Device and method

Abstract

According to the present disclosure, it is possible to assist in appropriate training of a model so that the model can output information that is necessary and sufficient for a user. A device 10 according to the present disclosure comprises: an acquisition unit 11 that acquires question-and-answer data QA1 having a question sentence including a question element and an answer sentence including a plurality of answer elements related to the question element; and an output unit 13 that, on the basis of the question-and-answer data QA1, outputs a prompt P to be input to a question-and-answer generation AI model 31 for generating one or more pieces of sequential question-and-answer data QA2 including at least a part of answer elements extracted from the answer sentence and a question element corresponding to the at least a part of answer elements.

Description

Apparatus and method

[0001] One aspect of the present disclosure relates to an apparatus and a method.

[0002] Technologies that utilize questions and answers contained in text have been disclosed. For example, Patent Literature 1 discloses a question and answer management device that identifies and extracts question content parts and answer content parts from text, associates the extracted question content parts with the answer content parts, and stores them in a storage device.

[0003] Japanese Patent Application Publication No. 11-3335

[0004] By inputting a question into a generative artificial intelligence (AI) model, content that serves as an answer to the question is generated. However, the answer generated by the generative AI model may not be sufficient for the question and may not be an appropriate answer. Therefore, an object of the present disclosure is to provide an apparatus and method that can support appropriate learning of a model that can output necessary and sufficient information for a user.

[0005] The device disclosed herein includes an acquisition unit that acquires question-and-answer data having a question sentence including a question element and an answer sentence including a plurality of answer elements related to the question element, and an output unit that outputs, based on the question-and-answer data, a prompt to be input to a question-and-answer generation AI model for generating one or more pieces of sequential question-and-answer data including at least some of the answer elements extracted from the answer sentence and question elements corresponding to the at least some of the answer elements.

[0006] According to one aspect of the present disclosure, it is possible to support appropriate learning of a model that enables output of necessary and sufficient information for a user.

[0007] Fig. 1 is a block diagram showing a configuration of a processing system including an apparatus according to an embodiment of the present disclosure. Fig. 2 is a diagram showing an example of the configuration of question-and-answer data. Fig. 3 is a diagram showing an example of sequential question-and-answer data in response to a prompt. Fig. 4 is a diagram showing an example of a prompt output by an apparatus according to an embodiment of the present disclosure. Fig. 5 is a flowchart showing the procedure of a process for acquiring sequential question-and-answer data and a learning process performed by a processing system. Fig. 6 is a diagram showing an example of a hardware configuration of an apparatus according to an embodiment of the present disclosure.

[0008] The present disclosure will be described with reference to the accompanying drawings. Whenever possible, the same parts are designated by the same reference numerals and redundant description will be omitted.

[0009] 1 is a block diagram showing the configuration of a processing system including an apparatus according to an embodiment of the present disclosure. The processing system 1 shown in Fig. 1 includes a terminal 5, an apparatus 10, a question-and-answer database 21, a prompt database 22, a sequential question-and-answer database 23, and a server apparatus 30, which are configured to be able to communicate with each other via a network including a wireless communication network and a fixed communication network. The apparatus 10 constitutes a generating apparatus that generates sequential question-and-answer data based on question-and-answer data acquired from the question-and-answer database 21.

[0010] The terminal 5 is a device used by a user who wishes to obtain information about the question-and-answer data. The information about the question-and-answer data that a user wishes to obtain is, for example, information about an answer that a user wants to obtain when the answer is the same as or similar to a question included in the question-and-answer data. The terminal 5 is, for example, a personal computer, a smartphone, a tablet terminal, a feature phone, a server device, a game console, or the like. Note that while only one terminal is illustrated in FIG. 1 , the processing system 1 may include any number of terminals 5, two or more.

[0011] The server device 30 is a device that enables the provision of content using at least one generative AI model. A generative AI model is a model that can generate content in response to a prompt including input information, according to the instructions, context, question, and output format indicated by the prompt, and return the content as response information. The prompt can also include input information, in which case the generative AI model generates response information targeted at the input information. The generative AI model may be, for example, an interactive AI model that includes a large language model (LLM) and a user interface (UI) for interacting with the user, enabling text or voice chat with the user. Examples of such generative AI models include ChatGPT, GPT (registered trademark)-3.5, GPT-4V, PaLM2, etc.

[0012] In this embodiment, the server device 30 is capable of providing a content provision function using a question and answer generation AI model 31 and a response generation AI model 32 as multiple interactive AI models. The question and answer generation AI model 31 and the response generation AI model 32 may be stored in the server device 30, or may be stored in another device connected to the server device 30 via a network, and configured to enable information exchange with a user via the server device 30. Note that although only one server device 30 is illustrated in FIG. 1 , the processing system 1 may include multiple server devices 30.

[0013] The question and answer generation AI model 31 is a generation AI model that generates one or more pieces of sequential question and answer data (described later). The response generation AI model 32 outputs an answer sentence corresponding to a question sentence as response information to the terminal 5 in response to a prompt including the question sentence transmitted from the terminal 5.

[0014] The question and answer database 21 stores question and answer data having questions and answers. FIG. 2 is a diagram showing an example of the structure of question and answer data. The question and answer data QA1 shown in FIG. 2 is question and answer data for a mobile phone rate plan. The question and answer data QA1 included in the question and answer database 21 may be, for example, information obtainable from a Q&A (Question and Answer) page on a web page. The question and answer data may be stored in advance in the question and answer database 21 under the control of the terminal 5 or another terminal.

[0015] As shown in FIG. 2 , the question-and-answer data includes at least one question, such as Question 1, Question 5, or Question 6, and at least one answer, such as Answer 1, Answer 5, or Answer 6. The answer includes multiple answer elements related to the content of the question. In the example shown in FIG. 2 , the content of question Question 5 is the provision conditions of a pricing plan. Answer 5 corresponding to question Question 5 includes multiple answer elements related to the provision conditions of the pricing plan, which is the content of the question, such as summary information on the provision conditions, detailed information on each provision condition, and additional information on the provision conditions. In this way, answer elements refer to elements that differ from each other in terms of attributes, information resolution, type, meaning, etc. Answer elements may be elements obtained by dividing an answer into attributes, information resolution, type, meaning, etc.

[0016] Note that some of the answer sentences included in the question-and-answer data may contain only one answer element related to the content of the question sentence. For example, the answer sentence Answer1 contains summary information about a pricing plan as one answer element. In this embodiment, at least one answer sentence among the question-and-answer data acquired by the device 10 contains multiple answer elements.

[0017] Although not shown in FIG. 2 , the question-and-answer data may include judgment label information for a set of question sentences and answer sentences. Hereinafter, a set of question sentences and answer sentences may be referred to as a question-and-answer set. The judgment label information is information that can be compared with user input information by the judgment unit 12 described below. The judgment label information includes, for example, knowledge level, target age, target gender, target area, target period, etc. The judgment label information is associated in advance with each question-and-answer set included in the question-and-answer data. Note that the judgment label information may be information estimated by a generative AI model included in the server device 30 based on the question-and-answer set.

[0018] 1 stores prompt templates. The prompt database 22 stores prompt templates for generating one or more pieces of sequential question-and-answer data, which will be described later. The templates will be described in detail later.

[0019] The sequential question-and-answer database 23 stores one or more pieces of sequential question-and-answer data generated by the device 10. FIG. 3 is a diagram illustrating an example of sequential question-and-answer data for a prompt. As shown in FIG. 3, the sequential question-and-answer data QA2 includes at least some answer elements extracted from an answer sentence and question elements corresponding to the at least some answer elements. In the sequential question-and-answer data QA2, question elements and answer elements correspond one-to-one. An answer element in the sequential question-and-answer data QA2 is at least one answer element selected from multiple answer elements included in the answer sentence of the question-and-answer data. A question element is an element including a question to which the answer element is an answer. For example, if an answer element includes any of information such as summary information, detailed information, additional information, or notes, the question element is an element set according to the type of information in the answer element. The question element in the question sentence included in the question-and-answer data QA1 and the question element included in the sequential question-and-answer data QA2 do not have to be identical.

[0020] In the example shown in Fig. 3, only some of the multiple answer elements contained in the answer sentence Answer5 corresponding to the question sentence Question5 are extracted and decomposed into three answer elements. Each answer element in the sequential question-and-answer data QA2 indicates, for example, summary information on the provision conditions as Answer5(1), detailed information on each provision condition as Answer5(2), and additional information on the provision conditions as Answer5(3). In this way, in the sequential question-and-answer data QA2, the answer sentence is divided into each answer element.

[0021] In the example shown in Figure 3, some answer elements are extracted from the multiple answer elements contained in the answer sentence Answer6 corresponding to the question sentence Question6 and broken down into four answer elements. Each answer element in the sequential question-and-answer data QA2, for example, indicates summary information about the discount service and benefits as Answer6(1), detailed information about the discount service and benefits as Answer6(2), notes about the discount service as Answer6(3), and notes about the discount service and benefits that do not overlap with Answer6(3) as Answer6(4). In this way, in the sequential question-and-answer data QA2, the answer sentences are divided by answer element.

[0022] If a question-and-answer set in the question-and-answer data QA1 does not have multiple answer elements, the sequential question-and-answer data QA2 includes a set (hereinafter referred to as a sequential question-and-answer set) in which the question-and-answer set is reused as is for the sequential question-and-answer data. In the example shown in Figure 3, one answer element included in the answer sentence Answer1 corresponding to the question sentence Question1 is extracted and not broken down. Each answer element in the sequential question-and-answer data QA2 indicates, for example, summary information of the fee plan "xxx" as Answer1(1).

[0023] The sequential answer sentences, which are answer sentences in the sequential question-and-answer data QA2, contain some answer elements, but are supplemented with appropriate words and phrases to avoid grammatical errors. Furthermore, the sequential question sentences, which are question sentences in the sequential question-and-answer data QA2, contain question elements corresponding to some answer elements, but are supplemented with appropriate words and phrases to avoid grammatical errors. Details of the method for generating sequential question-and-answer data will be described later.

[0024] The device 10 includes, as functional components, an acquisition unit 11, a determination unit 12, an output unit 13, a data generation unit 14, and a learning unit 15. The device 10 transmits a prompt including question-and-answer data QA1 acquired from a question-and-answer database 21 to the server device 30. The device 10 also receives response information from the server device 30 in response to the prompt. The response information here includes, for example, one or more pieces of sequential question-and-answer data QA2. The device 10 then transmits the sequential question-and-answer data QA2 to the server device 30 as training data. The functions of each functional unit of the device 10 will be described in detail below.

[0025] The acquisition unit 11 acquires question and answer data QA1 having question sentences and answer sentences from a question and answer database 21 that stores the question and answer data QA1 in advance.

[0026] The acquisition unit 11 acquires usage pattern information based on at least one of user input information and extracted information extracted from question and answer data. The acquisition unit 11 acquires input information input by a user to the terminal 5 from the terminal 5. The input information includes user attribute information and theme information the user wants to use. The attribute information includes the user's age, gender, occupation, address, information on apps they normally use, etc. The theme information includes information on themes and topics according to the needs of the user who wants to ask a question. The themes and topics include background information for each specialty, as well as types of products and services.

[0027] The acquisition unit 11 extracts extracted information from the acquired question-and-answer data. The extracted information includes words and phrases contained in the questions and answers of the question-and-answer data and background information related to the words and phrases. The words and phrases here include proper nouns such as product names and service names, and indicators indicating feature quantities such as target age, target area, and target period.

[0028] The acquisition unit 11 acquires usage mode information relating to usage modes of the sequential question-and-answer data QA2. This usage mode information is information relating to usage modes of the sequential question-and-answer data QA2. For example, the usage mode information may include extracted data such as subjects using the data, constraints, output formats, etc., or a combination of these, or may include a collection (domain) of the extracted data. The usage mode information may include usage domain information as the collection of the extracted data. The acquisition unit 11 may acquire and organize the usage mode information based on information included in at least one of the user's input information and the extracted information.

[0029] The determination unit 12 determines whether to generate sequential question-and-answer data QA2 based on the user's input information. For example, the device 10 selects a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated from the question-and-answer data acquired by the acquisition unit 11, and generates the sequential question-and-answer data QA2 from the selected question-and-answer set. The determination unit 12 determines whether to generate sequential question-and-answer data QA2 based on the determination label information included in the question-and-answer data QA1 acquired by the acquisition unit 11 and the user's input information. For example, the determination unit 12 determines whether the question-and-answer data QA1 includes a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated, based on the determination label information and the user's input information.

[0030] As an example, a case will be described in which the user's input information includes a knowledge level indicating that the user is an expert who is well-versed in the theme and topic of the question-and-answer data QA1. For example, if the knowledge level of the determination label information corresponding to at least one question-and-answer set in the question-and-answer data QA1 is higher than the knowledge level of the user's input information, the determination unit 12 determines that the question-and-answer data QA1 contains a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated. In other words, if at least one question-and-answer set in the question-and-answer data QA1 contains a question and an answer that are estimated to be unknown even to an expert, the determination unit 12 determines that the question-and-answer data QA1 contains a question and an answer that are estimated to be unknown to an expert.

[0031] For example, if the knowledge level of the judgment label information corresponding to all question-and-answer sets in the question-and-answer data is equal to or lower than the knowledge level of the user's input information, the determination unit 12 determines that the question-and-answer data QA1 does not contain any question-and-answer sets for which sequential question-and-answer data QA2 must be generated. In other words, if all the question-and-answer sets in the question-and-answer data QA1 contain questions and answers that are presumed to be known by an expert, the determination unit 12 determines that the question-and-answer data QA1 does not contain any question-and-answer sets for which sequential question-and-answer data QA2 must be generated.

[0032] Note that when the determination unit 12 determines whether a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated is present in the question-and-answer data QA1, the number of question-and-answer sets is not limited. In the above example, the knowledge level of the determination label information is used. However, the determination unit 12 may make the determination by comparing any one of the indicators, such as knowledge level, knowledge level, target age, target gender, target domain, and target period, included in the determination label information with the same indicator in the user's input information. The determination unit 12 may make the determination by comparing multiple indicators, such as knowledge level, knowledge level, target age, target gender, target domain, and target period, included in the determination label information with the same multiple indicators in the user's input information. In this case, the determination unit 12 may determine that a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated is present in the question-and-answer data QA1 based on the determination result for any one indicator or the determination results for all indicators.

[0033] The output unit 13 outputs a prompt to be input to the question and answer generation AI model 31 based on the question and answer data. To output the prompt, the output unit 13 acquires a prompt template based on the usage mode information acquired by the acquisition unit 11.

[0034] The templates stored in the prompt database 22 include usage mode information regarding usage modes for the sequential question-and-answer data QA2. The template format is, for example, text data. The prompt database 22 has multiple prompt templates according to the type of usage mode information. The prompt database 22 has multiple prompt templates according to the entity that uses the data, constraints, output format, etc.

[0035] The template expresses, for example, in text, the instructions to be executed by the question and answer generation AI model, the tasks to be executed by the question and answer generation AI model, the background and context to be taken into consideration by the question and answer generation AI model, fields for entering input information to the question and answer generation AI model, and the output format of response information from the question and answer generation AI model. The instructions and tasks to be executed by the question and answer generation AI model include, for example, being given the task of generating sequential question and answer data QA2 from question and answer data QA1 in accordance with constraints. The background and context to be taken into consideration by the question and answer generation AI model include, for example, detailed descriptions of the question and answer data QA1 and the sequential question and answer data QA2. The fields for entering input information to the question and answer generation AI model are blank fields in the template and are input fields into which input is made by the output unit 13. The output format of the response information from the question and answer generation AI model may include, for example, assigning codes to the sequential question sentences and sequential answer sentences in the sequential question and answer data QA2, and displaying the sequential question sentences and sequential answer sentences as a set. The template may include, for example, information other than the question and answer data in the prompt. The prompt database 22 may include multiple prompt templates depending on, for example, the entity that processes the data, the task content of the data processing, etc.

[0036] The output unit 13 acquires from the prompt database 22 a template having the highest degree of similarity between the usage mode information acquired by the acquisition unit 11 and the usage mode information set in the template. The degree of similarity is, for example, the number of matches between the types of items included in the usage mode information acquired by the acquisition unit 11 and the types of items included in the usage mode information set in the template.

[0037] 4 is a diagram illustrating an example of a prompt output by an apparatus according to an embodiment of the present disclosure. As illustrated in FIG. 4 , the output unit 13 outputs a prompt P based on the question-and-answer data QA1 and a predetermined prompt template. The output unit 13 may, for example, store the prompt P in the prompt database 22 or display it on a display device. The output unit 13 applies the question-and-answer data QA1 acquired by the acquisition unit 11 to the portion of the question-and-answer data QA1 that is not entered in the template of the prompt P. Here, if the determination unit 12 determines that there is question-and-answer data including a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated, the output unit 13 may extract only the question-and-answer set for which sequential question-and-answer data QA2 needs to be generated from the question-and-answer data QA1 acquired by the acquisition unit 11.

[0038] The output unit 13 may add text information specifying the content of the sequential question-and-answer data QA2 generated by the question-and-answer generation AI model 31 and the order (priority) of the sequential question-and-answer sets to the prompt P, based on the usage mode information acquired by the acquisition unit 11. For example, if the usage mode information includes information such as the user's age being 30 or older, the output unit 13 may output a prompt P including an instruction to subdivide answer elements so as to enrich the content of information related to fees, an instruction to arrange information related to fees at the top of the order of other information, etc.

[0039] In the example shown in Figure 4, prompt P includes text indicating the commands and tasks to be executed by the question and answer generation AI model 31, such as "You are given the task of breaking down the given question and answer data into a multi-turn exchange. In accordance with the constraints, convert the given question and answer set into an exchange in which information is answered over multiple turns rather than a single turn of question and answer." It also includes the following constraints: "- Consists of a Question and an Answer, and follows the following output format: - The first Question (Q1) shall ask for an overview; - The first Answer (A1) shall answer a summary; - The second and subsequent Questions (Q2, Q3, ...) shall follow on from the first, and shall not duplicate proper nouns with Q1; - The second and subsequent Answers (A2, A3, ...) shall contain content that was unclear in the first one."

[0040] The prompt P includes the statement "You are a data creation professional" as text indicating the background and context that the question and answer generation AI model 31 should take into consideration. The prompt P also includes the statement "{{"Question1": Q1, "Answer1": A1}, {"Question2": Q2, "Answer2": A2}, ... {"Questionn": Qn, "Answern": An},}" as the output format of the response information from the question and answer generation AI model 31.

[0041] The prompt P is generated by inserting the question and answer data QA1 into the input information entry field for the question and answer generation AI model 31 by the output unit 13, thereby generating the following questions: "{"Question 1": "Please tell me about the outline of the rate plan "xxx"", "Answer 1": "It is a rate plan that packages a voice plan, ISP service, and data communication."}," "{"Question 5": "What are the conditions for offering this plan?", "Answer 5": "The conditions for offering this plan are that the contract name is an individual / corporate name aged 12 or older, the payment method is direct debit or credit card payment, and the use of a compatible model. Note that xxx is a 5G contract, but with the exception of some models, it can also be used with 4G-compatible models."}," and "{"Question 6": "Are there any discount services or benefits?", "Answer 6": "Discount services include the yyy card payment discount, the zzz Hikari set discount, and the 5G set discount. However, the 0.5GB plan is not eligible for any discounts. It is also not eligible for the long-term user appreciation benefit, and is not eligible for the Minna Zzz Discount, but it is counted in the number of group lines.

[0042] 1 inputs a prompt P to the question-and-answer generation AI model 31 to generate one or more pieces of sequential question-and-answer data QA2. The data generation unit 14 generates the sequential question-and-answer data QA2 when the determination unit 12 determines that the sequential question-and-answer data QA2 should be generated. The data generation unit 14 inputs the prompt P output by the output unit 13 to the question-and-answer generation AI model 31. The data generation unit 14 acquires the sequential question-and-answer data QA2 output from the question-and-answer generation AI model 31.

[0043] The data generation unit 14 outputs the acquired sequential question-and-answer data QA2. For example, the data generation unit 14 stores the acquired sequential question-and-answer data QA2 in the sequential question-and-answer database 23. The data generation unit 14 may also output the acquired sequential question-and-answer data QA2 to the learning unit 15.

[0044] The learning unit 15 trains the response generation AI model 32 that enables sequential responses based on the sequential question-and-answer data QA2 generated by the data generation unit 14. Sequential responses include, for example, responding by matching a minimum number of answer elements (e.g., one) to one question element. To this end, the learning unit 15 first obtains the sequential question-and-answer data QA2 generated by the data generation unit 14 from the data generation unit 14 or the sequential question-and-answer database 23.

[0045] The learning unit 15 converts the sequential question-and-answer data QA2 into a structure and format that can be learned by the response generative AI model 32. For example, if the format that the response generative AI model 32 requires during learning is the json format, the learning unit 15 converts the sequential question-and-answer data QA2 into the json format. Note that the output unit 13 may include a description in the prompt to convert the data structure and output format of the sequential question-and-answer data QA2 output from the question-and-answer generative AI model 31 into a structure and format that can be learned by the response generative AI model 32. If the sequential question-and-answer data QA2 generated by the data generation unit 14 already has a structure and format that can be learned by the response generative AI model 32, the learning unit 15 does not need to perform this conversion process.

[0046] The learning unit 15 trains the response generative AI model 32 based on the sequential question-and-answer data QA2 converted into a structure and format that the response generative AI model 32 can learn from. The learning unit 15 may generate prompts including the sequential question-and-answer data QA2 and input them to the response generative AI model 32 so as to fine-tune the response generative AI model 32. The learning unit 15 may train the response generative AI model 32 with a matching table that indicates question elements and answer elements included in the sequential question-and-answer data QA2.

[0047] The response generation AI model 32 learns using the sequential question-and-answer data QA2 sent by the learning unit 15 as training data. Because the sequential question-and-answer data QA2 has a one-to-one correspondence between question elements and answer elements, when a question sentence including an element that is the same as or similar to a question element is input, the response generation AI model 32 learns to select an answer element that corresponds to the question element in the sequential question-and-answer data QA2 and output an answer sentence including the answer element.

[0048] For example, when a question sentence is input from the terminal 5, the response generation AI model 32 extracts question elements contained in the question sentence. The response generation AI model 32 selects a question element in the sequential question-and-answer data QA2 that has the highest similarity to the extracted question element. The response generation AI model 32 may calculate the similarity between the extracted question element and one or more question elements in the sequential question-and-answer data QA2. The response generation AI model 32 selects an answer element corresponding to the selected question element. The response generation AI model 32 generates an answer sentence based on the question sentence input to the terminal 5 and the selected answer element. The response generation AI model 32 outputs the generated answer sentence to the terminal 5. The response generation AI model 32 may further generate an answer sentence based on the calculated similarity. The response generation AI model 32 generates an answer sentence based on the similarity, and can include in the answer sentence (information) a logical explanation regarding the matters considered in response to the question sentence, or content that looks as if it were thought up by a human. Note that the device 10 may accept information including a question sentence input from the terminal 5 and relay it to the response generation AI model 32 of the server device 30, or may accept information including an answer sentence output from the response generation AI model 32 of the server device 30 and relay it to the terminal 5.

[0049] When a question is input from the terminal 5, the response generation AI model 32 may output information (score) indicating the degree of likelihood that one or more answer elements correspond to the question. The score may be calculated, for example, based on vector data of the answer elements and vector data of the question. The score may also be calculated by other methods. The response generation AI model 32 may output the score in association with each answer element. Based on the score and a predetermined threshold, the response generation AI model 32 may select only answer elements having a score equal to or greater than the threshold to be output as a prompt. In this case, the response generation AI model 32 can answer the question based on the score, thereby generating an answer (information) that includes a logical explanation regarding the factors considered in the question or content that appears to be thought up by a human.

[0050] The processing procedure by the processing system 1 and the device 10 configured as above, i.e., the flow of the processing method according to this embodiment, will be described below. Fig. 5 is a flowchart showing the procedure of the acquisition process and learning process of sequential question-and-answer data.

[0051] In the processing method, first, in step S1, the acquisition unit 11 acquires question-and-answer data QA1 and input information. The acquisition unit 11 acquires the question-and-answer data QA1 from the question-and-answer database 21. The acquisition unit 11 acquires the user's input information from the terminal 5.

[0052] Next, in step S2, the acquisition unit 11 acquires usage mode information. The acquisition unit 11 extracts extracted information from the acquired question and answer data QA1. The acquisition unit 11 acquires usage mode information based on information included in at least one of the user input information and the extracted information.

[0053] Next, in step S3, the determination unit 12 determines whether or not to generate sequential question-and-answer data QA2 based on the user's input information. For example, the determination unit 12 determines whether or not the question-and-answer data QA1 contains a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated based on the determination label information and the user's input information.

[0054] For example, if the knowledge levels of the judgment label information corresponding to all question-and-answer sets in the question-and-answer data QA1 are equal to or lower than the knowledge level of the user's input information, the determination unit 12 determines that the question-and-answer data QA1 does not contain any question-and-answer sets for which sequential question-and-answer data QA2 must be generated, and determines not to generate the sequential question-and-answer data QA2 (step S3: NO). In this case, the flowchart shown in FIG. 5, which is the processing method performed by the processing system 1 and the device 10, is terminated.

[0055] For example, if the knowledge level of the judgment label information corresponding to at least one question-and-answer set in the question-and-answer data QA1 is higher than the knowledge level of the user's input information, the judgment unit 12 determines that a question-and-answer set for which sequential question-and-answer data QA2 needs to be generated exists in the question-and-answer data QA1, and determines to generate sequential question-and-answer data QA2 (step S3: YES).

[0056] If the determination unit 12 determines that the sequential question-and-answer data QA2 should be generated (step S3: YES), the output unit 13 acquires a prompt template at step S4. The output unit 13 acquires the prompt template based on the usage mode information acquired by the acquisition unit 11.

[0057] Next, in Step S5, the output unit 13 outputs a prompt P. The output unit 13 outputs the prompt P based on the question-and-answer data QA1 and a predetermined prompt template.

[0058] Next, in step S6, the data generation unit 14 outputs the prompt P output by the output unit 13 to the question and answer generation AI model 31, and inputs the prompt P to the question and answer generation AI model 31.

[0059] Next, in step S7, the data generation unit 14 acquires one or more pieces of sequential question-and-answer data QA2 from the question-and-answer generative AI model 31. The data generation unit 14 acquires the sequential question-and-answer data QA2 generated in the question-and-answer generative AI model 31 based on the prompt P input in step S6. The data generation unit 14 stores the acquired sequential question-and-answer data QA2 in the sequential question-and-answer database 23. Note that if only the process of acquiring the sequential question-and-answer data QA2 is executed, the flowchart may end at step S7.

[0060] Next, in step S8, the learning unit 15 converts the sequential question-and-answer data QA2 into a structure and format that can be learned by the response generative AI model 32. The learning unit 15 acquires the sequential question-and-answer data QA2 generated by the data generation unit 14 from the data generation unit 14 or the sequential question-and-answer database 23. The learning unit 15 determines whether the acquired sequential question-and-answer data QA2 needs to be converted into a structure and format that can be learned by the response generative AI model 32. If it is determined that the sequential question-and-answer data QA2 needs to be converted into a structure and format that can be learned by the response generative AI model 32, the learning unit 15 converts the sequential question-and-answer data QA2 into a structure and format that can be learned by the response generative AI model 32. If it is determined that the sequential question-and-answer data QA2 does not need to be converted into a structure and format that can be learned by the response generative AI model 32, the learning unit 15 omits the processing of step S8.

[0061] Next, in step S9, the learning unit 15 outputs the converted sequential question-and-answer data QA2 as learning data (teacher data) to the response generation AI model 32.

[0062] Next, in step S10, the learning unit 15 causes the response generative AI model 32 to learn the sequential question-and-answer data QA2 converted as learning data. This enables the response generative AI model 32 to output an appropriate answer sentence in response to a question sentence input from the terminal 5. The learning unit 15 may, for example, receive a notification from the response generative AI model 32 that learning has been completed. When step S10 is completed, the flowchart shown in FIG. 5 , which is a processing method by the processing system 1 and the device 10, ends.

[0063] Next, the effects of the device and method disclosed herein will be described with reference to an example of a conventional problem. By using question-and-answer data, which includes a set of question and answer content, as training data, it is possible to train a generative AI model to output content including answers in response to input data including questions. Here, when question-and-answer sentences written on the web or in instructions are imported as training data, a single question often has a set of answer sentences containing multiple answers. Therefore, a generative AI model trained using such question-and-answer sentences as training data may output insufficient (inappropriate) content, such as excessive information, in response to input data including a single question. To train a model (generative AI model) that enables interactions that sequentially extract information to facilitate user understanding, it is necessary to prepare training data suitable for learning.

[0064] The device 10 of the present disclosure includes an acquisition unit 11 that acquires question-and-answer data QA1 having a question sentence including a question element and an answer sentence including multiple answer elements related to the question element, and an output unit 13 that outputs a prompt P to be input to a question-and-answer generation AI model 31 that generates, based on the question-and-answer data QA1, one or more pieces of sequential question-and-answer data QA2 including at least some answer elements extracted from the answer sentence and question elements corresponding to the at least some answer elements. In this case, even if the answer sentence of the question-and-answer data QA1 includes multiple answer elements (a large amount of data), the output unit 13 can output a prompt P that can be input (instructed) to the question-and-answer generation AI model 31 so as to generate sequential question-and-answer data QA2 in which the question element and at least some of the answer elements are associated. The sequential question-and-answer data QA2 obtained by inputting the prompt P to the question-and-answer generation AI model 31 is suitable for training a response generation AI model 32, which is an example of a model that enables interaction to sequentially extract information in a manner that is easy for a user to understand. Therefore, the device 10 and the above-described method (acquisition method) can support appropriate learning of a model that can output necessary and sufficient information for the user.

[0065] The device 10 of the present disclosure further includes a data generation unit 14 that generates sequential question-and-answer data QA2 by inputting the prompt P into the question-and-answer generation AI model 31. In this case, the data generation unit 14 can prepare for learning of the response generation AI model 32 by generating (obtaining) the sequential question-and-answer data QA2.

[0066] The device 10 of the present disclosure further includes a learning unit 15 that trains a response generation AI model 32 that enables sequential responses based on the generated sequential question-and-answer data QA2. In this case, the response generation AI model 32 can sequentially retrieve information in a manner that is easy for the user to understand.

[0067] Furthermore, in the device 10 of the present disclosure, the output unit 13 outputs the prompt P based on the question-and-answer data QA1 and a predetermined prompt template. In this case, the output unit 13 can easily generate the prompt P by inserting the question-and-answer data QA1 into the predetermined prompt template.

[0068] In the device 10 of the present disclosure, the template includes usage mode information regarding usage modes for the sequential question-and-answer data QA2. In this case, since the template including the usage mode information is prepared in advance, the output unit 13 can easily generate a prompt P that conforms to the usage mode information.

[0069] Furthermore, in the device 10 of the present disclosure, the acquisition unit 11 acquires usage mode information based on at least one of the user's input information and the extracted information extracted from the question-and-answer data QA1, and the output unit 13 acquires a prompt template based on the acquired usage mode information. In this case, the user can be prompted to acquire a prompt template that is in line with the usage mode information of the user or the question-and-answer data, and therefore the output unit 13 can output a more appropriate prompt P in accordance with a prompt template that is in line with the needs of the user who will ultimately use the response generation AI model 32 and the content of the information to be input to the question-and-answer generation AI model 31.

[0070] Furthermore, in the device 10 of the present disclosure, the output unit 13 outputs the prompt P based on the question-and-answer data QA1 and usage mode information regarding usage modes for the sequential question-and-answer data QA2. In this case, the user can be prompted to create a prompt P that is in line with the usage mode information, and therefore, a prompt P can be generated that enables the generation of sequential question-and-answer data QA2 that better matches the user's needs.

[0071] The device 10 of the present disclosure further includes a determination unit 12 that determines whether to generate sequential question-and-answer data QA2 based on information input by the user, and the data generation unit 14 generates the sequential question-and-answer data QA2 when the determination unit 12 determines that the sequential question-and-answer data QA2 should be generated. In this case, the device 10 can generate sequential question-and-answer data QA2 that includes content that matches the needs of the user.

[0072] Furthermore, in the device 10 of the present disclosure, the learning unit 15 converts the sequential question-and-answer data QA2 into a structure and format that can be learned by the response generative AI model 32. In this case, even if the sequential question-and-answer data QA2 output by the question-and-answer generative AI model 31 does not have a structure and format that is suitable for learning by the response generative AI model 32, the learning unit 15 can convert the sequential question-and-answer data QA2 into an appropriate structure and format depending on the configuration and function of the response generative AI model 32.

[0073] The device and method of the present disclosure have the following configuration.

[0074] [1] An apparatus comprising: an acquisition unit that acquires question-and-answer data having a question sentence including a question element and an answer sentence including a plurality of answer elements related to the question element; and an output unit that outputs, based on the question-and-answer data, a prompt to be input to a question-and-answer generation AI model that generates one or more pieces of sequential question-and-answer data including at least some of the answer elements extracted from the answer sentence and question elements corresponding to the at least some of the answer elements.

[0075] [2] The device according to [1], further comprising a data generation unit that inputs the prompt to the question and answer generation AI model to generate the sequential question and answer data.

[0076] [3] The device according to [2], further comprising a learning unit that trains a response generation AI model that enables sequential responses based on the generated sequential question-and-answer data.

[0077] [4] The device according to any one of [1] to [3] above, wherein the output unit outputs the prompt based on the question and answer data and a predetermined template of the prompt.

[0078] [5] The device according to [4], wherein the template includes usage mode information regarding usage modes for the sequential question-and-answer data.

[0079] [6] The device described in [5], wherein the acquisition unit acquires the usage mode information based on at least one of user input information and extracted information extracted from the question and answer data, and the output unit acquires the template of the prompt based on the acquired usage mode information.

[0080] [7] The device according to any one of [1] to [6], wherein the output unit outputs the prompt based on usage mode information relating to a usage mode for the sequential question-and-answer data and the question-and-answer data.

[0081] [8] The device according to [2] or [3] above, further comprising a determination unit that determines whether or not to generate the sequential question-and-answer data based on user input information, wherein the data generation unit generates the sequential question-and-answer data when the determination unit determines that the sequential question-and-answer data should be generated.

[0082] [9] The device according to [3], wherein the learning unit converts the sequential question-and-answer data into a structure and format that can be learned by the response generation AI model.

[0083]

[10] A method comprising: acquiring question and answer data having an answer sentence including a question sentence and a plurality of answer elements related to the content of the question sentence; and outputting, based on the question and answer data, prompts to be input to a question and answer generation AI model for generating one or more pieces of sequential question and answer data including at least some of the answer elements extracted from the answer sentence and question elements corresponding to the at least some of the answer elements.

[0084] The block diagrams used to explain the above embodiments show functional blocks. These functional blocks (components) are realized by any combination of hardware and / or software. Furthermore, the method for realizing each functional block is not particularly limited. That is, each functional block may be realized using a single device that is physically or logically coupled, or may be realized using two or more physically or logically separated devices that are connected directly or indirectly (e.g., via wire, wirelessly, etc.) and these multiple devices. The functional block may also be realized by combining the single device or multiple devices with software.

[0085] Functions include, but are not limited to, judgment, determination, assessment, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, consideration, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating, mapping, and assignment. For example, a functional block (component) that performs transmission is called a transmitting unit or transmitter. As mentioned above, there are no particular limitations on how these functions are implemented.

[0086] For example, the device 10 constituting the conversion system according to an embodiment of the present disclosure may function as a computer that performs processing of the control method of the present disclosure. FIG. 6 is a diagram illustrating an example of the hardware configuration of the device 10 according to an embodiment of the present disclosure. The device 10 described above may be physically configured as a computer including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like. The device 10 may be configured as a computer including at least one processor such as a CPU or a GPU, or may be configured as a computer including multiple processors or may include multiple computer devices. The terminal 5 and the server device 30 may also have a similar hardware configuration.

[0087] In the following description, the term "apparatus" can be interpreted as a circuit, a device, a unit, etc. The hardware configuration of apparatus 10 may be configured to include one or more of the apparatuses shown in the drawings, or may be configured to exclude some of the apparatuses.

[0088] Each function of the device 10 is realized by loading specified software (programs) onto hardware such as the processor 1001 and memory 1002, causing the processor 1001 to perform calculations, control communication via the communication device 1004, and control at least one of reading and writing data in the memory 1002 and storage 1003.

[0089] The processor 1001 controls the entire computer by running, for example, an operating system. The processor 1001 may be configured by a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, a register, etc. For example, the above-mentioned acquisition unit 11, determination unit 12, output unit 13, data generation unit 14, learning unit 15, etc. may be realized by the processor 1001.

[0090] The processor 1001 also reads programs (program codes), software modules, data, etc. from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes in accordance with the programs. The programs used are those that cause a computer to execute at least some of the operations described in the above-described embodiments. For example, the acquisition unit 11, the determination unit 12, the output unit 13, the data generation unit 14, and the learning unit 15 may be implemented by a control program stored in the memory 1002 and running on the processor 1001, and similar implementations may be used for other functional blocks. While the above-described various processes have been described as being executed by a single processor 1001, they may also be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The programs may also be transmitted from a network via a telecommunications line.

[0091] The memory 1002 is a computer-readable recording medium and may be configured, for example, by at least one of a read-only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a random access memory (RAM), etc. The memory 1002 may also be called a register, a cache, a main memory (primary storage device), etc. The memory 1002 can store executable programs (program codes), software modules, etc. for implementing a control method according to an embodiment of the present disclosure.

[0092] Storage 1003 is a computer-readable recording medium, and may be composed of at least one of, for example, an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy (registered trademark) disk, a magnetic strip, etc. Storage 1003 may also be referred to as an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, a server, or other appropriate medium including at least one of memory 1002 and storage 1003.

[0093] The communication device 1004 is hardware (transmission / reception device) for communicating between computers via at least one of a wired network and a wireless network, and is also referred to as, for example, a network device, a network controller, a network card, a communication module, etc. The communication device 1004 may be configured to include a high-frequency switch, a duplexer, a filter, a frequency synthesizer, etc. to realize at least one of frequency division duplex (FDD) and time division duplex (TDD). For example, the above-mentioned acquisition unit 11 and learning unit 15 may be realized by the communication device 1004.

[0094] The input device 1005 is an input device (e.g., a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that accepts input from the outside. The output device 1006 is an output device (e.g., a display, a speaker, an LED lamp, etc.) that outputs to the outside. Note that the input device 1005 and the output device 1006 may be integrated into one device (e.g., a touch panel).

[0095] Furthermore, each device, such as the processor 1001 and the memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or may be configured using different buses between each device.

[0096] The device 10 may also be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, the processor 1001 may be implemented using at least one of these pieces of hardware.

[0097] The notification of information is not limited to the aspects / embodiments described in the present disclosure and may be performed using other methods. For example, the notification of information may be performed by physical layer signaling (e.g., Downlink Control Information (DCI) and Uplink Control Information (UCI)), higher layer signaling (e.g., Radio Resource Control (RRC) signaling, Medium Access Control (MAC) signaling, broadcast information (Master Information Block (MIB) and System Information Block (SIB))), other signals, or a combination thereof. Furthermore, the RRC signaling may be referred to as an RRC message, and may be, for example, an RRC Connection Setup message, an RRC Connection Reconfiguration message, or the like.

[0098] The order of the procedures, sequences, flowcharts, etc. of each aspect / embodiment described in this disclosure may be changed unless it is consistent. For example, the methods described in this disclosure present elements of various steps using an example order, and are not limited to the particular order presented.

[0099] Input and output information may be stored in a specific location (for example, memory) or may be managed using a management table. Input and output information may be overwritten, updated, or added to. Output information may be deleted. Input information may be sent to another device.

[0100] The determination may be made based on a value represented by one bit (0 or 1), a Boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0101] The aspects / embodiments described in this disclosure may be used alone, in combination, or switched depending on the implementation. Notification of predetermined information (e.g., notification that "X is true") is not limited to explicit notification, but may be implicit (e.g., not notifying the predetermined information).

[0102] Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the spirit and scope of the present disclosure as defined by the claims. Therefore, the description of the present disclosure is intended to be illustrative and does not have any limiting meaning on the present disclosure.

[0103] Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0104] Software, instructions, information, etc. may also be transmitted or received over a transmission medium. For example, if software is transmitted from a website, server, or other remote source using wired technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and / or wireless technologies (such as infrared, microwave), then these wired and / or wireless technologies are included within the definition of transmission media.

[0105] The information, signals, etc. described in this disclosure may be represented using any of a variety of different technologies. For example, data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0106] Note that terms described in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings. For example, at least one of a channel and a symbol may be a signal (signaling). Furthermore, a signal may be a message. Furthermore, a component carrier (CC) may be called a carrier frequency, a cell, a frequency carrier, etc.

[0107] Furthermore, the information, parameters, etc. described in the present disclosure may be expressed using absolute values, may be expressed using relative values ​​from a predetermined value, or may be expressed using other corresponding information. For example, a radio resource may be indicated by an index.

[0108] The names used for the above-described parameters are not intended to be limiting in any way. Furthermore, the mathematical expressions using these parameters may differ from those explicitly disclosed in this disclosure. The various channels (e.g., PUCCH, PDCCH, etc.) and information elements may be identified by any suitable names, and therefore the various names assigned to these various channels and information elements are not intended to be limiting in any way.

[0109] In this disclosure, the terms "Mobile Station (MS)," "user terminal," "User Equipment (UE)," "terminal," and the like may be used interchangeably.

[0110] A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other suitable terminology.

[0111] As used in this disclosure, the terms "determining" and "determining" may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, inquiring (e.g., searching in a table, database, or other data structure), ascertaining, and the like. "Determining" and "determining" may also include receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in memory), and the like. Furthermore, "judgment" and "decision" can include regarding resolving, selecting, choosing, establishing, comparing, etc. as having been "judged" or "decided." In other words, "judgment" and "decision" can include regarding some action as having been "judged" or "decided." Furthermore, "judgment (decision)" can be interpreted as "assuming," "expecting," "considering," etc.

[0112] The terms "connected," "coupled," or any variation thereof, refer to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, "connected" may be read as "access." As used in this disclosure, two elements may be considered to be "connected" or "coupled" to each other using one or more wires, cables, and / or printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and optical (both visible and invisible) range, as some non-limiting and non-exhaustive examples.

[0113] As used in this disclosure, the phrase "based on" does not mean "based only on," unless expressly stated otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0114] As used in this disclosure, any reference to an element using a designation such as "first," "second," etc. does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient method of distinguishing between two or more elements. Thus, a reference to a first and a second element does not imply that only two elements may be employed or that the first element must in some way precede the second element.

[0115] When the terms "include," "including," and variations thereof are used in this disclosure, these terms are intended to be inclusive, similar to the term "comprising." Furthermore, when the term "or" is used in this disclosure, it is not intended to be an exclusive or.

[0116] In this disclosure, where articles are added by translation, such as a, an, and the in English, the disclosure may include that the nouns following these articles are in the plural form.

[0117] In the present disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "coupled" may also be interpreted in the same way as "different."

[0118] 1...processing system, 5...terminal, 10...device, 11...acquisition unit, 12...determination unit, 13...output unit, 14...data generation unit, 15...learning unit, 21...question and answer database, 22...prompt database, 23...sequential question and answer database, 30...server device, 31...generative AI model for questions and answers, 32...generative AI model for responses, P...prompt, QA1...question and answer data, QA2...sequential question and answer data.

Claims

1. An apparatus comprising: an acquisition unit that acquires question-and-answer data having a question sentence including a question element and an answer sentence including a plurality of answer elements related to the question element; and an output unit that outputs, based on the question-and-answer data, a prompt to be input to a question-and-answer generation AI model that generates one or more pieces of sequential question-and-answer data including at least some of the answer elements extracted from the answer sentence and question elements corresponding to the at least some of the answer elements.

2. The device according to claim 1, further comprising a data generation unit that inputs the prompts into the question and answer generation AI model to generate the sequential question and answer data.

3. The device according to claim 2, further comprising a learning unit that trains a response generation AI model that enables sequential responses based on the generated sequential question-and-answer data.

4. The device according to claim 1, wherein the output unit outputs the prompt based on the question and answer data and a predetermined template for the prompt.

5. The device according to claim 4, wherein the template includes usage information regarding usage patterns for the sequential question-and-answer data.

6. The device described in claim 5, wherein the acquisition unit acquires the usage mode information based on at least one of user input information and extracted information extracted from the question and answer data, and the output unit acquires the template of the prompt based on the acquired usage mode information.

7. The device according to claim 1, wherein the output unit outputs the prompt based on usage behavior information regarding usage behavior for the sequential question-and-answer data and the question-and-answer data.

8. The device according to claim 2, further comprising a determination unit that determines whether or not to generate the sequential question-and-answer data based on user input information, and the data generation unit generates the sequential question-and-answer data when the determination unit determines that the sequential question-and-answer data should be generated.

9. The device according to claim 3, wherein the learning unit converts the sequential question-and-answer data into a structure and format that can be learned by the response generation AI model.

10. A method comprising: acquiring question and answer data having an answer sentence including a question sentence and a plurality of answer elements related to the content of the question sentence; and outputting, based on the question and answer data, prompts to be input into a question and answer generation AI model for generating one or more pieces of sequential question and answer data including at least some of the answer elements extracted from the answer sentence and question elements corresponding to the at least some of the answer elements.