condition generating unit 305B that generates alternative conditions for the answer to a question when the answer to the question is not accepted by the negotiation partner. The negotiation unit 302B then notifies the negotiation partner of the alternative conditions and asks whether the alternative conditions are accepted. This provides an effect that, in addition to the effect provided by the information processing device 3, it is possible to flexibly negotiate while changing the conditions.

[0164] (Additional questions accepted) The question adding unit 307B of the information processing device 3B accepts an additional question from the survey requester regarding the answer generated by the information processing device 1B using the proxy answer language model 4B. The acceptance of the additional question will be described with reference to Fig. 16. Fig. 16 is a flow diagram showing an example of a process in which the information processing device 3B accepts the additional question.

[0165] In S31d of flow F3d shown in Fig. 16, the negotiation unit 302B receives an answer to the question from the information processing device 1B. Here, the negotiation unit 302B may receive the reliability of the answer from the information processing device 1B together with the answer. The reliability is determined in S14d of Fig. 15. Note that before the process of S31d, the processes of S31a to S34a in flow F3a of Fig. 12 are performed.

[0166] In S32d, the question adding unit 307B notifies the requester of the answer received in S32d. Here, the negotiation unit 302B may also notify the reliability of the answer. In addition, in S32d, the question adding unit 307B may send a message to the requester to prompt the requester to send an additional question about the notified answer. If the condition associated with the question specifies whether additional questions can be asked or the number of times additional questions can be accepted, the question adding unit 307B sends a message according to the specification. For example, if the number of times additional questions can be asked is specified, the question adding unit 307B may also include the number of times in the message. If the condition specifies that additional questions should not be accepted, the question adding unit 307B may include that effect in the message. In this case, NO is determined in S33d described below, and the processing of flow F3d ends.

[0167] In S33d, the question adding unit 307B judges whether or not an additional question has been received from the request source. If the result of S33d is NO, the process of flow F3d ends. After the process of flow F3d ends, the process of S36a and subsequent steps in the flow F3a of FIG. 12 is performed.

[0168] In S34d, the question adding unit 307B transmits the additional question received from the request source to the information processing device 1B, and causes the proxy answer language model 4B to generate an answer to the additional question. After that, the process returns to S31d, and the negotiation unit 302B receives the answer to the additional question.

[0169] As described above, the information processing device 3B includes a question adding unit 307B that accepts additional questions from the survey requester regarding the answers generated by the proxy answer language model 4B and causes the proxy answer language model 4B to generate answers to the additional questions. This provides the effect of enabling the requester to clarify unclear points regarding the answers and ask more in-depth additional questions, in addition to the effect provided by the information processing device 3.

[0170] (Evaluating and renegotiating responses) The answer evaluation unit 306B of the information processing device 3B evaluates the answer generated by the information processing device 1B using the proxy answer language model 4B. Then, the negotiation unit 302B renegotiates the conditions according to the evaluation result of the answer evaluation unit 306B. The evaluation and renegotiation of the answer will be described below with reference to Fig. 17. Fig. 17 is a flow diagram showing an example of a process in which the information processing device 3B evaluates and renegotiates the answer.

[0171] In S31e of the flow F3e shown in Fig. 17, the negotiation unit 302B receives a response to the question from the information processing device 1B. Note that before the process of S31e, the processes of S31a to S34a in the flow F3a of Fig. 12 are performed.

[0172] In S32e, the answer evaluation unit 306B evaluates the answer received in S31e. The answer evaluation method may be determined in advance. For example, the answer evaluation unit 306B may evaluate the received answer based on whether or not the answer satisfies a condition associated with the question. As a specific example, if the condition specifies that each answer is to be confirmed by the user and corrected as necessary, but the received answer does not indicate that the user has confirmed it, the answer evaluation unit 306B evaluates the answer as low. The evaluation may be performed by determining which of a plurality of stages such as high, medium, or low the answer falls into. The answer evaluation unit 306B may also calculate an index value indicating the goodness or badness of the answer. The answer evaluation unit 306B may also evaluate the answer based on the reliability notified from the information processing device 1B. For example, the answer evaluation unit 306B may use the average reliability of each answer as the evaluation value of those answers.

[0173] In S33e, the negotiation unit 302B judges whether or not to negotiate on the remuneration based on the evaluation result of S32e. For example, the negotiation unit 302B may judge not to negotiate if the evaluation result of S32e is within a predetermined allowable range, and may judge to negotiate if it is outside the allowable range. The negotiation unit 302B may also negotiate on conditions other than the remuneration (for example, whether additional questions can be asked and the allowable number of times, whether the user needs to confirm and correct the answer, etc.). If the judgment in S33e is YES, the process proceeds to S34e, where the conditions for the answer to the question are renegotiated. On the other hand, if the judgment in S33e is NO, the process proceeds to S37e without renegotiating.

[0174] In S34e, the negotiation unit 302B causes the alternative condition generating unit 305B to reset the remuneration amount. The method of resetting the remuneration amount may be predetermined. For example, the alternative condition generating unit 305B may reset the remuneration amount so that it is within a range of a predetermined standard remuneration amount. Also, for example, the alternative condition generating unit 305B may cause the negotiation language model 5B to reset the remuneration amount. In this case, the alternative condition generating unit 305B may generate a query that instructs to reset the remuneration amount based on the question and conditions previously transmitted and the evaluation result in S32e, and input the query to the negotiation language model 5B. Also, for example, the relationship between the evaluation result in S32e and the reduction rate of the remuneration may be modeled. In this case, the alternative condition generating unit 305B can reset the remuneration amount using the model. In addition to this, for example, the alternative condition generating unit 305B may reset the remuneration amount by taking into account the past response history and reliability of the information processing device 1B.

[0175] In S35e, the negotiation unit 302B notifies the information processing device 1B of the remuneration amount reset in S34e. In the information processing device 1B that has received the notification of the remuneration amount, the negotiation unit 208B determines whether or not to accept the remuneration amount, and notifies the information processing device 3B of the determination result.

[0176] The method of determining whether or not to accept the remuneration amount may be determined in advance. For example, the negotiation unit 208B may determine whether or not to accept the remuneration amount based on a response obtained by generating a query asking whether or not the notified remuneration amount should be accepted and inputting the query into the proxy response language model 4B. Also, for example, an acceptable range may be determined in advance for the conditions of the remuneration, etc. In this case, the negotiation unit 208B may determine that the notified remuneration amount is accepted if it is within an acceptable range, and may determine that the notified remuneration amount is not accepted if it is beyond the acceptable range. Also, the negotiation unit 208B may present the notified remuneration amount to the user of the information processing device 1B and allow the user to select whether or not to accept the remuneration amount.

[0177] In S36e, the negotiation unit 302B judges whether the re-set remuneration amount has been accepted. If the judgment in S36e is NO, the process returns to S33e, and if the judgment in S36e is YES, the process proceeds to S37e. In S37e, the negotiation unit 302B notifies the requester of the determined remuneration amount, and the process of flow F3e ends. After this, the requester provides the determined remuneration amount to the user of the information processing device 1B.

[0178] As described above, the information processing device 3B includes the answer evaluation unit 306B that evaluates the answer generated by the proxy answer language model 4B, and the negotiation unit 302B renegotiates the conditions of the answer to the question based on the result of the evaluation. This provides the effect of being able to automatically review the conditions depending on the quality of the generated answer, in addition to the effect provided by the information processing device 3.

[0179] [Modifications] The executing entity of each process described in each of the above exemplary embodiments is arbitrary and is not limited to the above examples. In other words, the functions of the information processing devices 1, 1A, 1B, 2, 3, and 3B can be realized by a plurality of devices (which can also be called processors) that can communicate with each other. For example, each process described in the flow charts of Figs. 2, 4, 7, 8, and 12 to 17 can be shared and executed by a plurality of processors. In other words, the executing entity of each process in each of the above exemplary embodiments may be one processor or a plurality of processors.

[0180] [Software implementation example] Some or all of the functions of the information processing devices 1, 1A, 1B, 2, 3, 3B (hereinafter also referred to as "the above devices") may be realized by hardware such as an integrated circuit (IC chip), or by software.

[0181] In the latter case, each of the above devices is realized, for example, by a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter, referred to as computer C) is shown in Fig. 18. Fig. 18 is a block diagram showing the hardware configuration of computer C that functions as each of the above devices.

[0182] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores programs (adjustment program / response program / intermediation program) P for operating the computer C as each of the above devices. In the computer C, the processor C1 reads and executes the programs P from the memory C2 to realize the functions of each of the above devices.

[0183] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0184] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may further include a communication interface for transmitting and receiving data to and from other devices. The computer C may further include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0185] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can obtain the program P via such a recording medium M. Furthermore, the program P can be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also obtain the program P via such a transmission medium.

[0186] [Additional Note 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.

[0187] (Appendix A1) An information processing device comprising: an identification means for identifying data related to a specific user's business as business-related data; and an adjustment means for adjusting a language model that has been machine-learned to output answers to queries for the specific user using the business-related data.

[0188] (Appendix A2) The information processing device described in Appendix A1, wherein the identification means uses hierarchical information indicating a hierarchical structure of an organization to which the specified user belongs to identify a department of a hierarchy having a specified relationship with the hierarchy to which the specified user's department belongs, and identifies the business-related data related to the identified department.

[0189] (Appendix A3) The information processing device according to claim 1, wherein the identifying means identifies the business-related data according to a confidentiality level set for each piece of data related to the business of the predetermined user.

[0190] (Appendix A4) An information processing device comprising: a reception means for receiving a query input from a specified user; and a response means for generating an answer to the query using a language model that is machine-learned to output an answer to the query, the response means generating an answer to the query received by the reception means using business-related data related to the business of the specified user and a language model adjusted for the specified user, or generating an answer to the query received by the reception means using the business-related data and the language model.

[0191] (Appendix A5) The information processing device described in Appendix A4, wherein the language model is a proxy answer language model that has been machine-learned to be able to generate an answer on behalf of the specified user to an input question, and the receiving means includes an answer feasibility determination means that receives an input of a question for the specified user and determines whether or not to cause the proxy answer language model to generate an answer to the question.

[0192] (Appendix A6) The information processing device described in Appendix A5, wherein the answer feasibility determination means determines that an answer to the question should be generated in the proxy answer language model corresponding to the specified user when the specified user satisfies a condition associated with the question.

[0193] (Appendix A7) An information processing device as described in Appendix A5 or A6, comprising: an alternative condition generating means for generating alternative conditions to replace the conditions associated with the question; and a negotiation means for notifying a sender of the question of the alternative conditions and inquiring whether or not to accept the alternative conditions.

[0194] (Appendix A8) A survey system including: an information processing device according to any one of appendices A4 to A7; and an intermediary device that accepts a request for a survey to collect answers to a specified question, and negotiates with the information processing device to have the language model generate the answer to the question.

[0195] (Appendix B1) An adjustment method in which at least one processor executes an identification process to identify data related to a specific user's work as work-related data, and an adjustment process to adjust a language model machine-learned to output answers to queries for the specific user using the work-related data.

[0196] (Appendix B2) The adjustment method described in Appendix B1, in which, in the identification process, the at least one processor identifies a department in a hierarchy that has a specified relationship with the hierarchy to which the department to which the specified user belongs, using hierarchical information indicating the hierarchical structure of the organization to which the specified user belongs, and identifies the business-related data related to the identified department.

[0197] (Appendix B3) The adjustment method according to appendix B1 or B2, wherein in the identification process, the at least one processor identifies the business-related data according to a confidentiality level set for each piece of data related to the specified user's business.

[0198] (Appendix B4) A response method in which at least one processor executes a reception process for receiving a query input from a specified user, and a response process for generating an answer to the query received in the reception process using a language model that is machine-learned to output an answer to the query, and that uses business-related data related to the business of the specified user to generate an answer to the query received in the reception process using a language model adjusted for the specified user, or using the business-related data and the language model to generate an answer to the query received in the reception process.

[0199] (Appendix B5) The language model is a proxy answer language model that has been machine-learned to be capable of generating an answer on behalf of the specified user to an input question, and in the reception process, the at least one processor receives an input of a question for the specified user, and executes an answer feasibility determination process in which the at least one processor determines whether or not to cause the proxy answer language model to generate an answer to the question.

[0200] (Appendix B6) The response method described in Appendix B5, wherein, in the answer feasibility determination process, the at least one processor determines that an answer to the question should be generated in the proxy answer language model corresponding to the specified user if the specified user satisfies a condition associated with the question.

[0201] (Appendix B7) The response method described in Appendix B5 or B6, including an alternative condition generation process in which the at least one processor generates alternative conditions to replace the conditions associated with the question, and a negotiation process in which the sender of the question is notified of the alternative conditions and inquires as to whether or not to accept the alternative conditions.

[0202] (Appendix C1) An adjustment program that causes a computer to function as an identification means for identifying data related to a specified user's work as work-related data, and as an adjustment means for adjusting a language model machine-learned to output answers to queries for the specified user using the work-related data.

[0203] (Appendix C2) The adjustment program described in Appendix C1, wherein the identification means uses hierarchical information indicating a hierarchical structure of an organization to which the specified user belongs to identify a department of a hierarchy that has a specified relationship with the hierarchy to which the department to which the specified user belongs, and identifies the business-related data related to the identified department.

[0204] (Appendix C3) The adjustment program according to appendix C1 or C2, wherein the identifying means identifies the business-related data according to a confidentiality level set for each piece of data related to the business of the specified user.

[0205] (Appendix C4) A response program that causes a computer to function as a reception means for receiving a query input from a specified user, and a response means for generating an answer to the query using a language model that has been machine-learned to output an answer to the query, the response means generating an answer to the query received by the reception means using business-related data related to the business of the specified user and a language model adjusted for the specified user, or generating an answer to the query received by the reception means using the business-related data and the language model.

[0206] (Appendix C5) The language model is a proxy answer language model that has been machine-learned to be able to generate an answer on behalf of the specified user to an input question, and the receiving means receives an input of a question for the specified user, and causes the computer to function as answer feasibility determination means that determines whether to cause the proxy answer language model to generate an answer to the question.

[0207] (Appendix C6) The response program described in Appendix C5, wherein the answer feasibility determination means determines that an answer to the question should be generated in the proxy answer language model corresponding to the specified user when the specified user satisfies a condition associated with the question.

[0208] (Appendix C7) The response program described in appendix C5 or C6, which causes the computer to function as an alternative condition generating means for generating alternative conditions to replace the conditions associated with the question, and as a negotiation means for notifying the sender of the question of the alternative conditions and inquiring whether or not to accept the alternative conditions.

[0209] [Additional Note 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.

[0210] (Appendix D1) An information processing device comprising at least one processor, the at least one processor executing an identification process for identifying data related to a specific user's business as business-related data, and an adjustment process for adjusting a language model that has been machine-learned to output answers to queries, for the specific user, using the business-related data.

[0211] (Appendix D2) The information processing device described in Appendix D1, wherein, in the identification process, the at least one processor identifies a hierarchy having a specified relationship with the hierarchy to which the specified user belongs using hierarchical information indicating a hierarchical structure of an organization to which the specified user belongs, and identifies the business-related data related to the identified hierarchy.

[0212] (Appendix D3) The information processing device according to claim 1 or 2, wherein the at least one processor, in the identification process, identifies the business-related data according to a confidentiality level set for each piece of data related to the specified user's business.

[0213] (Appendix D4) An information processing device comprising at least one processor, the at least one processor executing a reception process for receiving a query input from a specified user, and a response process for generating an answer to the query received by the reception process using a language model adjusted for the specified user using business-related data related to the business of the specified user, or generating an answer to the query received by the reception process using the business-related data and the language model, the language model being machine-learned to output an answer to the query.

[0214] (Appendix D5) The information processing device described in Appendix D4, wherein the language model is a proxy answer language model that has been machine-learned to be able to generate an answer on behalf of the specified user to an input question, and the at least one processor, in the reception process, receives an input of a question for the specified user, and executes an answer feasibility determination process that determines whether or not to cause the proxy answer language model to generate an answer to the question.

[0215] (Appendix D6) The information processing device described in Appendix D5, wherein, in the answer feasibility determination process, the at least one processor determines that an answer to the question should be generated in the proxy answer language model corresponding to the specified user when the specified user satisfies a condition associated with the question.

[0216] (Appendix D7) The information processing device described in Appendix D5 or D6, wherein the at least one processor executes an alternative condition generation process that generates alternative conditions to replace the conditions associated with the question, and a negotiation process that notifies the sender of the question of the alternative conditions and inquires whether or not to accept the alternative conditions.

[0217] The information processing device may further include a memory, and the memory may store a program for causing the at least one processor to execute each of the processes.

[0218] (Appendix E1) A non-temporary recording medium having recorded thereon a program that causes a computer to function as an information processing device, the non-temporary recording medium having recorded thereon an adjustment program that causes the computer to execute an identification process that identifies data related to a specific user's business as business-related data, and an adjustment process that uses the business-related data to adjust a language model that has been machine-learned to output answers to queries for the specific user.

[0219] (Appendix E2) A non-transient recording medium having recorded thereon a response program that causes a computer to function as an information processing device, the response program causing the computer to execute a reception process that receives query input from a specified user, and a response process that generates an answer to the query received by the reception process using a language model that is machine-learned to output an answer to the query, and that uses business-related data related to the business of the specified user and is adjusted for the specified user, or generates an answer to the query received by the reception process using the business-related data and the language model. [Explanation of symbols]

[0220] 1. Information processing device 101 Specification part (specification means) 102 Adjustment section (adjustment means) 2. Information processing equipment 201 Reception section (reception means) 202 Response unit (response means) 1A Information processing equipment 101A Specification part (specification means) 202A Response unit (response means) 4A Language Model 1B Information processing equipment 201B Reception section (reception means) 202B Response section (response means) 206B Answerability judgment unit (answerability judgment means) 207B Alternative condition generation unit (alternative condition generation means) 208B Negotiations Section (Means of Negotiations) 3B Information processing device (intermediary device) 4B Language model for proxy answers 5B Negotiation Language Model 7B Survey System

Claims

1. A means for identifying data related to a task of a predetermined user as task-related data; and an adjustment means for adjusting a language model that has been machine-learned to output an answer to a query, for the specified user, using the business-related data.

2. The information processing apparatus according to claim 1 , wherein the adjustment means adjusts the language model for the predetermined user by registering the task-related data as data to be referred to when an answer is generated by the language model.

3. 3. The information processing device according to claim 1, wherein the identification means identifies a hierarchy having a predetermined relationship with the hierarchy to which the specified user belongs using hierarchical information indicating a hierarchical structure of an organization to which the specified user belongs, and identifies the business-related data related to the identified hierarchy.

4. The information processing apparatus according to claim 1 , wherein the identifying unit identifies the business-related data in accordance with a confidentiality level set for each piece of data related to the business of the predetermined user.

5. A reception means for receiving a query input from a predetermined user; An information processing device comprising: a language model trained by machine learning to output an answer to a query; and a response means for generating an answer to a query received by the reception means using a language model adjusted for the specified user using business-related data related to the specified user's business, or for generating an answer to a query received by the reception means using the business-related data and the language model.

6. the language model is a proxy answer language model that has been machine-learned to generate an answer on behalf of the predetermined user to an input question, The receiving means receives an input of a question for the predetermined user, The information processing apparatus according to claim 5 , further comprising an answer acceptability determining means for determining whether or not to cause the proxy answer language model to generate an answer to the question.

7. 7. The information processing device according to claim 6, wherein the answer feasibility determination means determines that an answer to the question is to be generated in the proxy answer language model corresponding to the specified user when the specified user satisfies a condition associated with the question.

8. an alternative condition generating means for generating an alternative condition to replace the condition associated with the question; 8. The information processing apparatus according to claim 6, further comprising: a negotiation means for notifying a sender of the question of the alternative condition and inquiring whether or not the alternative condition is acceptable.

9. An information processing device according to any one of claims 5 to 7; and an intermediary device that receives a request for a survey to collect answers to a predetermined question, and negotiates with the information processing device to have the language model generate the answer to the question.

10. At least one processor An identification process of identifying data related to a business of a predetermined user as business-related data; An adjustment method that executes an adjustment process that adjusts a language model that has been machine-learned to output an answer to a query for the specified user using the business-related data.

11. Computer, Identification means for identifying data related to a business of a predetermined user as business-related data; and An adjustment program that functions as an adjustment means that adjusts a language model that has been machine-learned to output an answer to a query for the specified user using the business-related data.

Images