Prompt generation device and question and answer support device

Abstract

You can get accurate answers from the generated AI. [Solution] The prompt creation device comprises an input unit for inputting a user's question, a storage unit for storing creation information, a creation unit for creating prompts, a transmission unit for sending the created prompts to a generation AI, and a receiving unit for receiving answers to prompts from the generation AI. The prompt creation information comprises the user's basic information and multiple detailed information items classified into multiple categories. Each detailed information item has a title and a description associated with it. The creation unit creates a prompt (first prompt) that asks for an answer to select the most suitable detailed information based on the question, basic information, and the titles attached to the multiple detailed information items. Based on the question, basic information, and the answer to the first prompt from the generation AI (first answer), the device adds the description information of the selected detailed information to create a prompt that asks for an answer to the question.

Description

Technical Field

[0001] The present invention relates to a technique for obtaining an answer required by a user by using generative AI (Artificial Intelligence) using a large language model (LLM: Large Language models), and particularly relates to a technique for creating a prompt to be input to the LLM.

Background Art

[0002] Conventionally, a question-and-answer system using an LLM has been provided. An LLM is a language model constructed by a large amount of data and deep learning technology, and is a kind of generative AI. In an LLM, when a question called a prompt is input, an answer generated based on the question is output, but often an answer lacking accuracy for the question is output (hallucination occurs).

[0003] In this regard, Patent Document 1 discloses a technique for generating a prompt with reference information added to an input question in order to obtain an accurate answer.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, there is still room for improvement in a method for obtaining an accurate answer from an LLM, for example, a method for optimizing a prompt for obtaining an accurate answer from an LLM.

[0006] The present invention has been made in view of the above circumstances, and an object thereof is to provide a technique for obtaining an accurate answer from generative AI using an LLM.

Means for Solving the Problems

[0007] To achieve the above objective, the information processing apparatus according to the present invention, in one embodiment, A prompt generation device connected to a generation AI via a communication network, An input section for entering questions from the user, A memory unit that stores prompt creation information for creating prompts, A prompt creation unit creates a prompt related to the question based on the aforementioned question and the prompt creation information, A prompt sending unit that sends the created prompt to the generation AI, A response receiving unit that receives a response to a prompt from the aforementioned generating AI, Equipped with, The prompt creation information comprises basic user information and detailed information categorized into multiple categories. Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information is associated with descriptive information indicating the content of the detailed information. The prompt generation unit includes a first prompt generation unit that generates a prompt (hereinafter referred to as the "first prompt") that asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and the plurality of detailed information items. The prompt transmission unit includes a first prompt transmission unit that transmits the first prompt, The response receiving unit includes a first response receiving unit that receives a response to the first prompt (hereinafter referred to as the "first response"), The prompt generation unit includes a second prompt generation unit that generates a prompt (hereinafter referred to as the "second prompt") requesting an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. This is the gist of it.

[0008] To achieve the above objective, the information processing apparatus according to the present invention, in one embodiment, A question-and-answer support device connected to a generation AI via a communication network, A question input section where users can enter their questions, A memory unit that stores prompt creation information for creating prompts, A prompt creation unit creates a prompt based on the aforementioned question and the prompt creation information, A prompt sending unit that sends the created prompt to the generation AI, The response receiving unit receives a response to a prompt (hereinafter referred to as "intermediate response") from the aforementioned generating AI, A response output unit that creates a final response to present to the user based on the aforementioned interim response and outputs the created final response, Equipped with, The prompt creation information comprises basic user information and detailed information categorized into multiple categories. Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information is associated with descriptive information indicating the content of the detailed information. The prompt generation unit includes a first prompt generation unit that generates a prompt (hereinafter referred to as the "first prompt") that asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and the plurality of detailed information items. The prompt transmission unit includes a first prompt transmission unit that transmits the first prompt, The response receiving unit includes a first response receiving unit that receives a response to the first prompt (hereinafter referred to as the "first response"), The prompt generation unit includes a second prompt generation unit that generates a prompt (hereinafter referred to as the "second prompt") requesting an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. The prompt transmission unit includes a second prompt transmission unit that transmits the second prompt, The aforementioned response receiving unit includes a second response receiving unit that receives a response to the second prompt (hereinafter referred to as the "second response"), The answer output unit creates the final answer based on the second answer and outputs the created final answer. This is the gist.

Effect of the Invention

[0009] According to the present invention, an accurate answer can be obtained from a generative AI using an LLM.

Brief Description of the Drawings

[0010] [Figure 1] It is a schematic configuration diagram of an information processing system according to an embodiment of the present invention. [Figure 2] It is a diagram schematically showing data stored in a storage unit of an information processing apparatus according to an embodiment of the present invention. [Figure 3] (a) is a diagram showing in detail the configuration of a prompt creation unit of an information processing apparatus according to an embodiment of the present invention, (b) is a diagram showing in detail the configuration of a prompt transmission unit of an information processing apparatus according to an embodiment of the present invention, and (c) is a diagram showing in detail the configuration of a prompt transmission unit of an information processing apparatus according to an embodiment of the present invention. [Figure 4] It is a sequence diagram showing the operation of an information processing system according to an embodiment of the present invention. [Figure 5] It is a flowchart showing in detail the flow of optimal prompt selection processing in step S30 of FIG. 4. [Figure 6] It is a flowchart showing in detail the flow of optimal answer acquisition processing in step S40 of FIG. 4. [Figure 7] It is a flowchart showing in detail the flow of additional information acquisition processing in step S50 of FIG. 4. [Figure 8] It is a flowchart showing in detail the flow of answer creation processing in step S60 of FIG. 4. [Figure 9] It is a hardware configuration diagram of an information processing apparatus according to an embodiment of the present invention.

Mode for Carrying Out the Invention

[0011] Embodiments of the present invention will be described below with reference to the drawings.

[0012] <Configuration of the Information Processing System> Figure 1 is a schematic diagram of an information processing system 100 according to an embodiment of the present invention. The information processing system 100 is an information processing system in which an information processing device 1 uses a generation AI 2 to assist in creating answers to user questions. Specifically, the information processing device 1 is a server device that reads the user's question intent, creates an appropriate prompt, sends the created prompt to the generation AI 2, receives an answer to the prompt from the generation AI 2, creates a final answer to the user's question based on the received answer, and presents the created final answer to the user (hereinafter referred to as "question and answer support processing"). In other words, the information processing device 1 is a prompt creation device that creates prompts to be input to the generation AI 2, and is also a question and answer support device that uses the generation AI 2 to present a final answer to the user's question to the user.

[0013] In this embodiment, a user refers to a user who can use the services provided by the information processing system 100, and includes both (1) users as an organization (e.g., a company, group, etc.) (hereinafter referred to as "organizational users") and (2) users as individuals (hereinafter referred to as "individual users"). For example, if a company uses the information processing system 100 as one of its internal systems, it corresponds to an organizational user (1), and the user information (user ID that can uniquely identify the user within the information processing system 100, user attribute information, etc.) includes not only user information for the entire organization but also user information for employees belonging to the company. Even organizational users log in to the information processing system 100 using an individual user ID assigned by the organization. On the other hand, if an individual uses the information processing system 100, they are an individual user (2), and the user information consists only of the individual user information.

[0014] As shown in Figure 1, the information processing system 100 comprises an information processing device 1, a generation AI 2, an intermediary system 3, and a user terminal 4. The information processing device 1, the generation AI 2, the intermediary system 3, and the user terminal 4 are each connected to each other so as to be able to communicate via a communication network 5. The communication network 5 is a wired or wireless communication network, such as the Internet or a public network.

[0015] Information processing device 1 is the core server device of the information processing system 100. As described above, information processing device 1 performs question and answer support processing in response to user questions and presents the final answer to the user.

[0016] Generative AI2 is a generative AI that uses LLM, such as ChatGPT, Gemini, Claude, or Copilot. Generative AI2 may consist of one or more AIs. If multiple Generative AI2s are used, they may consist of different types of Generative AI2s or of the same type. Furthermore, if multiple Generative AI2s consist of the same type, they may consist of different versions of Generative AI2s.

[0017] Intermediary system 3 serves as the input / output interface when a user accesses the information processing system 100. Intermediary system 3 is, for example, a chat tool system. Examples of chat tool systems include Chatwork, LINE WORKS, Microsoft Teams, and Slack. Users input questions into the information processing system 100 and receive final answers from the information processing system 100 via the chat tool.

[0018] User terminal 4 is an information terminal used by the user. User terminal 4 could be, for example, a personal computer, a smartphone, or a tablet device. The user uses user terminal 4 to access the information processing system 100 and obtain the final answer to their question.

[0019] <Configuration of the information processing device> As shown in Figure 1, the information processing device 1 functionally comprises a question input unit 11, a storage unit 12, a prompt creation unit 13, a prompt transmission unit 14, an answer receiving unit 15, and an answer output unit 16.

[0020] The question input unit 11 inputs questions from users received via the intermediary system 3.

[0021] The memory unit 12 stores information for creating prompts (hereinafter referred to as "prompt creation information"). The prompt creation information is used when the information processing device 1 creates prompts based on user questions. Figure 2 is a schematic diagram showing the contents of the memory unit 12.

[0022] As shown in Figure 2, the memory unit 12 is configured to include at least a learning model database (hereinafter referred to as "learning model DB") 17, a prompt database (hereinafter referred to as "prompt DB") 18, a reference database (hereinafter referred to as "reference DB") 19, and a history database 20 (hereinafter referred to as "history DB") 20.

[0023] The learning model DB17 stores learning model information for each user. The learning model information consists of information such as the learning model description d1, the number of conversation rallies, the selection of the generating AI2, and the setting of the API key. Of these, the learning model description d1 consists of the user's basic information. The user's basic information includes information such as the user's attribute information and an overview of the information system that the user builds using the information processing system 100. For example, if the user is company X and builds an inquiry system targeting internal users, this would include company information of company X, industry information, and an overview of the inquiry system. The learning model description d1 is information that the prompt creation unit 13, which will be described later, uses as a prompt that indicates the user's basic information when creating a prompt.

[0024] Furthermore, the learning model description d1 includes, as an overview of the information system, the titles d21 of each of the multiple prompt information d2 stored in the prompt DB18 (described later), and the titles d31 of each of the multiple reference information d3 stored in the reference DB19, as a table of contents.

[0025] The prompt DB18 stores multiple prompt information entries d2 for each user. Prompt information d2 is text information created for each of the multiple types (categories) classified by the user. Prompt information d2 can be added, modified, or deleted sequentially as needed by the user.

[0026] For example, if company X uses the information processing system 100 as an inquiry system for internal departments (e.g., Human Resources Department, IT Department, Marketing Department, etc.), prompt information d2 may be created for each department, such as prompt information d2 for Human Resources, prompt information d2 for the IT Department, prompt information d2 for the Marketing Department, etc.

[0027] The prompt information d2 is configured by associating a title d21 that concisely shows the content of the prompt information d2 with detailed information d22 that shows the content of the prompt information d2 in detail. Here, the title d21 is information that the information processing device 1 refers to when selecting the appropriate prompt information d2 in response to a question from the user. More specifically, the title d21 is configured to include title information that can be uniquely identified for user management purposes and prompt summary information that is referred to when the generating AI 2 selects the prompt information d2. For example, if the user asks a question about "sales" or "sales," the prompt information d2 in the prompt summary information of the title d21 that has the keyword "sales" or "sales" or a similar word attached to it (for example, a title d21 that shows text information about "sales and sales") will be selected. In other words, in this embodiment, the prompt information d2 is dynamically selected based on the title d21. The detailed information d22 is information that, when the prompt creation unit 13 creates a prompt after the prompt information d2 has been selected, is used as a prompt related to the question. In this embodiment, the prompt summary information is limited to 100 characters, but the number of characters is not limited to this.

[0028] In this way, the prompt DB18 allows for the creation of prompt information d2 for each user according to the scale, type, or level of the services provided by the user, enabling the setting of prompt information d2 that can flexibly respond to the diverse needs of each user. Furthermore, even after system operation, it is possible to add, modify, or delete prompt information d2 to optimize the response from the generating AI2, making it easy to customize according to the situation.

[0029] Furthermore, in the prompt DB18, a default prompt information d2 that is not classified into any type (category) may be provided. In this case, if the user's question does not fall under any of the prompt information d2 types (categories), the detailed information d22 of the default prompt information d2 may be used as the prompt.

[0030] For example, suppose the user is company X and has built an inquiry system for its internal users. In this system, if the generating AI2 cannot select a title d21 of the multiple prompt information d2 recorded by the user based on the question entered by the internal user, the default prompt information d2 is dynamically selected. In this case, the detailed information d22 of the default prompt information d2 may include text information such as, "We do not have information to answer your question. Please contact the Information Systems Department. The contact address is infosystem@xxx.co.jp. Please send an email with the subject line 'No response in the inquiry system'." This text information may be output as a message to the user terminal 4.

[0031] Reference DB19 stores reference information d3 for each user. Reference information d3 is additional information to the final response to the user. It is used when adding standardized information such as URLs (Uniform Resource Locator) or boilerplate text in addition to the response from Generating AI2. Reference information d3 is text information created for each of the multiple types (categories) classified by the user. Reference information d3 can be added, modified, or deleted sequentially as needed by the user.

[0032] For example, if the information processing system 100 is used as an internal human resources system for a company, reference information d3 might be reference information d3 that contains URLs where forms and templates for various application documents are stored.

[0033] Reference information d3 is composed of a title d31 that concisely indicates the content of the reference information d3, and detailed information d32, such as the URL where the content of the reference information d3 exists and a standard phrase indicating the content of the reference information d3. Here, title d31 is the information referenced when the information processing device 1 selects reference information d3 that matches the user's question. More specifically, title d31 is composed of title information that can be uniquely identified for user management, and reference summary information that is referenced when the generating AI 2 selects prompt information d2. For example, when a user asks a question about "leave application," reference information d3 with keywords such as "leave" or "application" or similar words attached to the reference summary information of title d31 (for example, title d31 indicating text information about "leave application") is dynamically selected from among multiple prompt information d3. Detailed information d32 is used as additional information to be added to the final answer when the prompt creation unit 13 creates a prompt after the reference information d3 has been selected. The reference summary information for this embodiment is limited to 100 characters, but the number of characters is not limited to this.

[0034] In this way, the reference DB19 can create standardized answers as additional information depending on the type of service provided by the user, allowing for the configuration of reference information d3 that can flexibly respond to the diverse needs of each user. Furthermore, even after system operation, it is possible to add, modify, or delete reference information d3 in response to changes in the system environment, making it easy to customize according to the situation.

[0035] The history DB 20 stores history information that shows each user's questions and final answers in chronological order. In this embodiment, the answers to user-initiated questions are described, but the history information may also be used to initiate a dialogue initiated by the information processing system 100. In other words, the system may present the user with relevant questions and suggestions based on the user's past dialogue logs.

[0036] Furthermore, the memory unit 12 stores a program 21 that performs question and answer support processing.

[0037] Returning to Figure 1, the prompt creation unit 13 creates a prompt based on the question received by the question input unit 11. When creating a prompt, it uses the prompt creation information stored in the storage unit 12. In this embodiment, the prompt creation unit 13 creates prompts in multiple stages. Specifically, the prompt creation unit 13, when subdivided, has a configuration as shown in Figure 3(a), which includes a first prompt creation unit 31 that creates the first prompt, a second prompt creation unit 32 that creates the second prompt, and a third prompt creation unit 33 that creates the third prompt. The contents of the first, second, and third prompts will be described later.

[0038] The prompt transmission unit 14 transmits the prompt created by the prompt creation unit 13 to the generation AI 2. In this embodiment, the prompt transmission unit 14 transmits the prompt to the generation AI 2 in multiple steps. Specifically, the prompt transmission unit 14 is further subdivided as shown in Figure 3(b), and consists of a first prompt transmission unit 41 that transmits the first prompt, a second prompt transmission unit 42 that transmits the second prompt, and a third prompt transmission unit 43 that transmits the third prompt.

[0039] The response receiving unit 15 receives responses to prompts sent by the prompt sending unit 14. In this embodiment, since the prompt sending unit 14 sends prompts in multiple parts, the response receiving unit 15 receives multiple responses. Specifically, the response receiving unit 15 can be further subdivided as shown in Figure 3(c), and consists of a first response receiving unit 51 that receives the first response corresponding to the first prompt, a second response receiving unit 52 that receives the second response corresponding to the second prompt, and a third response receiving unit 53 that receives the third response corresponding to the third prompt. The contents of the first, second, and third responses will be described later.

[0040] The response output unit 16 creates a final response to present to the user based on the responses received by the response receiving unit 15, and sends the created final response to the user terminal 4.

[0041] Specifically, the information processing device 1 described above employs the hardware configuration shown in Figure 9.

[0042] Figure 9 is a hardware configuration diagram of the information processing device 1. The information processing device 1 consists of a processor 101, memory 102, storage 103, input / output interface 104, and communication interface 105. The processor 101 consists of, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The memory 102 consists of, for example, ROM (Read Only Memory) or RAM (Random Access Memory). The storage 103 consists of, for example, an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The input / output interface 104 consists of, for example, input devices such as a touch panel, keyboard, mouse, camera, and microphone, and output devices such as a liquid crystal display, an EL (Electro Luminescence) display, and a speaker. The communication interface 105 is capable of sending and receiving various data via the communication network 5, and can be either wired or wireless, and the communication protocol is not limited.

[0043] In addition to being stored in the memory 102 and storage 103 mentioned above, program 21 can also be recorded on computer-readable recording media such as external hard disks, flexible disks, CDs (Compact Discs), DVDs (Digital Versatile Discs), and BDs (Blu-ray® Discs), or distributed via the communication network 5.

[0044] In this embodiment, the information processing device 1 is configured as a separate device from the intermediary system 3, but the information processing device 1 may also be configured to have the functions of the intermediary system 3. In other words, the information processing device 1 may be configured to have an input / output interface for when a user uses the information processing system 100.

[0045] Furthermore, the information processing device 1 may consist of a single physical device, or it may be a system in which multiple devices are connected to a network. If it consists of multiple devices, they may all be installed in the same location, or they may be distributed and installed in multiple locations.

[0046] <Operation of information processing systems / Q&A support processing> Next, the operation of the information processing system 100 will be explained using Figure 4. Figure 4 is a sequence showing the flow of question and answer support processing executed by the information processing device 1 of the information processing system 100. As a prerequisite, the user is logged into the information processing system 100 (for example, logged into the information processing system 100 in conjunction with logging into the intermediary system 3).

[0047] When a user inputs a question via the user terminal 4 and the intermediary system 3 (step S10), the input question is transmitted from the user terminal 4 to the information processing device 1 (step S20). The question may be input as text or as voice input.

[0048] When the information processing device 1 receives a question from the user terminal 4, it first creates a first prompt, sends the created first prompt to the generation AI 2, and simultaneously performs an optimal prompt selection process to receive a first response to the first prompt from the generation AI 2 (step S30). The optimal prompt selection process is performed by the functions of the first prompt creation unit 31, the first prompt transmission unit 41, and the first response receiving unit 51 described above.

[0049] • Optimal prompt selection process Here, we will explain the details of the optimal prompt selection process using Figure 5. Figure 5 is a flowchart showing the flow of the optimal prompt selection process in step S30 of Figure 4.

[0050] When the information processing device 1 receives a question from the user terminal 4 (step S31), it refers to the learning model DB 17 to obtain the learning model description d1 for the logged-in user (step S32) and obtains all the titles d21 of the user's prompt information d2 (step S33). For example, when it receives a question Q1 from a user of company X, it obtains the learning model description d1 for company X stored in the learning model DB 17 and obtains all the titles d21 of the prompt information d2 for company X stored in the prompt DB 18.

[0051] Next, the information processing device 1 creates a first prompt (step S34) based on the question, learning model description d1, and all titles d21 of the prompt information d2 acquired in steps S31 to S33, and sends the created first prompt to the generating AI2 (step S35).

[0052] Here, the first prompt is a prompt that asks for a response to select the title d21 that best matches the content of the user's question from all the titles d21 obtained in step S33. In other words, it is a prompt that asks the user to select the best prompt information d2 from among the multiple prompt information d2 that the user has regarding the received user question, based on the title d21. For example, in the case of X company's learning model description d1 and question Q1, the first prompt consists of the learning model description d1 and question Q1 of X company, and is a prompt that asks for a response to select the best title d21 that matches question Q1 from among the multiple prompt information d2 of X company. In this case, keywords included in question Q1 may be extracted to create the first prompt.

[0053] Next, the information processing device 1 receives a first response from the generating AI 2 that corresponds to the first prompt (step S36).

[0054] Here, the first response is the response from the generating AI2, and it is the response that contains the title d21 of the prompt information d2. For example, if the first prompt is a prompt that asks for a response that selects the title that is best suited to question Q1 from among the titles d21 of multiple prompt information d2 of Company X mentioned above, then the first response will be, for example, a response that contains the title d21 indicating title A (see Figure 2). As a result, the best prompt information d2 for the question can be selected from the prompt DB18. For example, if the first response includes a title d21 indicating title A, then the prompt information d2 with the title d21 indicating title A can be selected.

[0055] As described above, according to the optimal prompt selection process of this embodiment, the information processing device 1 creates a first prompt that asks for a response to select the title d21 of the optimal prompt information d2 based on the user's question, and the generating AI 2 provides a response that selects the title d21 of the optimal prompt information d2 based on the first prompt. Therefore, by using the generating AI 2, even if the user's question is ambiguous, prompt information d2 that matches the user's intent to ask the question can be easily used.

[0056] Furthermore, since the generating AI2 selects the title d21 of the optimal prompt information d2 based on the first prompt, the generating AI2 can reduce the amount of tokens required for its response without having to retrieve all the prompt information d2 stored in the prompt DB18.

[0057] Furthermore, if the title d21 of prompt information d2 is not included in the first response, that is, if there is no prompt information d2 in prompt DB18 that matches the content of the question, prompt information d2 will not be selected. In this case, if a default prompt information d2 has been created in prompt DB18, the default prompt information d2 may be selected. In this way, even if prompt information d2 is not selected, the default prompt information d2 is selected, so that some kind of answer can always be provided to the user's question.

[0058] Returning to Figure 4, the information processing device 1 then creates a second prompt, sends the created second prompt to the generating AI2, and performs an optimal response acquisition process to receive a second response to the second prompt from the generating AI2 (step S40). The optimal response acquisition process is performed by the functions of the second prompt creation unit 32, the second prompt transmission unit 42, and the second response receiving unit 52 described above. • Processing to obtain the optimal answer Here, we will explain the details of the optimal response acquisition process using Figure 6. Figure 6 is a flowchart showing the flow of the optimal response acquisition process in step S40 of Figure 4.

[0059] The information processing device 1 obtains a question from the user (step S41), obtains the user's learning model description d1 (step S42), and obtains detailed information d22 of the optimal prompt information d2 based on the first answer obtained in step S36 of the optimal prompt selection process (see Figure 5) (step S43). For example, if the first answer is a title d21 indicating title A, then detailed information d22 indicating prompt A is obtained (see Figure 2).

[0060] Next, the processing unit 1 creates a second prompt (step S44) based on the question, learning model description d1, and detailed information d22 obtained in steps S41 to S43, and sends the created second prompt to the generating AI2 (step S45).

[0061] Here, the second prompt is a prompt that adds the acquired detailed information d22 and asks for the user's answer to the question. For example, if the learning model description d1 is for company X, question Q1 is the question, and detailed information d22 shows prompt A, then the second prompt will include the learning model description d1 for company X, question Q1, and detailed information d22 showing prompt A, and will ask for the answer to question Q1.

[0062] Furthermore, in this case, since the second prompt is created immediately following the first prompt, the content of the second prompt that overlaps with the content of the first prompt (in this case, the description of Company X's learning model and question Q1) may be shortened by adding it as prompt history information that is the same as the corresponding part of the first prompt. By doing so, the amount of tokens required for the response from generating AI2 can be reduced.

[0063] Next, the information processing device 1 receives a second response from the generating AI 2 that corresponds to the second prompt (step S46).

[0064] Here, the second response is the answer from the generating AI2 to the user's question. For example, if the second prompt consists of a description of Company X's learning model d1, question Q1, and detailed information d22 indicating prompt A, then the second response will be the response a1 containing the answer to question Q1.

[0065] As described above, according to the optimal answer processing of this embodiment, the information processing device 1 creates a second prompt that seeks an answer to the question by taking into account the detailed information d22 of the prompt information d2 that is optimal for the user's question, and the generating AI 2 provides an answer to the question based on the second prompt, so that even if the user's question is ambiguous, an answer that matches the user's intent can be obtained.

[0066] Furthermore, since the generating AI2 takes into account the detailed information d22 of the dynamically selected title prompt information d2 based on the second prompt when providing a response, the generating AI2 can reduce the amount of tokens required for its response without having to retrieve all the prompt information d2 recorded in the prompt DB18.

[0067] Furthermore, if default prompt information d2 exists in the first response from the generated AI2, the second prompt may be created by taking into account the detailed information d22 of the default prompt information d2.

[0068] Returning to Figure 4, the information processing device 1 then creates a third prompt, sends the created third prompt to the generating AI2, and performs additional information acquisition processing to receive a third response to the third prompt from the generating AI2 (step S50). The additional information acquisition processing is performed by the functions of the third prompt creation unit 33, the third prompt transmission unit 43, and the third response receiving unit 53 described above.

[0069] • Processing to retrieve additional information Here, we will explain the contents of the additional information acquisition process in detail using Figure 7. Figure 7 is a flowchart that shows in detail the flow of the additional information acquisition process in step S50 of Figure 4.

[0070] The information processing device 1 obtains a question from the user (step S51), obtains the user's learning model description d1 (step S52), and obtains all the titles d31 of the user's reference information d3 (step S53) by referring to the reference DB 19, similar to steps S31 and S32 of the optimal prompt selection process (see Figure 5). For example, if a question Q1 is received from a user of company X, the information processing device 1 obtains the learning model description d1 of company X stored in the learning model DB 17 and obtains all the titles d31 of the reference information d3 of company X stored in the reference DB 19.

[0071] Next, the information processing device 1 creates a third prompt (step S54) based on the question, the learning model description d1, and all the titles d31 of the reference information d3 obtained in steps S51 to S53, and sends the created third prompt to the generating AI2 (step S55).

[0072] Here, the third prompt is a prompt that asks the user to select a title d31 that matches the content of the user's question from among all the titles d31 obtained in step S53. In other words, if there is an optimal reference information d3 among the multiple reference information d3 that the user has regarding the user's question, the prompt asks the user to select the reference information d31 based on its title d31. For example, the third prompt consists of a learning model description d1 of company X and question Q1, and if there is a title d31 that matches question Q1 among the titles d31 of multiple reference information d3 of company X, the prompt asks the user to select the matching title d31. In this case, keywords included in question Q1 may be extracted to create the third prompt.

[0073] Furthermore, in this process, since the third prompt is created consecutively with the first and second prompts, the content of the third prompt that overlaps with the content of the first and second prompts (in this case, the description of Company X's learning model and question Q1) may be shortened by adding it as prompt history information that is the same as the corresponding parts of the first and second prompts. By doing so, the amount of tokens required for the response from generating AI2 can be reduced.

[0074] Next, the information processing device 1 receives a third response from the generating AI 2 that corresponds to the third prompt (step S56).

[0075] Here, the third response is the response from the generating AI2, and if there is at least one title d31 that matches the content of the question, the response will contain the title d31 of the reference information d3. For example, if the third prompt is a prompt that asks for a response to select a title d31 that matches question Q1 from among the titles d31 of multiple reference information d3 of Company X mentioned above, the third response will be, for example, a response containing a title d31 that indicates title A (see Figure 2). As a result, it is possible to select the reference information d3 that matches the question from the reference DB19. For example, if the third response contains a title d31 that indicates title A, it is possible to select the reference information d3 with the title d31 that indicates title A. This makes it possible to obtain the detailed information d32 of the reference information d3 as additional information.

[0076] As described above, according to the additional information acquisition process of this embodiment, the information processing device 1 creates a third prompt that prompts the user to select the title d31 of the matching reference information d3 based on the user's question, and the generating AI 2 responds by selecting the title d31 of the matching reference information d3 based on the third prompt. Therefore, by using the generating AI 2, it is possible to easily select reference information d3 that matches the user's question intent.

[0077] Furthermore, since the generating AI2 selects the title d31 of the optimal reference information d3 based on the third prompt, the generating AI2 can reduce the amount of tokens required for its response without having to retrieve all the reference information d3 recorded in the prompt DB18.

[0078] Furthermore, if there is no optimal title d31 to select in the third response from generating AI2, it is not necessary to obtain the detailed information d32 of reference information d3 as additional information.

[0079] Returning to Figure 4, finally, the information processing device 1 executes an answer creation process to create a final answer to present to the user (step S60), and then transmits the created final answer to the user terminal 4 (step S70). The answer creation process is performed by the function of the answer output unit 16 described above.

[0080] As a result, the final response is output via the user terminal 4 and the intermediary system 3 (step S80), so the user can view or hear the final response via the user terminal 4 and the intermediary system 3.

[0081] • Answer generation process Here, we will explain the contents of the response creation process in detail using Figure 8. Figure 8 is a flowchart showing the flow of the response creation process in step S60 of Figure 4.

[0082] The information processing device 1 obtains the second answer obtained in the optimal answer acquisition process (see Figure 6) (step S61), and obtains the third answer obtained in the additional information acquisition process (see Figure 7) (step S62). For example, the second answer is the answer a1 to question Q1, and the third answer is the answer d31 which indicates title A.

[0083] Next, the information processing device 1 determines whether the title d31 is included in the third answer obtained in step S62 (step S63). If the title d31 is included in the third answer (step S63: YES), the information processing device 1 refers to the reference DB 19 to obtain the detailed information d32 of the reference information d3 for the title d31 included in the third answer (step S64), and then adds the detailed information d32 to the second answer to create the final answer (step S65). For example, the detailed information d32 of the reference information d31 that has title d31 indicating title A is obtained, and the detailed information d32 contains URL information, "URL If "A" is included, the final answer A1 is the answer a1 to question Q1 and "URL The answers consist of responses that include "A".

[0084] In contrast, if the third answer does not have the title d31 (step S63: NO), the information processing device 1 creates the second answer as the final answer (step S66). For example, if the second answer is answer a1 to question Q1, and the third answer does not have the title d31, then the final answer A1 is only answer a1 to question Q1.

[0085] As described above, according to the answer creation process of this embodiment, the information processing device 1 creates a final answer to present to the user based on the second and third answers created by the generating AI 2. Therefore, even if the user's question is ambiguous, the information processing device 1 can provide the user with a final answer that matches the user's intent. In particular, if the most appropriate URL or standard phrase is added to the final answer in response to the question, the final answer that matches the user's intent can be presented more carefully and specifically.

[0086] In this embodiment, the question-and-answer support processing employs a configuration in which the information processing device 1 sends prompts to the generating AI 2 three times and receives answers, as shown in Figure 4. In this case, prompts may be sent to the same generating AI 2 all three times, or prompts may be sent to different generating AI 2s. In this embodiment, the first transmission and reception to the generating AI 2 (first prompt and first answer) and the third transmission and reception to the generating AI 2 (third prompt and third answer) require an answer that selects the optimal one from multiple options, so the number of tokens is small. In the second transmission and reception to the generating AI 2 (second prompt and second answer), the answer to the question itself is required, so the number of tokens is expected to be large. Therefore, the generating AI 2 may be selected based on the number of tokens. Alternatively, the generating AI 2 may be selected from the viewpoint that the second prompt is more important than the first and third prompts. That is, since the second prompt requires higher answer accuracy than the first and third prompts, the generating AI 2 to which the second prompt is sent may be selected to be more accurate than the generating AI 2 to which the first and third prompts are sent.

[0087] As described above, according to the information processing system 100 of this embodiment, by having the user prepare data for the learning model DB17, prompt DB18, and reference DB19 in advance, the system can dynamically create the optimal prompt in response to the user's question, and obtain an answer from the generating AI2 based on the created prompt, thereby providing the user with an accurate answer that matches the user's intent.

[0088] Furthermore, according to the information processing system 100 of this embodiment, the information processing device 1 and the generating AI 2 are configured to repeatedly send multiple prompts (first prompt, second prompt, and third prompt in this embodiment) and provide answers based on each prompt. As a result, the generating AI 2 can obtain only the necessary information at each stage by going through multiple steps, without acquiring all the information stored in the memory unit 12 (learning model information, prompt information d2, and reference information d3 in this embodiment), and can provide answers based on the acquired information. Consequently, the amount of tokens required for the answers from the generating AI 2 to the information processing device 1 can be reduced. For example, according to existing information processing systems, if a party capable of creating prompts, such as a prompt engineer, asks a question, approximately 128,000 tokens are required to obtain an accurate answer. In contrast, according to the information processing system 100 of this embodiment, the number of tokens required to obtain an accurate answer can be reduced to 4,000 tokens (1,000 tokens for the first prompt answer, 2,000 tokens for the second prompt answer, and 1,000 tokens for the third prompt answer).

[0089] Furthermore, according to the information processing system 100 of this embodiment, in addition to the response from the generating AI2, standardized information can be added as part of the response, making it possible to provide individualized and specific answers that match the user's intent to ask a question.

[0090] Furthermore, the data in Prompt DB18 and Reference DB19 can be added, modified, or deleted as needed according to the user's usage, allowing the user to utilize Generation AI2 while immediately reflecting their own customized information.

[0091] Furthermore, according to the information processing system 100 of this embodiment, for example, an intermediary system 3 such as a commonly used chat tool is used as an input / output interface to input questions and output answers, so that users can use the information processing system 100 without being aware of the generating AI 2 or feeling any discomfort.

[0092] Furthermore, according to the information processing system 100 of this embodiment, prompts are sent to the generating AI2 in stages over multiple steps, so the total number of tokens can be reduced, and the cost of using the generating AI2 can be lowered.

[0093] <Other variations> In the information processing system 100 of the above embodiment, instead of a prompt DB 18 that stores multiple prompt information d2, multiple RAGs (Retrieval-augmented generation) may be provided. Each RAG shall include at least a title that concisely indicates the contents of the RAG, a vector database of the search entities of the RAG, and a generation AI. Then, in response to a user's question, the system shall select the most suitable RAG from among the titles of the multiple RAGs, and then use the selected optimal RAG and the generation AI to obtain an answer to the question. In other words, the system may select the most suitable RAG from among a number of pre-prepared RAGs and obtain the answer from the generation AI by using the selected RAG.

[0094] Furthermore, in the information processing system 100 of the above embodiment, instead of the reference DB 19, a database that collects data whose values ​​change in real time (hereinafter referred to as "real-time DB") may be used. For example, when the information processing system 100 is used as an in-house sales support system, a real-time DB that manages product inventory may be used, and if a user asks a question about product A, the current inventory quantity of product A may be added as additional information to the final answer. <Summary> In summary, the embodiments of the present invention are as follows: (1) According to the prompt generation device according to an embodiment of the present invention, A prompt generation device (e.g., information processing device 1) connected via a communication network (e.g., communication network 5) to a generation AI (e.g., generation AI 2), A question input section (for example, question input section 11) where the user enters a question, A memory unit (for example, memory unit 12) that stores prompt creation information for creating a prompt, A prompt creation unit (for example, a prompt creation unit 13) creates a prompt related to the question based on the question and the prompt creation information, A prompt sending unit (for example, prompt sending unit 14) that sends the created prompt to the generating AI, The response receiving unit (for example, response receiving unit 15) receives a response to a prompt from the generating AI, Equipped with, The prompt creation information comprises basic user information (e.g., learning model information including a learning model description d1) and multiple detailed information classified into multiple categories (e.g., prompt information d2), Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information (e.g., title d21) is associated with descriptive information indicating the content of the detailed information (e.g., detailed information d22), The prompt generation unit includes a first prompt generation unit (for example, a first prompt generation unit 31) that generates a prompt (hereinafter referred to as the "first prompt") which asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and the plurality of detailed information items. The prompt transmitting unit includes a first prompt transmitting unit (for example, a first prompt transmitting unit 41) that transmits the first prompt, The response receiving unit includes a first response receiving unit (for example, a first response receiving unit 51) that receives a response to the first prompt (hereinafter referred to as the "first response"), The prompt generation unit includes a second prompt generation unit (for example, a second prompt generation unit 42) that generates a prompt (hereinafter referred to as a "second prompt") requesting an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. This will be the basic structure.

[0095] Furthermore, according to the prompt generation device of the present invention, in the above basic configuration, The aforementioned detailed information consists of multiple different text data (for example, text data that serves as a command prompt to instruct the generating AI2). This configuration is preferable. Furthermore, according to the prompt generation device of the present invention, in the above basic configuration, The aforementioned detailed information is comprised of multiple different databases (e.g., RAG). This configuration is preferable.

[0096] Here, the prompt creation device according to the embodiment of the present invention performs each of the functions of the prompt creation device according to the embodiment of the present invention by executing the prompt creation method according to the embodiment of the present invention and the prompt creation program according to the embodiment of the present invention.

[0097] In other words, according to the prompt generation method of the embodiment of the present invention, A method for creating a prompt using a prompt creation device (e.g., information processing device 1) connected via a communication network (e.g., communication network 5) to a generation AI (e.g., generation AI 2), The prompt generation device is It includes a storage unit (for example, storage unit 12) that stores prompt creation information for creating prompts, A question input step where a question from the user is entered (for example, a question input unit 11, and question and answer support processing steps S10 and S20), A prompt creation step (for example, a prompt creation unit 13, steps S30 to S50 of the question and answer support processing) that creates a prompt related to the question based on the question and the prompt creation information, A prompt transmission step (for example, prompt transmission unit 14, question and answer processing steps S30 to S50) that sends the created prompt to the generating AI, The response receiving step (for example, response receiving unit 15, question and answer processing steps S30 to S50) receives a response to a prompt from the generating AI, Equipped with, The prompt creation information comprises basic user information (e.g., learning model description d1) and multiple detailed information categorized into multiple categories (e.g., prompt information d2), Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information (e.g., title d21) is associated with descriptive information indicating the content of the detailed information (e.g., detailed information d22), The prompt creation step includes a first prompt creation step (for example, a first prompt creation unit 31, a step S30 of the question and answer support processing) that creates a prompt (hereinafter referred to as the "first prompt") that asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and each of the plurality of detailed information items. The prompt transmission step includes a first prompt transmission step (for example, a first prompt transmission unit 41, a question and answer support processing step S30) for transmitting the first prompt, The response receiving step includes a first response receiving step (for example, a first response receiving unit 51, a step S30 of the question and answer support processing) for receiving a response to the first prompt (hereinafter referred to as the "first response"), The prompt creation step includes a second prompt creation step (for example, a second prompt creation unit 32, step S40 of the question-answer support processing) which creates a prompt (hereinafter referred to as the "second prompt") that requests an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. This will be the second basic structure.

[0098] Furthermore, according to the prompt creation program of the embodiment of the present invention, A prompt creation program (e.g., program 21) is generated by a prompt creation device (e.g., information processing device 1) connected via a communication network (e.g., communication network 5) to a generation AI (e.g., generation AI 2), The prompt generation device is It includes a storage unit (for example, storage unit 12) that stores prompt creation information for creating prompts, A question input step where a question from the user is entered (for example, a question input unit 11, and question and answer support processing steps S10 and S20), A prompt creation step (for example, a prompt creation unit 13, steps S30 to S50 of the question and answer support processing) that creates a prompt related to the question based on the question and the prompt creation information, A prompt transmission step (for example, prompt transmission unit 14, question and answer processing steps S30 to S50) that sends the created prompt to the generating AI, The response receiving step (for example, response receiving unit 15, question and answer processing steps S30 to S50) receives a response to a prompt from the generating AI, The prompt generation device is made to execute the above, The prompt creation information comprises basic user information (e.g., learning model description d1) and multiple detailed information categorized into multiple categories (e.g., prompt information d2), Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information (e.g., title d21) is associated with descriptive information indicating the content of the detailed information (e.g., detailed information d22), The prompt creation step includes a first prompt creation step (for example, a first prompt creation unit 31, a step S30 of the question and answer support processing) that creates a prompt (hereinafter referred to as the "first prompt") that asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and each of the plurality of detailed information items. The prompt transmission step includes a first prompt transmission step (for example, a first prompt transmission unit 41, a question and answer support processing step S30) for transmitting the first prompt, The response receiving step includes a first response receiving step (for example, a first response receiving unit 51, a step S30 of the question and answer support processing) for receiving a response to the first prompt (hereinafter referred to as the "first response"), The prompt creation step includes a second prompt creation step (for example, a second prompt creation unit 32, step S40 of the question-answer support processing) which creates a prompt (hereinafter referred to as the "second prompt") that requests an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. This will be the third basic structure.

[0099] (2) According to the question and answer support device according to an embodiment of the present invention, A question-and-answer support device (e.g., information processing device 1) connected via a communication network (e.g., communication network 5) to a generation AI (e.g., generation AI 2), A question input section (for example, question input section 11) where the user enters a question, A memory unit (for example, memory unit 12) that stores prompt creation information for creating a prompt, A prompt creation unit (for example, prompt creation unit 13) creates a prompt based on the aforementioned question and the prompt creation information, A prompt sending unit (for example, prompt sending unit 14) that sends the created prompt to the generating AI, The response receiving unit (for example, response receiving unit 15) receives a response to a prompt (hereinafter referred to as "intermediate response") from the generating AI, A response output unit (for example, response output unit 16) creates a final response to present to the user based on the aforementioned interim response and outputs the created final response, Equipped with, The prompt creation information comprises basic user information (e.g., learning model information including a learning model description d1) and multiple detailed information classified into multiple categories (e.g., prompt information d2), Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information (e.g., title d21) is associated with descriptive information indicating the content of the detailed information (e.g., detailed information d22), The prompt generation unit includes a first prompt generation unit (for example, a first prompt generation unit 31) that generates a prompt (hereinafter referred to as the "first prompt") which asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and the plurality of detailed information items. The prompt transmitting unit includes a first prompt transmitting unit (for example, a first prompt transmitting unit 41) that transmits the first prompt, The response receiving unit includes a first response receiving unit (for example, a first response receiving unit 51) that receives a response to the first prompt (hereinafter referred to as the "first response"), The prompt generation unit includes a second prompt generation unit (for example, a second prompt generation unit 42) that generates a prompt (hereinafter referred to as a "second prompt") requesting an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. The prompt transmitting unit includes a second prompt transmitting unit (for example, a second prompt transmitting unit 42) that transmits the second prompt. The response receiving unit includes a second response receiving unit (for example, a second response receiving unit 52) ​​that receives a response to the second prompt (hereinafter referred to as the "second response"). The response output unit creates the final response based on the second response and outputs the created final response. This will be the fourth basic structure. According to the question-and-answer support device of the present invention, in the fourth basic configuration described above, The prompt creation information further comprises multiple reference pieces of information (for example, reference piece d3) classified into multiple categories, Each of the aforementioned multiple pieces of reference information is information in which a title indicating the content of the reference information (e.g., title d31) is associated with a URL or a standard phrase indicating the content of the reference information (e.g., detailed information d32). The prompt generation unit includes a third prompt generation unit (for example, a third prompt generation unit 33) that generates a prompt (hereinafter referred to as the "third prompt") that causes the user to select the most suitable reference information based on the titles attached to the question, the basic information, and the multiple reference information. The prompt transmitting unit includes a third prompt transmitting unit (for example, a third prompt transmitting unit 43) that transmits the third prompt. The response receiving unit includes a third response receiving unit (for example, a third response receiving unit 53) that receives a response to the third prompt (hereinafter referred to as the "third response"). The response output unit, if a third response exists, adds the selected URL or standard phrase of the reference information to create the final response and outputs the created final response. This configuration is preferable.

[0100] Furthermore, the question-and-answer support device according to the embodiment of the present invention performs each function of the prompt creation device according to the embodiment of the present invention by executing the question-and-answer support method according to the embodiment of the present invention and by executing the question-and-answer support program according to the embodiment of the present invention.

[0101] In other words, according to the question and answer support method according to an embodiment of the present invention, A method for providing question-and-answer support using a question-and-answer support device (e.g., information processing device 1) connected via a communication network (e.g., communication network 5) to a generating AI (e.g., generating AI 2), The aforementioned question and answer support device, It includes a storage unit (for example, storage unit 12) that stores prompt creation information for creating prompts, A question input step where a question from the user is entered (for example, a question input unit 11, and question and answer support processing steps S10 and S20), A prompt creation step (for example, a prompt creation unit 13, steps S30 to S50 of the question and answer support processing) that creates a prompt based on the aforementioned question and the prompt creation information, A prompt transmission step (for example, prompt transmission unit 14, question and answer processing steps S30 to S50) that sends the created prompt to the generating AI, The response receiving step (for example, response receiving unit 15, question and answer processing steps S30 to S50) receives a response to a prompt (hereinafter referred to as "intermediate response") from the generating AI, A response output step (for example, response output unit 16, question and answer processing steps S60-S70) creates a final response to present to the user based on the aforementioned interim response and outputs the created final response, Equipped with, The prompt creation information comprises basic user information (e.g., learning model description d1) and multiple detailed information categorized into multiple categories (e.g., prompt information d2), Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information (e.g., title d21) is associated with descriptive information indicating the content of the detailed information (e.g., detailed information d22), The prompt creation step includes a first prompt creation step (for example, a first prompt creation unit 31, a step S30 of the question and answer support processing) that creates a prompt (hereinafter referred to as the "first prompt") that asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and each of the plurality of detailed information items. The prompt transmission step includes a first prompt transmission step (for example, a first prompt transmission unit 41, a question and answer support processing step S30) for transmitting the first prompt, The response receiving step includes a first response receiving step (for example, a first response receiving unit 51, a step S30 of the question and answer support processing) for receiving a response to the first prompt (hereinafter referred to as the "first response"), The prompt creation step includes a second prompt creation step (for example, a second prompt creation unit 32, step S40 of the question-and-answer support processing) which creates a prompt (hereinafter referred to as the "second prompt") that requests an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. The prompt transmission step includes a second prompt transmission step (for example, a second prompt transmission unit 42, a question and answer support processing step S40) for transmitting the second prompt, The response receiving step includes a second response receiving step (for example, a second response receiving unit 52, step S40 of the question and answer support processing) for receiving a response to the second prompt (hereinafter referred to as the "second response"), The aforementioned answer output step creates the final answer based on the second answer and outputs the created final answer (for example, steps S50 and S60 of the question and answer support processing). This will be the fifth basic structure.

[0102] Furthermore, according to the question-and-answer support program according to an embodiment of the present invention, A question-and-answer support program (e.g., program 21) provided by a question-and-answer support device (e.g., information processing device 1) connected via a communication network (e.g., communication network 5) to a generation AI (e.g., generation AI 2), The aforementioned question and answer support device, It includes a storage unit (for example, storage unit 12) that stores prompt creation information for creating prompts, A question input step where a question from the user is entered (for example, a question input unit 11, and question and answer support processing steps S10 and S20), A prompt creation step (for example, a prompt creation unit 13, steps S30 to S50 of the question and answer support processing) that creates a prompt based on the aforementioned question and the prompt creation information, A prompt sending step that sends the created prompt to the generating AI, The response receiving step (for example, response receiving unit 15, question and answer processing steps S30 to S50) receives a response to a prompt (hereinafter referred to as "intermediate response") from the generating AI, A response output step (for example, response output unit 16, question and answer processing steps S60-S70) creates a final response to present to the user based on the aforementioned interim response and outputs the created final response, Equipped with, The prompt creation information comprises basic user information (e.g., learning model description d1) and multiple detailed information categorized into multiple categories (e.g., prompt information d2), Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information (e.g., title d21) is associated with descriptive information indicating the content of the detailed information (e.g., detailed information d22), The prompt creation step includes a first prompt creation step (for example, a first prompt creation unit 31, a step S30 of the question and answer support processing) that creates a prompt (hereinafter referred to as the "first prompt") that asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and each of the plurality of detailed information items. The prompt transmission step includes a first prompt transmission step (for example, a first prompt transmission unit 41, a question and answer support processing step S30) for transmitting the first prompt, The response receiving step includes a first response receiving step (for example, a first response receiving unit 51, a step S30 of the question and answer support processing) for receiving a response to the first prompt (hereinafter referred to as the "first response"), The prompt creation step includes a second prompt creation step (for example, a second prompt creation unit 32, step S40 of the question-and-answer support processing) which creates a prompt (hereinafter referred to as the "second prompt") that requests an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. The prompt transmission step includes a second prompt transmission step (for example, a second prompt transmission unit 42, a question and answer support processing step S40) for transmitting the second prompt, The response receiving step includes a second response receiving step (for example, a second response receiving unit 52, step S40 of the question and answer support processing) for receiving a response to the second prompt (hereinafter referred to as the "second response"), The aforementioned answer output step creates the final answer based on the second answer and outputs the created final answer (for example, steps S50 and S60 of the question and answer support processing). This will be the sixth basic structure.

[0103] While embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above. Various modifications and changes, as well as combinations of embodiments of the present invention, can be made to the embodiments of the present invention without departing from the spirit of the invention, and such modifications, changes, and combinations are also included within the technical scope of the present invention. Furthermore, the actions and effects described in the embodiments of the invention are merely a list of the most preferred actions and effects resulting from the present invention, and the actions and effects according to the present invention are not limited to those described in the embodiments of the present invention. [Explanation of Symbols]

[0104] 1. Information Processing Device 2 Generation AI 3. Intermediary System 4. User terminals 5. Communication Network 11 Question Input Section 12 Storage section 13. Prompt Creation Section 14. Prompt sending unit 15. Response Receiving Section 16. Answer Output Section 17. Learning Model Database 18 Prompt DB 19 Reference DB 20 History DB 21 Programs 100 Information Processing Systems

Claims

1. A prompt generation device connected to a generation AI via a communication network, A question input section where users can enter their questions, A memory unit that stores prompt creation information for creating prompts, A prompt creation unit creates a prompt related to the question based on the aforementioned question and the prompt creation information, A prompt transmission unit that sends the created prompt to the generating AI, The response receiving unit receives a response to a prompt from the generating AI, Equipped with, The prompt creation information comprises basic user information and detailed information categorized into multiple categories. Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information is associated with descriptive information indicating the content of the detailed information. The prompt generation unit includes a first prompt generation unit that generates a prompt (hereinafter referred to as the "first prompt") which asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and each of the plurality of detailed information items. The prompt transmission unit includes a first prompt transmission unit that transmits the first prompt, The response receiving unit includes a first response receiving unit that receives a response to the first prompt (hereinafter referred to as the "first response"), The prompt generation unit includes a second prompt generation unit that generates a prompt (hereinafter referred to as the "second prompt") requesting an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. A prompt generation device characterized by the following features.

2. The aforementioned detailed information consists of multiple different text data. The prompt generation device according to claim 1, characterized in that it is a feature of the present invention.

3. The aforementioned detailed information is comprised of multiple different databases. The prompt generation device according to claim 1, characterized in that it is a feature of the present invention.

4. A question-and-answer support device connected to a generation AI via a communication network, A question input section where users can enter their questions, A memory unit that stores prompt creation information for creating prompts, A prompt creation unit creates a prompt based on the aforementioned question and the prompt creation information, A prompt transmission unit that sends the created prompt to the generating AI, The response receiving unit receives a response to a prompt (hereinafter referred to as "intermediate response") from the generating AI, A response output unit that creates a final response to present to the user based on the aforementioned interim response and outputs the created final response, Equipped with, The prompt creation information comprises basic user information and detailed information categorized into multiple categories. Each of the aforementioned multiple pieces of detailed information is information in which a title indicating the content of the detailed information is associated with descriptive information indicating the content of the detailed information. The prompt generation unit includes a first prompt generation unit that generates a prompt (hereinafter referred to as the "first prompt") which asks for a response to select the most appropriate detailed information based on the titles attached to the question, the basic information, and each of the plurality of detailed information items. The prompt transmission unit includes a first prompt transmission unit that transmits the first prompt, The response receiving unit includes a first response receiving unit that receives a response to the first prompt (hereinafter referred to as the "first response"), The prompt generation unit includes a second prompt generation unit that generates a prompt (hereinafter referred to as the "second prompt") requesting an answer to the question based on the question, the basic information, and the explanatory information of the detailed information selected in the first answer. The prompt transmission unit includes a second prompt transmission unit that transmits the second prompt, The response receiving unit includes a second response receiving unit that receives a response to the second prompt (hereinafter referred to as the "second response"), The response output unit creates the final response based on the second response and outputs the created final response. A question-and-answer support device characterized by the following features.

5. The prompt creation information further comprises multiple reference pieces categorized into multiple categories, Each of the aforementioned multiple pieces of reference information is information in which a title indicating the content of the reference information is associated with a URL or a standard phrase indicating the content of the reference information. The prompt generation unit includes a third prompt generation unit that generates a prompt (hereinafter referred to as the "third prompt") that causes the user to select the most suitable reference information based on the titles assigned to the question, the basic information, and the multiple reference information items. The prompt transmission unit includes a third prompt transmission unit that transmits the third prompt, The response receiving unit includes a third response receiving unit that receives a response to the third prompt (hereinafter referred to as the "third response"), The response output unit, if a third response exists, adds the selected URL or standard phrase of the reference information to create the final response and outputs the created final response. The question and answer support device according to feature 4.

Images