Information processing system, information processing method, and information processing program

Abstract

To obtain highly accurate natural language sentences tailored to individual users for specific purposes. [Solution] The information processing system includes: an input information acquisition unit that acquires input information including natural language sentences input by a user to obtain natural language sentences based on first knowledge for a specific purpose; a model selection unit that selects a tuning model from one or more large-scale language models, which is a large-scale language model that has been fine-tuned using second knowledge different from the first knowledge for a specific purpose; a related knowledge acquisition unit that acquires related knowledge related to the input information from the first knowledge; and an output information generation unit that generates output information including related knowledge and natural language sentences corresponding to the input information using the tuning model.

Description

Technical Field

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

Background Art

[0002] Patent Document 1 describes a technique of inputting text into a natural language processing model and outputting information from a subsequent processing algorithm (classification model, summarization model, etc.) connected to the subsequent stage of the natural language processing model. Further, it is described that the natural language processing model is learned in two stages: pre-training and fine-tuning according to the subsequent processing algorithm.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Here, in order to obtain highly accurate natural language sentences by a large language model in a specific application such as creating a council response of a local government, the fine-tuning described in Patent Document 1 can be considered. However, there are problems such as the commission of fine-tuning by individual users (for example, for each local government) who want to obtain natural language sentences in a specific application is not cost-effective, and fine-tuning by oneself is difficult without specialized technology. Retrieval Augmented Generation (RAG) that allows a general large language model to refer to the knowledge possessed by individual users in a specific application can also be considered, but there is room for improvement in obtaining highly accurate natural language sentences suitable for individual users in a specific application.

[0005] This disclosure has been made in view of the above-mentioned issues, and one exemplary purpose is to provide a technology for obtaining highly accurate natural language sentences tailored to individual users in specific applications. [Means for solving the problem]

[0006] An information processing system relating to an illustrative aspect of this disclosure includes: an input information acquisition means for acquiring input information including a natural language sentence, which is input by a user to obtain a natural language sentence based on a first knowledge for a specific application; a model selection means for selecting a tuning model from one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application; a related knowledge acquisition means for acquiring related knowledge related to the input information from the first knowledge; and an output information generation means for generating output information including the related knowledge and a natural language sentence corresponding to the input information using the tuning model.

[0007] An example of an information processing method relating to this disclosure includes: an input information acquisition process in which at least one processor acquires input information including a natural language sentence input by a user to obtain a natural language sentence based on first knowledge for a specific application; a model selection process in which the at least one processor selects a tuning model from among one or more large-scale language models, which is a large-scale language model fine-tuned using a second knowledge different from the first knowledge for the specific application; an associated knowledge acquisition process in which the at least one processor acquires associated knowledge related to the input information from the first knowledge; and an output information generation process in which the at least one processor generates output information including the associated knowledge and a natural language sentence corresponding to the input information using the tuning model.

[0008] An illustrative aspect of this disclosure relates to an information processing program, which is a program that causes a computer to function as an information processing system, and causes the computer to function as: an input information acquisition means that acquires input information including natural language sentences input by a user to obtain natural language sentences based on first knowledge for a specific purpose; a model selection means that selects a tuning model from one or more large-scale language models, which is a large-scale language model that has been fine-tuned using second knowledge different from the first knowledge for the specific purpose; a related knowledge acquisition means that acquires related knowledge related to the input information from the first knowledge; and an output information generation means that generates output information including the related knowledge and natural language sentences corresponding to the input information using the tuning model. [Effects of the Invention]

[0009] One exemplary aspect of this disclosure is that it can provide a technology that can obtain highly accurate natural language sentences tailored to individual users in specific applications. [Brief explanation of the drawing]

[0010] [Figure 1] This is a block diagram showing the configuration of the information processing system related to this disclosure. [Figure 2] This is a flowchart showing the flow of the information processing method related to this disclosure. [Figure 3] This is a block diagram showing the configuration of the information processing system related to this disclosure. [Figure 4] This is a block diagram showing the functional configuration of the information processing device related to this disclosure. [Figure 5] This block diagram shows the configuration of the user terminal related to this disclosure. [Figure 6] This is a flowchart showing the flow of the information processing method related to this disclosure. [Figure 7] This figure shows an example of the model selection screen related to this disclosure. [Figure 8] This figure shows an example of the input screen for parliamentary questions related to this disclosure. [Figure 9]This figure shows an example of the output screen for parliamentary responses related to this disclosure. [Figure 10] This is a flowchart showing the flow of the information processing method related to this disclosure. [Figure 11] This diagram schematically illustrates a specific example of the conversion process related to this disclosure. [Figure 12] This is a flowchart showing the flow of the information processing method related to this disclosure. [Figure 13] This figure shows an example of the meeting minutes database management screen related to this disclosure. [Figure 14] This block diagram shows the hardware configuration of the computer that functions as each device related to this disclosure. [Modes for carrying out the invention]

[0011] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining some or all of the technologies (things or methods) employed in each of the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, embodiments obtained by appropriately omitting some of the technologies 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 that can be expected in that exemplary embodiment and do not define the scope of the present invention. That is, embodiments that do not produce the effects mentioned in each of the exemplary embodiments shown below may also be included in the scope of the present invention.

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

[0013] (Configuration of Information Processing System 1) The configuration of information processing system 1 will be described with reference to FIG. 1. FIG. 1 is a block diagram showing the configuration of information processing system 1. As shown in FIG. 1, information processing system 1 includes an input information acquisition unit 11, a model selection unit 12, a related knowledge acquisition unit 13, and an output information generation unit 14. The input information acquisition unit 11 is an example of a configuration that realizes input information acquisition means. The model selection unit 12 is an example of a configuration that realizes model selection means. The related knowledge acquisition unit 13 is an example of a configuration that realizes related knowledge acquisition means. The output information generation unit 14 is an example of a configuration that realizes output information generation means. Note that these units may be arranged in one computer or may be distributed and arranged in a plurality of computers. In other words, information processing system 1 may be constituted by one computer or may be constituted by a plurality of computers.

[0014] The input information acquisition unit 11 acquires input information including a natural language sentence that is input by a user in order to obtain a natural language sentence based on the first knowledge in a specific application. The specific application is a non-general-purpose application for generating a natural language sentence. Examples of the specific application include, but are not limited to, "creating a council response of a local government", "creating an answer to a question in a specific type of organization", etc. The specific type of organization may be, for example, a school, a hospital, a company, etc., but is not limited thereto.

[0015] The first knowledge is knowledge used in a specific application. For example, the first knowledge may be knowledge associated with a user. The user may be an individual or an organization. When the specific application is, for example, "create a council response for a local government", the first knowledge may be the record of past council responses in the local government to which the user belongs. Also, when the specific application is, for example, "create an answer to a question regarding a specific type of organization", the first knowledge may be the record of past council responses in a certain organization of the specific type to which the user belongs. However, the first knowledge is not limited to this.

[0016] Note that "input by the user" may mean that the content of the input information itself is input by the user, or it may mean that a file containing the input information is specified by the user.

[0017] The model selection unit 12 selects a tuning model, which is a large language model fine-tuned using second knowledge different from the first knowledge in a specific application, from among one or more large language models. Each of the one or more large language models is a model available to the information processing system 1. The one or more large language models include one or more tuning models. The tuning model is a model obtained by fine-tuning a general large language model using the second knowledge.

[0018] The second knowledge is knowledge used in a specific application and is different from the first knowledge. For example, the second knowledge may be knowledge associated with another user different from the user. When the specific application is, for example, "create a council response for a local government", the second knowledge may be the record of past council responses in another local government different from the local government to which the user belongs. Also, when the specific application is, for example, "create an answer to a question regarding a specific type of organization", the second knowledge may be the record of past answers in another organization of the specific type different from the organization of the specific type to which the user belongs. However, the second knowledge is not limited to this.

[0019] Furthermore, some or all of one or more large-scale language models may be connected to the information processing system 1 via a network, or may be included in the information processing system 1. If there is only one available large-scale language model, that is, the large-scale language model is a tuning model. In this case, the model selection unit 12 selects the tuning model. If there are multiple available large-scale language models, the multiple large-scale language models include one or more tuning models. The model selection unit 12 may select a tuning model from among the multiple large-scale language models, for example, one specified by user operation, or one of the predetermined tuning models. The model selection unit 12 may also select a tuning model from among the multiple large-scale language models according to the input information.

[0020] The related knowledge acquisition unit 13 acquires related knowledge related to the input information from the first knowledge. For example, the related knowledge acquisition unit 13 may acquire as related knowledge knowledge the knowledge that contains keywords included in the input information from among the multiple knowledge that constitute the first knowledge. Alternatively, the related knowledge acquisition unit 13 may acquire as related knowledge knowledge the knowledge that satisfies predetermined conditions of similarity with the input information from among the multiple knowledge that constitute the first knowledge. However, the method for acquiring related knowledge from the first knowledge is not limited to these. The first knowledge may be stored in a device connected to the information processing system 1 via a network, or it may be stored in a device included in the information processing system 1.

[0021] The output information generation unit 14 generates output information including natural language sentences corresponding to the relevant knowledge and input information using a tuning model. For example, the output information generation unit 14 inputs the relevant knowledge and input information into the tuning model. The output information generation unit 14 also obtains natural language sentences output from the tuning model in response to the input of the relevant knowledge and input information. The output information generation unit 14 may also generate output information including the obtained natural language sentences. For example, the output information generation unit 14 may include other information in the output information in addition to the obtained natural language sentences. This other information may include, but is not limited to, information indicating the referenced relevant knowledge. The output information generation unit 14 may also output the output information to other devices via a network or to an output device.

[0022] (Effects of Information Processing System 1) As described above, the information processing system 1 employs a configuration comprising: an input information acquisition unit 11 that acquires input information including natural language sentences input by the user to obtain natural language sentences based on first knowledge for a specific application; a model selection unit 12 that selects a tuning model from one or more large-scale language models, which is a large-scale language model fine-tuned using second knowledge different from the first knowledge for a specific application; a related knowledge acquisition unit 13 that acquires related knowledge related to the input information from the first knowledge; and an output information generation unit 14 that generates output information including related knowledge and natural language sentences corresponding to the input information using the tuning model. Therefore, according to the information processing system 1, in order to obtain natural language sentences based on first knowledge for a specific application, a tuning model that has already been fine-tuned using second knowledge for the same specific application can be used, and fine-tuning using first knowledge is not required. As a result, the effect is obtained that highly accurate natural language sentences tailored to individual users for specific applications can be obtained.

[0023] (Information processing method S1 flow) The flow of the information processing method S1 will be explained with reference to Figure 2. For example, if the information processing system 1 has at least one processor, that at least one processor executes the information processing method S1. Figure 2 is a flowchart showing the flow of the information processing method S1. As shown in Figure 2, the information processing method S1 includes an input information acquisition process S11, a model selection process S12, a related knowledge acquisition process S13, and an output information generation process S14.

[0024] In the input information acquisition process S11, at least one processor (for example, the input information acquisition unit 11) acquires input information, including natural language sentences, that are input by the user to obtain natural language sentences based on first knowledge for a specific purpose. The details of the input information acquisition process S11 are the same as those of the input information acquisition unit 11, so they will not be repeated.

[0025] In the model selection process S12, at least one processor (for example, the model selection unit 12) selects a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using second knowledge different from first knowledge for a specific application. The details of the model selection process S12 are the same as those of the model selection unit 12, so they will not be repeated.

[0026] In the related knowledge acquisition process S13, at least one processor (for example, the related knowledge acquisition unit 13) acquires related knowledge related to the input information from the first knowledge. The details of the related knowledge acquisition process S13 are the same as those of the related knowledge acquisition unit 13, so they will not be explained again.

[0027] In the output information generation process S14, at least one processor (for example, the output information generation unit 14) generates output information including natural language sentences corresponding to the relevant knowledge and input information using a tuning model. The details of the output information generation process S14 are the same as those of the output information generation unit 14, so they will not be explained again.

[0028] (Effects of information processing method S1) As described above, the information processing method S1 employs a configuration that includes: an input information acquisition process S11 in which at least one processor acquires input information including natural language sentences input by a user to obtain natural language sentences based on first knowledge for a specific application; a model selection process S12 in which at least one processor selects a tuning model from one or more large-scale language models, which is a large-scale language model fine-tuned using second knowledge different from the first knowledge for a specific application; an associated knowledge acquisition process S13 in which at least one processor acquires associated knowledge related to the input information from the first knowledge; and an output information generation process S14 in which at least one processor generates output information including associated knowledge and natural language sentences corresponding to the input information using the tuning model. Therefore, the same effects as the information processing system 1 can be obtained with the information processing method S1.

[0029] [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 function as those described in the above-described exemplary embodiment are denoted by the same reference numerals, and their descriptions are omitted as appropriate. The scope of application of each technology adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology adopted in this exemplary embodiment can also be adopted in other exemplary embodiments included in this disclosure, to the extent that no particular technical problems arise. Furthermore, each technology shown in the drawings referenced to describe this exemplary embodiment can also be adopted in other exemplary embodiments included in this disclosure, to the extent that no particular technical problems arise.

[0030] (Overview of Information Processing System 1A) Information processing system 1A is a system that provides a service for generating natural language sentences for specific purposes. In this exemplary embodiment, the explanation will focus on an example where the specific purpose is to generate responses for a local government assembly. Hereafter, the service for generating assembly responses will also be referred to as the assembly response generation service. The local government to which the user of the assembly response generation service belongs will also be referred to as the first local government. In other words, the user is a person related to the first local government (for example, an employee). Hereafter, the user may also be simply referred to as the first local government.

[0031] Furthermore, the input information submitted to the parliamentary response creation service includes natural language sentences representing parliamentary questions. The output information generated by the parliamentary response creation service also includes natural language sentences representing parliamentary responses. In addition, the parliamentary response creation service references first knowledge to generate parliamentary responses. First knowledge is knowledge about past parliamentary responses in the first local government. Furthermore, a large-scale language model (hereinafter referred to as the tuning model) fine-tuned using second knowledge is used to generate parliamentary responses. Second knowledge is knowledge about past parliamentary responses in a second local government, distinct from the first local government. In other words, the second local government is the local government that provides the second knowledge for generating the tuning model. Note that the number of first and second local governments is not limited to one; there may be multiple. Furthermore, in addition to the parliamentary response creation service, information processing system 1A may also provide a service for generating natural language sentences for other specific purposes.

[0032] (Configuration of Information Processing System 1A) The configuration of the information processing system 1A will be explained with reference to Figure 3. Figure 3 is a block diagram showing the configuration of the information processing system 1A. The information processing system 1A includes an information processing device 10, a user terminal 20, a model storage device 30, meeting minutes databases 40_1, 40_2, 40_3, ..., and a user information storage device 50. When there is no need to specifically distinguish each of the meeting minutes databases 40_1, 40_2, 40_3, ..., they will simply be referred to as meeting minutes database 40. The information processing device 10 is connected to the user terminal 20, the model storage device 30, the meeting minutes database 40, and the user information storage device 50 via a network NW. The network NW may include, but is not limited to, a wireless LAN (Local Area Network), a wired LAN, a WAN (Wide Area Network), a public telephone network, a mobile data communication network, or a combination of some or all of these. The functional configuration of the information processing device 10 and the user terminal 20 will be described later with a different diagram.

[0033] (Model memory 30) The model memory device 30 stores multiple large-scale language models. These multiple large-scale language models include a general-purpose model M0, a tuning model Ma, and a tuning model Mb. The general-purpose model M0 is a general-purpose large-scale language model. Although Figure 3 shows one general-purpose model M0, this does not limit the number of general-purpose models that the model memory device 30 stores; there may be multiple such models. Furthermore, the general-purpose model M0 is not limited to being stored in the model memory device 30; it may also be stored in an external device of the information processing system 1A.

[0034] Tuning Model Ma is a large-scale language model fine-tuned for the purpose of generating parliamentary responses to parliamentary questions. Past meeting minutes from the second local government (an example of second knowledge) were used for the fine-tuning of Tuning Model Ma. Tuning Model Mb is a large-scale language model fine-tuned for the purpose of generating answers to internal questions within the local government. Internal Q&A from the second local government (an example of second knowledge) was used for the fine-tuning of Tuning Model Mb.

[0035] Furthermore, the "second local government" related to past meeting minutes used for fine-tuning tuning model Ma and the "second local government" related to internal Q&A used for fine-tuning tuning model Mb may be the same or different. Also, although Figure 3 shows two tuning models Ma and Mb, this does not limit the number of tuning models stored in the model memory device 30; the number may be 1 or 3 or more. Also, although Figure 3 shows one tuning model Ma and one Mb for each specific use, this does not limit the number of tuning models fine-tuned for a particular use; the number may be 1 or 3 or more. For example, in the use of preparing parliamentary responses, multiple tuning models fine-tuned using meeting minutes from multiple different second local governments may be stored. Also, in the use of preparing parliamentary responses, multiple tuning models fine-tuned using multiple meeting minutes from the same second local government, each covering at least a portion of a different period, may be stored.

[0036] (Meeting minutes database 40) The minutes database 40 stores minutes associated with each first local government that uses the information processing system 1A. These minutes are an example of first knowledge. For example, minutes database 40_1 stores the minutes of the council of local government L1 (an example of the first local government) for fiscal year 2023, fiscal year 2022, etc. Also, for example, minutes database 40_2 stores the minutes of the council of local government L2 (an example of the first local government) for fiscal year 2023, fiscal year 2022, etc. Also, for example, minutes database 40_3 stores the minutes of the council of local government L3 (an example of the first local government) for fiscal year 2023, fiscal year 2022, etc. Although Figure 3 shows three meeting minutes databases 40_1 to 40_3, this does not limit the number of meeting minutes databases 40 included in the information processing system 1A; the number may be 1, 2, or 4 or more.

[0037] (User information storage device 50) The user information storage device 50 stores user information relating to users who use the information processing system 1A. For example, user information U1 indicates information relating to local government L1. Also, for example, user information U2 indicates information relating to local government L2. Also, for example, user information U3 indicates information relating to local government L3. User information U1 to U3 may each include user identification information that identifies the user (for example, information that identifies the first local government). Also, user information U1 to U3 may each include information indicating, for example, the type of contract.

[0038] The type of contract refers to the type of service agreement between the provider of the parliamentary response preparation service and the first local government. The type of contract may, for example, be based on the amount of consideration required to use the parliamentary response preparation service. Examples of such contract types include free membership, first-tier paid membership (paying a first-tier usage fee), and second-tier paid membership (paying a second-tier usage fee higher than the first-tier usage fee), but the types and number of contracts are not limited to these.

[0039] Although Figure 3 shows three user information entries U1 to U3, this does not limit the number of user information entries stored in the user information storage device 50; the number may be 1, 2, or 4 or more. Furthermore, the local governments L1 to L3 (an example of the first local government) that use the information processing system 1A may be the same as or different from the second local government related to the second knowledge used for fine-tuning the tuning model Ma or Mb. In other words, the second local government may also be able to use the information processing system 1A as a user.

[0040] (Information processing device 10) The configuration of the information processing device 10 will be explained with reference to Figure 4. Figure 4 is a block diagram showing the functional configuration of the information processing device 10. As shown in Figure 4, the information processing device 10 comprises a control unit 110, a storage unit 120, and a communication unit 130. The control unit 110 controls all parts of the information processing device 10. The storage unit 120 stores various types of information referenced by the control unit 110. The communication unit 130 communicates with external devices (for example, a user terminal 20, a model storage device 30, a meeting minutes database 40, a user information storage device 50, etc.) via a communication line NW. The communication unit 130 transmits data supplied from the control unit 110 to other devices and supplies data received from other devices to the control unit 110.

[0041] (Functional block included in the control unit 110) The control unit 110 includes, in addition to the input information acquisition unit 11, model selection unit 12, related knowledge acquisition unit 13, and output information generation unit 14 of the information processing system 1, a user information acquisition unit 15, a conversion unit 16, an evaluation acquisition unit 17, and an alert output unit 18. The user information acquisition unit 15 is an example of a configuration that realizes the user information acquisition means. The conversion unit 16 is an example of a configuration that realizes the conversion means. The evaluation acquisition unit 17 is an example of a configuration that realizes the evaluation means. The alert output unit 18 is an example of a configuration that realizes the alert output means.

[0042] The input information acquisition unit 11 is configured in the same manner as in the exemplary embodiment 1, and is configured as follows: The input information acquisition unit 11 acquires input information, including natural language sentences indicating parliamentary questions, which are entered by the user. The input information acquisition unit 11 may also acquire user identification information that identifies the user who enters the input information. The user identification information is information that identifies the user, and for example, includes information that identifies the first local government to which the user belongs.

[0043] The user information acquisition unit 15 acquires user information about the user. Specific examples of user information are as described above. For example, if the input information acquisition unit 11 acquires user identification information along with the input information, the user information acquisition unit 15 may acquire user information including the user identification information from the user information storage device 50.

[0044] The model selection unit 12 is configured in the same manner as in Exemplary Embodiment 1, and is further configured as follows: The model selection unit 12 changes the selectable large-scale language model from among a plurality of large-scale language models according to user information. For example, if the user information includes information indicating the type of contract, the model selection unit 12 may change the selectable large-scale language model according to the type of contract. This allows the user to use an appropriate large-scale language model according to the type of contract.

[0045] For example, if different contract types are offered depending on the amount of consideration paid for using the service, "changing the selectable large-scale language models" could mean that the more consideration paid, the more large-scale language models become available. Alternatively, "changing the selectable large-scale language models" could mean that the more consideration paid, the more large-scale language models that are better suited to specific uses become available. For example, tuning models Ma or Mb are better suited to specific uses than general-purpose model M0. As an example, free members could only select general-purpose model M0, while paid members could select general-purpose model M0, tuning models Ma, or Mb. This would allow users who wish to select a desired model from a wider range of large-scale language models, or from large-scale language models that are better suited to specific uses, to utilize an appropriate large-scale language model according to the type of contract.

[0046] The conversion unit 16 converts the data format of the first knowledge. For example, the conversion unit 16 may convert the first knowledge recorded in natural language sentences into first knowledge consisting of a predetermined set of items. For example, the conversion unit 16 may convert minutes recorded in natural language sentences into minutes consisting of sets of parliamentary questions and parliamentary answers. However, the data formats before and after conversion are not limited to the examples described above. This makes it possible to make the first knowledge into a data format suitable for acquiring related knowledge, which will be described later.

[0047] The output information generation unit 14 is configured in the same manner as in the exemplary embodiment 1, and is configured as follows: The output information generation unit 14 includes information indicating related knowledge in the output information. For example, the information indicating related knowledge to be included in the output information may be an overview such as the title of the related knowledge. This allows the user to know what kind of related knowledge the large-scale language model referenced to create the parliamentary response to the parliamentary question.

[0048] The evaluation unit 17 obtains the user's evaluation of the relevant knowledge included in the output information. For example, the evaluation may, but is not limited to, information indicating the degree of appropriateness as a reference in multiple stages. Also, for example, the evaluation unit 17 may store the obtained evaluation in association with the relevant knowledge. Specifically, the evaluation unit 17 may store the evaluation in the meeting minutes database 40 in association with the knowledge obtained as relevant knowledge from the meeting minutes. This makes the evaluation available for reference in subsequent related knowledge acquisition processes.

[0049] The related knowledge acquisition unit 13 is configured in the same manner as in Exemplary Embodiment 1, and is configured as follows: The related knowledge acquisition unit 13 acquires related knowledge related to the input information from the converted first knowledge. For example, the related knowledge acquisition unit 13 may extract keywords from the input information and acquire meeting minutes containing items similar to those keywords from the converted first knowledge as related knowledge. This makes it easier to acquire related knowledge related to the input information from the first knowledge.

[0050] Furthermore, the related knowledge acquisition unit 13 acquires related knowledge related to new input information based on the first knowledge and evaluation. For example, in order to acquire related knowledge related to new input information, the related knowledge acquisition unit 13 may acquire knowledge associated with evaluations that satisfy predetermined conditions from the minutes database 40. This makes it possible to acquire knowledge from the minutes in which the user's evaluation satisfies predetermined conditions (for example, evaluated as "good") as related knowledge. In addition, this makes it possible to avoid acquiring knowledge from the minutes in which the user's evaluation does not satisfy predetermined conditions (for example, evaluated as "bad") as related knowledge.

[0051] The alert output unit 18 outputs an alert for each piece of knowledge constituting the first knowledge that satisfies a predetermined alert condition. For example, suppose that date and time information is associated with each piece of knowledge constituting the minutes in the minutes database 40. The date and time information may be the date and time the meeting was held, the date and time the knowledge was registered or updated, etc. In this case, the alert condition may be a condition related to the date and time. For example, the alert condition may be that the date and time indicated by the date and time information is more than a predetermined period ago from the present (for example, more than 3 years ago). This makes it possible to output an alert for knowledge that is not appropriate to be referenced in the large-scale language model as related knowledge (for example, old knowledge). Note that the alert condition is not limited to the example described above.

[0052] (Configuration of user terminal 20) Figure 5 is a block diagram showing the configuration of the user terminal 20. As shown in Figure 5, the user terminal 20 comprises a control unit 210, a storage unit 220, a communication unit 230, an input unit 240, and a display unit 250. The control unit 210 controls all parts of the user terminal 20. The storage unit 220 stores various information that the control unit 210 refers to. The communication unit 230 communicates with external devices (e.g., an information processing device 10, etc.) of the user terminal 20 via a communication line NW. The communication unit 230 transmits data supplied from the control unit 210 to other devices and supplies data received from other devices to the control unit 210.

[0053] The input unit 240 is configured to receive input from the user terminal 20, and may include, for example, input devices such as a keyboard, mouse, touch panel, camera, and microphone. The display unit 250 is configured to display the screen output from the user terminal 20, and may include, for example, a display. The input unit 240 and the display unit 250 may also be integrally formed as a touch panel or the like. Furthermore, one or both of the input unit 240 and the display unit 250 are not limited to being built into the user terminal 20, but may also be connected externally via an interface such as USB (Universal Serial Bus).

[0054] The control unit 210 includes a UI (User Interface) unit 21. The UI unit 21 provides a user interface for using a service that generates natural language sentences for a specific purpose. For example, the UI unit 21 receives user operations for using the service and transmits them to the information processing device 10. Also, when the UI unit 21 receives a screen related to the service from the information processing device 10, it displays the received screen on the display unit 250. For example, the UI unit 21 may be implemented by executing an application program for using the service, which is stored in the storage unit 220. The application program may be an application dedicated to the service. Also, if the service is implemented as a web service, the application program may be a general-purpose web browser.

[0055] (Information processing method S1A flow) The information processing system 1A configured as described above executes the information processing method S1A. Figure 6 is a flowchart showing the flow of the information processing method S1A. As shown in Figure 6, the information processing method S1A includes steps S101 to S112.

[0056] In step S101, the UI unit 21 of the user terminal 20 accepts an operation to input user identification information. For example, user identification information may be assigned to the first local government or its related parties in connection with a contract to use the parliamentary response preparation service. As mentioned above, the user identification information includes, for example, information that identifies the first local government. The input information acquisition unit 11 transmits the input user identification information to the information processing device 10.

[0057] In step S102, the input information acquisition unit 11 of the information processing device 10 receives user identification information. The input information acquisition unit 11 also identifies user information that includes the received user identification information from among the user information stored in the user information storage device 50.

[0058] In step S103, the model selection unit 12 changes the selectable model from among multiple large-scale language models according to the user information. The model selection unit 12 also transmits information indicating the selectable model to the user terminal 20.

[0059] For example, as mentioned above, the model selection unit 12 may change the selectable models depending on the type of contract included in the user information. For example, let's consider an example where the type of contract is one of the free member, first paid member, or second paid member mentioned above. In this case, the free member may only be able to select the general-purpose model M0, the first paid member may be able to select either the general-purpose model M0 or the tuning model Ma, and the second paid member may be able to select any of the general-purpose model M0, the tuning model Ma, or Mb.

[0060] In step S104, the UI unit 21 of the user terminal 20 presents the user with selectable models and accepts an operation to specify one of the presented models. The UI unit 21 transmits information indicating the specified model to the information processing device 10.

[0061] Figure 7 shows an example of a model selection screen displayed on the display unit 250 of the user terminal 20 in step S104. As shown in Figure 7, the example screen G1 includes operation objects G11 to G13 corresponding to each selectable model. Operation object G11 accepts an operation to specify the general-purpose model M0. Operation object G12 accepts an operation to specify the tuning model Ma for creating parliamentary responses. Operation object G13 accepts an operation to specify the tuning model Mb for creating internal Q&A. In other words, in the example screen G1, the selectable models are the general-purpose model M0, the tuning model Ma, and the tuning model Mb.

[0062] Step S105 is an example of the model selection process. In step S105, the model selection unit 12 of the information processing device 10 selects a model specified by the user on the user terminal 20 from among multiple large-scale language models stored in the model storage device 30. The information processing device 10 also sends an input screen corresponding to the selected model to the user terminal 20. For example, if tuning model Ma for creating parliamentary responses is selected, an input screen for entering parliamentary questions is sent to the user terminal 20.

[0063] In step S106, the UI unit 21 of the user terminal 20 accepts an operation to input information including natural language text indicating a parliamentary question. The UI unit 21 also transmits the input information to the information processing device 10.

[0064] Figure 8 shows an example of a parliamentary question input screen displayed on the display unit 250 of the user terminal 20 in step S106. As shown in Figure 8, screen example G2 is displayed when a user who is a person related to "City A," which is an example of the first local government, specifies a tuning model Ma for creating a parliamentary response. Screen example G2 includes an input area G21 and an operation object G22. The input area G21 accepts input of natural language sentences indicating a parliamentary question. Here, natural language sentences indicating a parliamentary question, "I would like to ask about the promotion of DX in City A...", and natural language sentences indicating a command, "Create a draft response for...", are entered. The operation object G22 accepts an operation to instruct the creation of a parliamentary response.

[0065] Step S107 is an example of the input information acquisition process. In step S107, the input information acquisition unit 11 of the information processing device 10 acquires the input information entered by the user into the user terminal 20. The acquired input information includes natural language sentences indicating parliamentary questions.

[0066] Step S108 is an example of the related knowledge acquisition process. In step S108, the related knowledge acquisition unit 13 acquires related knowledge related to the parliamentary questions included in the input information from the minutes database 40 corresponding to the first local government related to the user. Here, for example, the minutes stored in the minutes database 40 may be converted minutes whose data format has been converted by the conversion unit 16. A detailed specific example of the conversion process will be described later with a different diagram.

[0067] Furthermore, if an evaluation is associated with the knowledge that constitutes the minutes in the minutes database 40, the related knowledge acquisition unit 13 may acquire related knowledge related to the input information from among the knowledge that constitutes the minutes and whose evaluation satisfies predetermined conditions.

[0068] Step S109 is an example of the output information generation process. In step S109, the output information generation unit 14 generates output information using a selected model by referring to the input information and related knowledge. For example, the output information generation unit 14 inputs input information including a natural language sentence indicating a parliamentary question and related knowledge related to the parliamentary question obtained from the minutes database 40 into a tuning model Ma for creating a parliamentary response. As a result, a natural language sentence indicating the parliamentary response is output from the tuning model Ma. The output information generation unit 14 also generates output information including the natural language sentence indicating the parliamentary response and information indicating the referenced related knowledge. The output information generation unit 14 also sends an output screen containing the output information to the user terminal 20.

[0069] In step S110, the UI unit 21 of the user terminal 20 displays the received output screen on the display unit 250. This presents the user with output information including the generated parliamentary response and the referenced related knowledge.

[0070] Figure 9 shows an example of a parliamentary response output screen displayed on the display unit 250 of the user terminal 20 in step S110. As shown in Figure 9, screen example G3 is output in response to an operation on the operation object G22 in screen example G2. Screen example G3 includes areas G31 and G32, and operation objects G33 and G34. Area G31 includes a natural language sentence representing the generated parliamentary response: "Regarding the promotion of DX, we believe that cooperation with the private sector, such as incorporating the latest technologies, is indispensable..." This allows the user to consider the actual parliamentary response based on the parliamentary response in area G31.

[0071] Furthermore, area G32 includes operation object G33 corresponding to "R5.3 (March 2023) Parliamentary Response," which indicates the referenced related knowledge, and operation object G34 corresponding to "R3.3 (March 2021) Parliamentary Response." Operation objects G33 and G34 may each accept an operation to instruct the display of a detailed screen (not shown) of the corresponding related knowledge. This allows the user to consider that the parliamentary response in area G31 was generated with reference to the related knowledge in area G32, and to examine the actual parliamentary response based on that parliamentary response.

[0072] In step S111, the UI unit 21 accepts an operation to input an evaluation of the related knowledge. For example, the UI unit 21 may accept an operation to input a multi-level evaluation of the related knowledge (for example, a two-level evaluation of "good" or "bad"). Alternatively, for example, the detailed screen of the relevant related knowledge displayed in response to an operation on operation object G33 or G34 in screen example G3 may include an operation object that accepts evaluations. The UI unit 21 transmits information indicating the input evaluation to the information processing device 10.

[0073] In step S112, the evaluation acquisition unit 17 of the information processing device 10 acquires information indicating an evaluation of the relevant knowledge entered by the user terminal 20. The evaluation acquisition unit 17 also associates the acquired evaluation information with the knowledge corresponding to the relevant knowledge and stores it in the minutes database 40. The stored evaluation may be referenced in the relevant knowledge acquisition process of step S108 when the information processing method S1A is executed again in the future.

[0074] (Information processing method S2 flow) The minutes referenced in the related knowledge acquisition process of step S108 described above may be converted minutes with a converted data format. The information processing system 1A may execute an information processing method S2 to convert the data format of the minutes. Figure 10 is a flowchart showing the flow of information processing method S2. As shown in Figure 10, information processing method S2 includes steps S201 to S204.

[0075] In step S201, the UI unit 21 of the user terminal 20 accepts an operation to register meeting minutes. This operation may be an operation to input the meeting minutes themselves, or an operation to specify a file in which the meeting minutes are recorded. The user performing this operation may be the same as, for example, a user who can use the meeting minutes creation service, or a different user. For example, the operation to register meeting minutes may be accepted by inputting user identification information that has been granted the authority to register meeting minutes. The UI unit 21 sends the meeting minutes to the information processing device 10.

[0076] In step S202, the conversion unit 16 of the information processing device 10 acquires the meeting minutes entered on the user terminal 20.

[0077] Step S203 is an example of a conversion process. In step S203, the conversion unit 16 converts the data format of the acquired meeting minutes. For example, the conversion unit 16 may convert the meeting minutes into a data format represented by a predetermined combination of items.

[0078] Figure 11 is a schematic diagram illustrating a specific example of the conversion process. In Figure 11, meeting minutes D1 represents the minutes of the A City Council meeting on June 7, 2024, and is a natural language data format recording the statements of each council member in chronological order. The conversion unit 16 analyzes meeting minutes D1 and converts it into meeting minutes D2, a data format represented by a combination of the following items: date and time, field, council question, and council answer. By referring to the converted meeting minutes D2, for example, if the input council question includes a keyword related to a field such as "childcare," it becomes possible to obtain relevant knowledge from the set of council questions and council answers whose field is "childcare." This has the advantage of improving the appropriateness of the relevant knowledge and reducing the computational cost of searching for relevant knowledge.

[0079] (Information processing method S3 flow) The information processing system 1A may execute an information processing method S3 that outputs an alert for the meeting minutes. For example, the information processing method S3 may be executed in response to an operation in which a user accesses a screen for managing meeting minutes. The user performing this operation may be the same as, for example, a user who can use the meeting minutes creation service, or a different user. For example, the operation to manage meeting minutes may be accepted by inputting user identification information that has been granted the authority to manage meeting minutes. Figure 12 is a flowchart showing the flow of the information processing method S3. As shown in Figure 12, the information processing method S3 includes steps S301 to S303.

[0080] In step S301, the alert output unit 18 of the information processing device 10 identifies knowledge that satisfies the alert conditions from among the knowledge constituting the meeting minutes stored in the meeting minutes database 40. For example, as described above, the alert output unit 18 may identify knowledge whose associated date and time information satisfies a predetermined condition (for example, it is more than a predetermined period before the present).

[0081] In step S302, the alert output unit 18 generates a screen containing the alert. For example, the alert output unit 18 may generate a screen containing a list of knowledge that constitutes the meeting minutes, and add information indicating an alert to the knowledge that satisfies the alert conditions. The alert output unit 18 also sends the screen containing the alert to the user terminal 20.

[0082] In step S303, the UI unit 21 of the user terminal 20 displays a screen including an alert on the display unit 250. Figure 13 shows an example of the minutes database management screen displayed on the display unit 250 of the user terminal 20 in step S303. As shown in Figure 13, example screen G4 includes the name, size, update date (an example of date and time information), and message of the document that constitutes the minutes (an example of the knowledge that constitutes the minutes). The message associated with the document "A City_Minutes_20210111" whose update date is more than 3 years ago (an example of an alert condition) includes the alert "Data older than 3 years is being referenced." The user can refer to the outputted alert and perform management such as deleting documents that are no longer appropriate. As a result, it is possible to mitigate situations in which parliamentary responses are generated by referring to inappropriate minutes.

[0083] (Effects of Information Processing System 1A) As described above, in Information Processing System 1A, the specific use is to generate responses to local government council meetings. The input information includes natural language sentences representing council questions, and the output information includes natural language sentences representing council responses. The first knowledge is knowledge about past council responses in the first local government related to the user, and the second knowledge is knowledge about past council responses in a second local government different from the first local government. Therefore, in addition to the effects of Information Processing System 1, Information Processing System 1A provides the effect that even in the first local government, where fine tuning is difficult, it is possible to obtain highly accurate council responses tailored to the first local government by using a large-scale language model that has been fine-tuned for the purpose of creating council responses.

[0084] Furthermore, information processing system 1A has multiple large-scale language models and is further equipped with a user information acquisition unit 15 that acquires user information about the user. The model selection unit 12 changes the selectable large-scale language model from among the multiple large-scale language models according to the user information. Therefore, with information processing system 1A, in addition to the effects achieved by information processing system 1, it is possible to use a model that is appropriate to the user information from among the finely tuned large-scale language models, and output information that is suitable for the user information can be obtained.

[0085] Furthermore, information processing system 1A is further equipped with a conversion unit 16 that converts the data format of the first knowledge, and the related knowledge acquisition unit 13 acquires related knowledge from the converted first knowledge. As a result, in addition to the effects achieved by information processing system 1, information processing system 1A makes it possible to convert the first knowledge into a data format suitable for acquiring related knowledge, resulting in improved accuracy of the related knowledge to be referenced and a reduction in the computational cost related to the related knowledge acquisition process.

[0086] Furthermore, in the information processing system 1A, the output information generation unit 14 is configured to include information indicating related knowledge in the output information. Therefore, in addition to the effects achieved by the information processing system 1, the information processing system 1A allows the user to recognize what kind of related knowledge was referenced in the large-scale language model to generate natural language sentences for a specific purpose. Moreover, the user can examine the generated natural language sentences while considering such referenced related knowledge.

[0087] Furthermore, the information processing system 1A is further equipped with an evaluation acquisition unit 17 that acquires the user's evaluation of the related knowledge included in the output information, and the related knowledge acquisition unit 13 acquires related knowledge related to new input information based on the first knowledge and evaluation. As a result, with the information processing system 1A, in addition to the effects achieved by the information processing system 1, the related knowledge acquired according to the user's evaluation is referenced in subsequent uses, so it is possible to obtain output information with even higher accuracy that is tailored to each individual user.

[0088] Furthermore, the information processing system 1A is configured to include an alert output unit 18 that outputs an alert for each piece of knowledge constituting the first knowledge that satisfies predetermined alert conditions. Therefore, in addition to the effects of the information processing system 1, the information processing system 1A allows for consideration of whether or not to exclude the knowledge for which an alert has been output, thereby reducing situations in which output information is accessed by referring to inappropriate related knowledge.

[0089] (modified version) In this exemplary embodiment, other uses may be applied instead of creating parliamentary responses or generating answers for internal Q&A as specific uses. Furthermore, other users may be applied instead of local governments as users. Also, the first and second knowledge can be various types of knowledge that can be referenced for specific uses, not limited to meeting minutes or past internal Q&A.

[0090] [Examples of implementation using software] Some or all of the functions of each device constituting information processing systems 1 and 1A (hereinafter also referred to as "the above devices") may be implemented by hardware such as integrated circuits (IC chips) or by software.

[0091] In the latter case, each of the above devices is implemented, for example, by a computer that executes instructions for a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as Computer C) is shown in Figure 14. Figure 14 is a block diagram showing the hardware configuration of Computer C, which functions as each of the above devices.

[0092] Computer C comprises at least one processor C1 and at least one memory C2. Memory C2 stores a program P that causes computer C to operate as each of the above-mentioned devices. In computer C, processor C1 reads program P from memory C2 and executes it, thereby realizing each of the above-mentioned devices.

[0093] For processor C1, for example, a CPU (Central Processing Unit), GPU (Graphic Processing Unit), DSP (Digital Signal Processor), MPU (Micro Processing Unit), FPU (Floating Point Number Processing Unit), PPU (Physics Processing Unit), TPU (Tensor Processing Unit), quantum processor, microcontroller, or a combination thereof can be used. For memory C2, for example, flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), or a combination thereof can be used.

[0094] Computer C may also be equipped with RAM (Random Access Memory) for loading program P at runtime and for temporarily storing various data. Furthermore, computer C may be equipped with communication interfaces for sending and receiving data with other devices. Additionally, computer C may be equipped with input / output interfaces for connecting input / output devices such as keyboards, mice, displays, and printers.

[0095] Furthermore, program P can be recorded on a non-temporary, tangible recording medium M that is readable by computer C. Such a recording medium M could be, for example, tape, disk, card, semiconductor memory, or programmable logic circuitry. Computer C can acquire program P via such a recording medium M. Program P can also be transmitted via a transmission medium. Such a transmission medium could be, for example, a communication network or broadcast waves. Computer C can also acquire program P via such a transmission medium.

[0096] Furthermore, each of the above functions of each of the above devices may be implemented by a single processor in a single computer, by multiple processors in a single computer working together, or by multiple processors in each of multiple computers working together. In addition, the programs for implementing each of the above functions in each of the above devices may be stored in a single memory in a single computer, distributed and stored in multiple memories in a single computer, or distributed and stored in multiple memories in each of multiple computers.

[0097] [Additional Note A] This disclosure includes the technologies described in the following appendices. However, the present invention is not limited to the technologies described in the following appendices, and various modifications are possible within the scope of the claims.

[0098] (Note A1) An input information acquisition means that acquires input information including natural language sentences, which are input by a user to obtain natural language sentences based on first knowledge for a specific purpose, A model selection means for selecting a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application; Related knowledge acquisition means for acquiring related knowledge related to the input information from the first knowledge, Output information generation means that generates output information including natural language sentences corresponding to the related knowledge and input information using the tuning model, An information processing system equipped with the following features.

[0099] (Appendix A2) The aforementioned specific use is for generating responses to local government councils, The aforementioned input information includes natural language sentences indicating parliamentary questions, The output information includes natural language text representing parliamentary responses, The aforementioned first knowledge is knowledge concerning past parliamentary responses in the first local government relating to the user, The second piece of knowledge is knowledge concerning past parliamentary responses in a second municipality, which is different from the first municipality. The information processing system described in Appendix A1.

[0100] (Note A3) There are multiple large-scale language models as described above. The system further comprises a user information acquisition means for acquiring user information relating to the aforementioned user, The model selection means changes the selectable large-scale language model from the plurality of large-scale language models according to the user information. The information processing system described in Appendix A1 or A2.

[0101] (Note A4) The system further comprises a conversion means for converting the data format of the first knowledge, The related knowledge acquisition means acquires the related knowledge from the first knowledge after conversion. The information processing system described in any one of the appendices A1 to A3.

[0102] (Note A5) The output information generation means includes information indicating the related knowledge in the output information. The information processing system described in any one of the appendices A1 through A4.

[0103] (Note A6) The system further comprises evaluation acquisition means for acquiring the user's evaluation of the related knowledge included in the output information, The related knowledge acquisition means acquires related knowledge related to new input information based on the first knowledge and the evaluation. The information processing system described in Appendix A5.

[0104] (Note A7) The system further includes an alert output means that outputs an alert for each piece of knowledge constituting the first knowledge that satisfies a predetermined alert condition. An information processing system described in any one of the appendices A1 through A6.

[0105] [Additional Notes B] This disclosure includes the technologies described in the following appendices. However, the present invention is not limited to the technologies described in the following appendices, and various modifications are possible within the scope of the claims.

[0106] (Note B1) At least one processor performs an input information acquisition process to acquire input information, including natural language sentences, which are input by a user to obtain natural language sentences based on first knowledge for a specific application. The at least one processor performs a model selection process in which it selects a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application. The at least one processor performs a related knowledge acquisition process that acquires related knowledge related to the input information from the first knowledge, The at least one processor performs an output information generation process that generates output information including natural language sentences corresponding to the related knowledge and the input information using the tuning model, An information processing method that includes this.

[0107] (Note B2) The aforementioned specific use is for generating responses to local government councils, The aforementioned input information includes natural language sentences indicating parliamentary questions, The output information includes natural language text representing parliamentary responses, The aforementioned first knowledge is knowledge concerning past parliamentary responses in the first local government relating to the user, The second piece of knowledge is knowledge concerning past parliamentary responses in a second municipality, which is different from the first municipality. The information processing method described in Appendix B1.

[0108] (Note B3) There are multiple large-scale language models as described above. The at least one processor further includes a user information acquisition process that acquires user information relating to the user, In the model selection process, the at least one processor changes the selectable large-scale language model from the plurality of large-scale language models according to the user information. The information processing method described in Appendix B1 or B2.

[0109] (Note B4) The at least one processor further includes a conversion process for converting the data format of the first knowledge, In the related knowledge acquisition process, the at least one processor acquires the related knowledge from the converted first knowledge. The information processing method described in any one of the appendices B1 to B3.

[0110] (Note B5) In the output information generation process, the at least one processor includes information indicating the related knowledge in the output information. The information processing method described in any one of the appendices B1 to B4.

[0111] (Note B6) The at least one processor further includes an evaluation acquisition process that acquires the user's evaluation of the relevant knowledge included in the output information, In the related knowledge acquisition process, the at least one processor acquires related knowledge related to new input information based on the first knowledge and the evaluation. The information processing method described in Appendix B5.

[0112] (Note B7) The at least one processor further includes an alert output process that outputs an alert with respect to each piece of knowledge constituting the first knowledge that satisfies a predetermined alert condition. The information processing method described in any one of the appendices B1 to B6.

[0113] [Additional Note C] This disclosure includes the technologies described in the following appendices. However, the present invention is not limited to the technologies described in the following appendices, and various modifications are possible within the scope of the claims.

[0114] (Note C1) A program that makes a computer function as an information processing system, The aforementioned computer, An input information acquisition means that acquires input information including natural language sentences, which are input by a user to obtain natural language sentences based on first knowledge for a specific purpose, A model selection means for selecting a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application; Related knowledge acquisition means for acquiring related knowledge related to the input information from the first knowledge, Output information generation means that generates output information including natural language sentences corresponding to the related knowledge and input information using the tuning model, An information processing program that functions as such.

[0115] (Note C2) The aforementioned specific use is for generating responses to local government councils, The aforementioned input information includes natural language sentences indicating parliamentary questions, The output information includes natural language text representing parliamentary responses, The aforementioned first knowledge is knowledge concerning past parliamentary responses in the first local government relating to the user, The second piece of knowledge is knowledge concerning past parliamentary responses in a second municipality, which is different from the first municipality. The information processing program described in Appendix C1.

[0116] (Note C3) There are multiple large-scale language models as described above. The aforementioned computer, Furthermore, it functions as a means for acquiring user information about the aforementioned user, The model selection means changes the selectable large-scale language model from the plurality of large-scale language models according to the user information. The information processing program described in Appendix C1 or C2.

[0117] (Note C4) The aforementioned computer, Further functioning as a conversion means for converting the data format of the aforementioned first knowledge, The related knowledge acquisition means acquires the related knowledge from the first knowledge after conversion. An information processing program described in any one of the appendices C1 to C3.

[0118] (Note C5) The output information generation means includes information indicating the related knowledge in the output information. An information processing program described in any one of the appendices C1 to C4.

[0119] (Appendix C6) The aforementioned computer, The output information is further configured to function as an evaluation acquisition means for acquiring the user's evaluation of the related knowledge included in the output information. The related knowledge acquisition means acquires related knowledge related to new input information based on the first knowledge and the evaluation. The information processing program described in Appendix C5.

[0120] (Note C7) The aforementioned computer, This further functions as an alert output means that outputs an alert for each piece of knowledge constituting the first knowledge that satisfies a predetermined alert condition. An information processing program described in any one of the appendices C1 to C6.

[0121] [Additional Note D] This disclosure includes the technologies described in the following appendices. However, the present invention is not limited to the technologies described in the following appendices, and various modifications are possible within the scope of the claims.

[0122] (Note D1) It comprises at least one processor, and the at least one processor is An input information acquisition process that acquires input information, including natural language sentences, which are input by the user to obtain natural language sentences based on first knowledge for a specific purpose, A model selection process that selects a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application described above. A related knowledge acquisition process that acquires related knowledge related to the input information from the first knowledge, An output information generation process that generates output information including natural language sentences corresponding to the related knowledge and the input information using the tuning model, An information processing system that performs [this action].

[0123] The information processing system may also include memory. Furthermore, the memory may store programs that cause at least one processor to execute each of the aforementioned processes.

[0124] (Note D2) The aforementioned specific use is for generating responses to local government councils, The aforementioned input information includes natural language sentences indicating parliamentary questions, The output information includes natural language text representing parliamentary responses, The aforementioned first knowledge is knowledge concerning past parliamentary responses in the first local government relating to the user, The second piece of knowledge is knowledge concerning past parliamentary responses in a second municipality, which is different from the first municipality. The information processing system described in Appendix D1.

[0125] (Note D3) There are multiple large-scale language models as described above. The aforementioned at least one processor, Further, a user information acquisition process is performed to acquire user information about the aforementioned user. In the model selection process, the at least one processor changes the selectable large-scale language model from the plurality of large-scale language models according to the user information. The information processing system described in Appendix D1 or D2.

[0126] (Note D4) The aforementioned at least one processor, Further conversion processes are performed to convert the data format of the aforementioned first knowledge, In the related knowledge acquisition process, the at least one processor acquires the related knowledge from the converted first knowledge. An information processing system described in any one of the appendices D1 to D3.

[0127] (Note D5) In the output information generation process, the at least one processor includes information indicating the related knowledge in the output information. An information processing system described in any one of the appendices D1 to D4.

[0128] (Note D6) The aforementioned at least one processor, Further, an evaluation acquisition process is performed to acquire the user's evaluation of the related knowledge included in the output information. In the related knowledge acquisition process, the at least one processor acquires related knowledge related to new input information based on the first knowledge and the evaluation. The information processing system described in Appendix D5.

[0129] (Note D7) The aforementioned at least one processor, Further, an alert output process is performed to output an alert for each piece of knowledge that constitutes the first knowledge and satisfies predetermined alert conditions. An information processing system described in any one of the appendices D1 to D6.

[0130] [Additional Note E] This disclosure includes the technologies described in the following appendices. However, the present invention is not limited to the technologies described in the following appendices, and various modifications are possible within the scope of the claims.

[0131] (Note E1) A program that makes a computer function as an information processing system, To the aforementioned computer, An input information acquisition process that acquires input information, including natural language sentences, which are input by the user to obtain natural language sentences based on first knowledge for a specific purpose, A model selection process that selects a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application described above. A related knowledge acquisition process that acquires related knowledge related to the input information from the first knowledge, An output information generation process that generates output information including natural language sentences corresponding to the related knowledge and the input information using the tuning model, A non-temporary recording medium that stores an information processing program that executes such a program. [Explanation of symbols]

[0132] 1. 1A Information Processing System 10 Information Processing Devices 11 Input Information Acquisition Unit 12 Model Selection Section 13. Related Knowledge Acquisition Department 14 Output Information Generation Unit 15. User Information Acquisition Unit 16 Conversion section 17 Evaluation Acquisition Department 18. Alert Output Section 20 User Terminals 30 Models of Storage Devices 40 Meeting Minutes Database 50 User information storage device 21 UI section 110, 210 Control Unit 120, 220 storage section 130, 230 Communications Department 240 Input section 250 Display section C1 Processor C2 Memory

Claims

1. An input information acquisition means for acquiring input information including natural language sentences, which are input by a user to obtain natural language sentences based on first knowledge for a specific purpose, A model selection means for selecting a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application, Related knowledge acquisition means for acquiring related knowledge related to the input information from the first knowledge, Output information generation means that generates output information including natural language sentences corresponding to the related knowledge and input information using the tuning model, An information processing system equipped with the following features.

2. The aforementioned specific use is for generating responses to local government councils, The aforementioned input information includes natural language sentences indicating parliamentary questions, The output information includes natural language text representing parliamentary responses, The first piece of knowledge is knowledge relating to past parliamentary responses in the first local government concerning the user, The second piece of knowledge is knowledge concerning past parliamentary responses in a second municipality, which is different from the first municipality. The information processing system according to claim 1.

3. There are multiple large-scale language models as described above. The system further comprises a user information acquisition means for acquiring user information relating to the aforementioned user, The model selection means changes the selectable large-scale language model from the plurality of large-scale language models according to the user information. The information processing system according to claim 1 or 2.

4. The system further comprises a conversion means for converting the data format of the first knowledge, The related knowledge acquisition means acquires the related knowledge from the first knowledge after conversion. The information processing system according to claim 1 or 2.

5. The output information generation means includes information indicating the related knowledge in the output information. The information processing system according to claim 1 or 2.

6. The system further comprises evaluation acquisition means for acquiring the user's evaluation of the related knowledge included in the output information, The related knowledge acquisition means acquires related knowledge related to new input information based on the first knowledge and the evaluation. The information processing system according to claim 5.

7. The system further includes an alert output means that outputs an alert for each piece of knowledge constituting the first knowledge that satisfies a predetermined alert condition. The information processing system according to claim 1 or 2.

8. At least one processor performs an input information acquisition process to acquire input information, including natural language sentences, which are input by a user to obtain natural language sentences based on first knowledge for a specific application. The at least one processor performs a model selection process in which it selects a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application. The at least one processor performs a related knowledge acquisition process that acquires related knowledge related to the input information from the first knowledge, The at least one processor performs an output information generation process that generates output information including natural language sentences corresponding to the related knowledge and the input information using the tuning model, An information processing method that includes this.

9. A program that makes a computer function as an information processing system, The aforementioned computer, An input information acquisition means for acquiring input information including natural language sentences, which are input by a user to obtain natural language sentences based on first knowledge for a specific purpose, A model selection means for selecting a tuning model from among one or more large-scale language models, which is a large-scale language model that has been fine-tuned using a second knowledge different from the first knowledge for the specific application, Related knowledge acquisition means for acquiring related knowledge related to the input information from the first knowledge, Output information generation means that generates output information including natural language sentences corresponding to the related knowledge and input information using the tuning model, An information processing program that functions as such.

Images