Information processing device, program, and method for returning an answer to a user-selected question

Abstract

An information processing device, program, and method are provided that can reduce the burden on a user of inputting a question as much as possible and instantly indicate the most appropriate answer to the question. [Solution] The information processing device has, as a registration stage, a question generation means for sending a first prompt including a document to be answered and an instruction statement instructing the generation of a question statement from the document to a language model and receiving multiple question statements; an answer generation means for sending a second prompt including the document, the question statement, and an instruction statement instructing the generation of an answer statement to the question statement to the language model and receiving the answer statements; a registration pair indication means for indicating to an operator multiple question-and-answer pairs consisting of the question statement and the answer statement; a registration pair selection means for allowing the operator to select multiple question-and-answer pairs to be registered; and a question-and-answer pair storage for registering the selected question-and-answer pairs.

Description

[Technical Field]

[0001] The present invention relates to technology for a FAQ (Frequently Asked Question) system that provides answers to questions from multiple users. [Background technology]

[0002] Conventional customer support FAQ systems include chatbots embedded in websites and apps. These FAQ systems can respond to a variety of questions regardless of the time of day, contributing to reduced service costs and improved customer satisfaction.

[0003] There is also a technology for FAQ systems that uses large-scale language models (LLMs) to infer answers to questions in natural language from users (see, for example, Patent Document 1). This technology determines whether the question itself is suitable for inputting into a large-scale language model to infer an answer. If it is determined to be inappropriate, a response that allows the user to delve deeper into the question is presented. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2025-060386 Summary of the Invention [Problem to be solved by the invention]

[0005] However, when an AI server for generating a large-scale language model is used in the background of an FAQ system, as in the technology of Patent Document 1, several issues arise. From the user's perspective, unless the question is entered as accurately as possible, the desired answer cannot be obtained. When using a large-scale language model, even if the user enters a question, they often have to re-enter the question depending on the answer, which places a burden on the user in entering the question. Furthermore, even if a user inputs an accurate question, it takes a relatively long time for the answer to be returned because a large-scale language model needs to make inferences. Users expect the answer to be output immediately after inputting a question. On the other hand, from the operator's perspective, because the large-scale language model allows free inference, there is a risk that an unexpected answer will be returned to a question. In particular, if a user enters a question that even the operator did not expect, there is a risk that an irrelevant answer will be returned.

[0006] Therefore, an object of the present invention is to provide an information processing device, program, and method that can reduce the burden on the user of inputting a question as much as possible and instantly display the most appropriate answer to the question. [Means for solving the problem]

[0007] According to the present invention, in an information processing device that searches for an answer sentence to a user's question sentence, At the registration stage, a question generation means for transmitting a first prompt including a document to be answered and an instruction for generating a question from the document to a language model, and for receiving a plurality of questions; an answer generation means for transmitting a second prompt including the document, the question, and an instruction to generate an answer to the question and to generate a citation of the document in the answer to the language model, and for receiving the answer; a registered pair indicating means for indicating to an operator a plurality of question and answer pairs each consisting of the question sentence, the answer sentence, and the quoted portion; a registration pair selection means for allowing an operator to select a plurality of question and answer pairs to be registered; a question and answer pair storage for registering the selected question and answer pairs; and In the operational stage, A search keyword acquisition means for acquiring a search keyword from a user; a question and answer pair search means for searching a plurality of question and answer pairs based on the correlation degree of each question and answer pair with respect to the search keyword by referring to the question and answer pair storage; Multiple questions based on multiple question-answer pairs are arranged in descending order of correlation. A question statement indicating means for indicating the question statement to a user; a desired question selection means for allowing a user to select a desired question; an answer statement indicating means for indicating to a user the answer statement and the quoted portion that form a question-and-answer pair for the desired question statement; The present invention is characterized by having the following.

[0009] According to another embodiment of the information processing device of the present invention, In the registration stage, for each user attribute designated in advance by the operator, the question generation means further includes a user attribute in the first prompt and instructs the user to generate a question taking the user attribute into consideration as an instruction; the answer sentence generation means further includes a user attribute in the second prompt and instructs the user to generate an answer sentence taking the user attribute into consideration as an instruction sentence; The question and answer pair storage registers the question and answer pairs for each user attribute, In the operational stage, The system further includes an attribute acquisition means for acquiring a user attribute of the user, The question statement indicating means indicates to the user only a plurality of questions based on a plurality of question and answer pairs based on the user attribute from the question and answer pair storage. It is also preferable.

[0010] According to another embodiment of the information processing device of the present invention, At the registration stage, the operator specifies persona attributes based on the creator of the document to be answered in advance, the question generation means further includes a persona attribute in the first prompt and instructs the user to generate a question taking the persona attribute into consideration as an instruction; The answer sentence generation means further includes a persona attribute in the second prompt and instructs the user to generate an answer sentence that takes the persona attribute into consideration as an instruction sentence. It is also preferable.

[0011] According to another embodiment of the information processing device of the present invention, Regarding the registration stage, the registered pair indicating means, when the operator selects one question and answer pair, calculates a correlation degree for each of the other question and answer pairs with the selected question and answer pair, and indicates to the operator the other multiple question and answer pairs in order of highest or lowest correlation degree; The registration pair selection means allows the operator to select a plurality of question and answer pairs to be registered. It is also preferable.

[0012] According to another embodiment of the information processing device of the present invention, Regarding the registration stage, The question generation means instructs the user to identify a plurality of topics from the context of the document to be answered and to generate a question for each topic. It is also preferable.

[0013] According to another embodiment of the information processing device of the present invention, The language model used in the question generation means and the language model used in the answer generation means may be the same or different. It is also preferable.

[0016] According to another embodiment of the information processing device of the present invention, In the operational stage, The question statement specifying means refers to the question and answer pair storage, searches for other multiple question and answer pairs that are similar to the question and answer pair of the desired question statement selected by the desired question statement selecting means under a predetermined condition, and further specifies the question statement in the other multiple question and answer pairs. It is also preferable.

[0017] According to another embodiment of the information processing device of the present invention, In the operational stage, The question statement indicating means calculates a correlation degree for each of the other question and answer pairs with respect to the question and answer pair of the desired question statement, and further searches for a plurality of other question and answer pairs whose correlation degree with the question and answer pair of the desired question statement is equal to or greater than a predetermined threshold as a predetermined condition, and indicates to the user the question statements in the other plurality of question and answer pairs in descending order of correlation degree. It is also preferable.

[0018] According to another embodiment of the information processing device of the present invention, the information processing device is a question and answer server that is accessed from terminals operated by an operator and a user, The language model is a generative AI server based on a pre-trained large-scale language model. It is also preferable.

[0019] According to the present invention, a program for causing a computer installed in a device that searches for answers to user questions to function includes: At the registration stage, a question generation means for transmitting a first prompt including a document to be answered and an instruction for generating a question from the document to a language model, and for receiving a plurality of questions; an answer generation means for transmitting a second prompt including the document, the question, and an instruction to generate an answer to the question and to generate a citation of the document in the answer to the language model, and for receiving the answer; a registered pair indicating means for indicating to an operator a plurality of question and answer pairs each consisting of the question sentence, the answer sentence, and the quoted portion; a registration pair selection means for allowing an operator to select a plurality of question and answer pairs to be registered; a question and answer pair storage for registering the selected question and answer pairs; and make it work, In the operational stage, A search keyword acquisition means for acquiring a search keyword from a user; a question and answer pair search means for searching a plurality of question and answer pairs based on the correlation degree of each question and answer pair with respect to the search keyword by referring to the question and answer pair storage; Multiple questions based on multiple question-answer pairs are arranged in descending order of correlation.A question statement indicating means for indicating the question statement to a user; a desired question selection means for allowing a user to select a desired question; an answer statement indicating means for indicating to a user the answer statement and the quoted portion that form a question-and-answer pair for the desired question statement; The present invention is characterized by the fact that it makes a computer function by using the above-mentioned method.

[0020] According to the present invention, there is provided a question answering method for a device that searches for answers to a user's question, comprising: The device is At the registration stage, a first step of sending a first prompt including a document to be answered and an instruction for generating a question from the document to a language model, and receiving a plurality of question sentences; a second step of sending a second prompt to a language model, the second prompt including the document, the question, and instructions for generating an answer to the question and a citation of the document in the answer, and receiving the answer; a third step of presenting to an operator a plurality of question-and-answer pairs each consisting of the question, the answer, and the quoted passage; a fourth step of having an operator select a plurality of question-answer pairs to be registered; a fifth step of registering the selected question-answer pair in a question-answer pair storage; Run In the operational stage, A sixth step of obtaining search keywords from the user; a seventh step of referring to the question and answer pair storage and searching for a plurality of question and answer pairs based on the correlation degree of each question and answer pair with respect to the search keyword; Multiple questions based on multiple question-answer pairs are arranged in descending order of correlation. The first step to be made clear to the user 8 Steps and The user is prompted to select a desired question. 9 Steps and The answer sentence and the citation that make up the question-answer pair for the desired question sentence are clearly indicated to the user. 10 Steps and The present invention is characterized by carrying out the following. [Effects of the Invention]

[0021] According to the information processing device, program, and method of the present invention, the burden on the user of inputting a question can be reduced as much as possible, and the most suitable answer to the question can be immediately displayed. [Brief explanation of the drawings]

[0022] [Figure 1] 1 is a diagram illustrating a system configuration including an information processing device according to the present invention. [Figure 2] FIG. 2 is a functional configuration diagram of the information processing device of the present invention at the registration stage; [Figure 3] FIG. 2 is an explanatory diagram of a question generation unit and an answer generation unit in the information processing device of the present invention. [Figure 4] 3 is an explanatory diagram of a registered pair specifying unit and a registered pair selecting unit in the information processing device of the present invention. FIG. [Figure 5] FIG. 10 is an explanatory diagram showing question and answer pairs in descending order of correlation in the registration pair selection unit of the present invention; [Figure 6] 10 is an explanatory diagram showing the registration pair selection unit of the present invention, which clearly shows question and answer pairs in descending order of correlation degree. FIG. [Figure 7] FIG. 2 is a functional configuration diagram of the information processing device of the present invention in an operation stage; [Figure 8] FIG. 1 is a first explanatory diagram of a user interface for the information processing device of the present invention. [Figure 9] FIG. 2 is a second explanatory diagram of a user interface for the information processing device of the present invention. [Figure 10] FIG. 10 is a diagram illustrating a function for collecting documents into a document storage. DETAILED DESCRIPTION OF THE INVENTION

[0023] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0024] FIG. 1 is a diagram showing the configuration of a system including an information processing device according to the present invention.

[0025] According to the system of Figure 1, the information processing device 1 of the present invention serves as a front end and communicates via a network between an operator-operated terminal 3 that manages questions and answers, and multiple user terminals 4 that send questions and receive answers. The information processing device 1 has a web interface as a website server, and communicates with an operator-operated terminal 3 and a user terminal 4 on which a browser and an application are installed. In other words, the information processing device 1 is a question and answer server that is accessed from the operator-operated terminal 3 and multiple user terminals 4. The operator operation terminal 3 manages questions and answers for "documents" to be answered via a user interface specific to the present invention to the information processing device 1. The user terminal 4 also transmits "questions" to the information processing device 1 via a user interface specific to the present invention and receives corresponding "answers."

[0026] An operator may be an information operations manager of a specific organization, such as a company, organization, or business. The operator is the person in charge of managing "documents to be answered" for users and publishing FAQs based on those documents. For example, the operator may publish a large number of "questions" for users about documents based on the company's products and services, and also disclose "answers" to the questions from users. Furthermore, users are people who need document-based FAQs, and may be an unspecified number of third parties, or a specified number of people (for example, employees who use an internal company system).

[0027] According to the system of FIG. 1, an information processing device 1 communicates with a generative artificial intelligence (AI) server 2, which serves as a backend and is used to create questions and answers, via a network. Generative AI Server 2 is based on pre-trained large language models (LLMs). Large language models are a type of deep learning model trained on large amounts of text data. They are based on neural network architectures such as Transformer and are capable of understanding context and generating natural language sentences. These are pre-trained with general linguistic knowledge from a large text corpus. Specifically, there are ChatGPT (registered trademark) from OpenAI, Claude (registered trademark) from Anthropic, Gemini (registered trademark) from Google, and Llama (registered trademark) from Meta, which are pre-trained with general linguistic knowledge from large text corpora.

[0028] The information processing device 1 of the present invention generates "question-answer pairs" consisting of anticipated questions and answers for "documents" to be answered in advance and stores them in storage. This allows multiple candidate questions to be immediately displayed in response to a user's access. When the user selects a desired question from these questions, the answer can be immediately displayed. In this way, the burden on the user of inputting a question is reduced as much as possible, and the most suitable answer to the question can be immediately displayed.

[0029] FIG. 2 is a functional configuration diagram of the information processing device of the present invention at the registration stage.

[0030] 2, the information processing device 1 of the present invention includes a document storage 100 and a question / answer pair storage 101. It also includes a question generation unit 111, an answer generation unit 112, a registration pair specification unit 113, and a registration pair selection unit 114 as a registration stage for registering a question and an answer in the question / answer pair storage 101. These functional components may be realized by executing a program that causes a computer installed in the information processing device to function. The processing flow of these functional components can also be understood as a question / answer method of the information processing device.

[0031] The document storage 100 stores a plurality of "documents" to be answered, which are received from the operator terminal 3. The document may be based on a wide variety of file formats, such as those listed below, but must contain at least text. pdf (registered trademark) doc (Word (registered trademark)) ppt (Powerpoint (registered trademark), graph data) xls (Excel (registered trademark), table data) csv (table text) txt (plain text) html (markup language text) xml (markup language text)

[0032] In addition, if the document includes an image, the image or the text that has been subjected to character recognition (Character Recognition / Reader) may be included in the document. Specifically, character recognition software that can be used includes open source software such as PaddleOCR (registered trademark) and Tesseract OCR (registered trademark), as well as cloud-based APIs (Application Programming Interfaces) from Google (registered trademark), Amazon (registered trademark), etc. These technologies allow for highly accurate text recognition by inputting a document.

[0033] The question and answer pair storage 101 registers question and answer pairs (question sentences and answer sentences) selected by an operator. The specific functions will be described later.

[0034] <Registration stage> [Question generation unit 111] The question generation unit 111 sends the first prompt to the generation AI server 2 (language model). The first prompt includes the following: First prompt = "Document" to be answered "Instruction statement" that instructs the generation of a question from the document Specifically, the first prompt is, for example, as follows: -------------------------------------------------------------------------------- Generate multiple questions for the user from the following #document. #document .... --------------------------------------------------------------------------------

[0035] As another first embodiment, the question generation unit 111 may specify, as an instruction, a plurality of topics from the context of the document to be answered, and instruct to generate a question for each topic. -------------------------------------------------------------------------------- Identify multiple topics from the context of the document below and generate a question for the user about each topic. #document .... -------------------------------------------------------------------------------- In this way, by inferring topics from documents in advance, multiple questions can be generated from the entire document without omission.

[0036] In addition, as another second embodiment, the question generation unit 111 may further include a "user attribute" in the first prompt and instruct the generation of a question that takes the user attribute into consideration as an instruction. -------------------------------------------------------------------------------- From the following #document, generate multiple questions for users with #user attributes. #User attributes: #document .... -------------------------------------------------------------------------------- In this way, even for the same document, multiple questions can be generated according to different user attributes.

[0037] The user attribute may be, for example, an age group specification such as "for teenagers" or "for people in their 70s and older," or a service attribute specification such as "for corporations" or "for individuals."

[0038] Furthermore, as another third embodiment, the operator may specify in advance a "persona attribute" based on the creator of the document to be answered. The question generation unit 111 further includes a persona attribute in the first prompt, and issues an instruction to generate a question that takes the persona attribute into consideration. -------------------------------------------------------------------------------- As someone with the #persona attribute, please generate multiple questions from the following #document to address to users with the #user attribute. #Persona attributes: #User attributes: #document .... -------------------------------------------------------------------------------- In this way, multiple questions can be generated depending on the position (persona attribute) of the document creator.

[0039] Persona attributes are attributes of the document creator, and may be attributes based on the organization or service, for example. For example, they may be company overviews or investor relations (IR) information. For example, in the case of documents related to telephone usage, there are various perspectives, such as those of a telephone company or a public facility, and questions can be generated according to each persona attribute.

[0040] The question generation unit 111 receives multiple questions by sending documents to be answered to the generation AI server 2.

[0041] FIG. 3 is an explanatory diagram of the question sentence generation unit and the answer sentence generation unit in the information processing device of the present invention.

[0042] 3, for example, the document is about telephone usage and is an explanatory document for the user. The question generation unit 111 sends a first prompt including the document to the generation AI server 2 and receives the generated multiple questions. -------------------------------------------------------------------------------- Question 1: "I would like to inquire with my telephone provider" Question 2: "I would like to inquire about telephone charges" ················ Question 9: "I can't make a call" --------------------------------------------------------------------------------

[0043] The received multiple question sentences are output to the answer sentence generation unit 112 and the registered pair specification unit 113.

[0044] [Answer sentence generation unit 112] The answer sentence generation unit 112 sends the second prompt to the generation AI server 2 (language model). The second prompt includes the following: Second prompt = document to be answered (same as first prompt) A plurality of questions (generated by the question generation unit 111) Instructions for generating an answer to a question Specifically, the second prompt is, for example, as follows: -------------------------------------------------------------------------------- Please generate answers to the user's questions for the following #document. #document .... #Question Question 1: "I would like to inquire with my telephone provider" Question 2: "I would like to inquire about telephone charges" ················ Question 9: "I can't make a call" --------------------------------------------------------------------------------

[0045] As another first embodiment, the answer sentence generation unit 112 may include an instruction sentence in the second prompt that, when generating an answer sentence to a question sentence, also generates a "quote passage" of the document for the answer sentence. -------------------------------------------------------------------------------- Please generate answers to the user's questions for the following #document. When doing so, please also generate a document citation (page number, line number range) for the generated answer. #document .... #Question Question 1: "I would like to inquire with my telephone provider" Question 2: "I would like to inquire about telephone charges" ················ Question 9: "I can't make a call" --------------------------------------------------------------------------------

[0046] As another second embodiment, the answer sentence generation unit 112 may further include a "user attribute" in the second prompt, similar to the question sentence generation unit 111. -------------------------------------------------------------------------------- From the following #document, generate an answer to a #question from a user with #user attributes. #User attributes: #document .... #Question Question 1: "I would like to inquire with my telephone provider" Question 2: "I would like to inquire about telephone charges" ················ Question 9: "I can't make a call" --------------------------------------------------------------------------------

[0047] Furthermore, as another third embodiment, the answer sentence generation unit 112 may further include a "persona attribute" in the second prompt, similar to the question sentence generation unit 111. -------------------------------------------------------------------------------- As someone with the #persona attribute, please use the following #document to generate an answer to a #question from a user with the #user attribute. #Persona attributes: #User attributes: #document .... #Question Question 1: "I would like to inquire with my telephone provider" Question 2: "I would like to inquire about telephone charges" ················ Question 9: "I can't make a call" --------------------------------------------------------------------------------

[0048] The above-mentioned user attributes and persona attributes may be added to the question and answer as hashtags, which allows the user attributes and persona attributes that generated the question and answer to be identified. In another embodiment, for example, if a persona attribute has already been assigned to a document, the same persona attribute may be automatically assigned to the question and answer sentences generated from that document.

[0049] Next, the answer sentence generation unit 112 receives an answer sentence for each question sentence from the generation AI server 2. For example, the answer sentence generation unit 112 acquires a plurality of answer sentences (including, for example, quoted passages) as follows: -------------------------------------------------------------------------------- Answer text Answer 1: "Phone calls are..., emails are..." :Quoted passage [Page 2, lines 2-10] Answer 2: "The telephone charges website is..." :Quoted passage [Page 5, lines 8-15] ················ Answer 9: "For inquiries regarding malfunctions..." :Quoted passage [Page 3, lines 5-12] -------------------------------------------------------------------------------- As described above, all of the questions and answers to the documents to be answered can be automatically constructed by the question generation unit 111 and the answer generation unit 104.

[0050] The generation AI server (language model) used by the question generation unit 111 and the generation AI server (language model) used by the answer generation unit 112 may be the same or different. By using different generation AI servers, answer sentences based on different perspectives may be obtained even for the same question sentence.

[0051] However, according to the present invention, the question and answer sentences are not generated all at once (simultaneously) by a single generation AI server. A feature of the present invention is that it requires two steps: first, generate a question sentence from the document to be answered, and then generate an answer sentence from that question sentence. This makes it possible to generate an answer sentence from a different perspective than the question sentence, which also leads to improved quality of the answer sentence.

[0052] FIG. 4 is an explanatory diagram of the registered pair specifying unit and registered pair selecting unit in the information processing device of the present invention.

[0053] [Registered Pair Explained Section 113] The registered pair revealing unit 113 reveals a plurality of "question and answer pairs" consisting of question sentences and answer sentences to the operator operation terminal 3. This allows the operator to recognize what kind of question sentences and answer sentences have been automatically generated for the document specified by the operator.

[0054] [Registered pair selection unit 114] The registration pair selection unit 114 allows the operator to select "plurality of question and answer pairs" to be registered, thereby eliminating unnecessary question sentences and answer sentences from the automatically generated question sentences and answer sentences.

[0055] In another embodiment, the registered pair specifying unit 113 and the registered pair selecting unit 114 may be mutually and repeatedly executed. That is, every time the registered pair selecting unit 114 selects a question-and-answer pair, the registered pair specifying unit 113 rearranges the question-and-answer pairs to be specified. When an operator selects one question-and-answer pair, the registered pair selection unit 114 calculates the "degree of correlation" for each of the other question-and-answer pairs with respect to the selected question-and-answer pair, and displays the other multiple question-and-answer pairs to the operator in order of "highest" or "lowest" correlation degree.

[0056] The correlation may be, for example, semantic ranking. Unlike keyword search, this uses technologies such as natural language processing and machine learning to understand the meaning, context, and search intent of the question sentence in the search query, and sorts the text (question sentence) in order of relevance. In particular, it inherits the advantages of fuzzy search and uses syntactic analysis / semantic analysis to rank the search results closest to the meaning of the search sentence. Specific examples include Microsoft's Azure OpenAI (registered trademark) and Google's BERT (Bidirectional Encoder Representations from Transformers).

[0057] FIG. 5 is an explanatory diagram showing the question and answer pairs in descending order of correlation degree in the registration pair selection unit of the present invention. (S51) First, it is assumed that question and answer pair 2 is selected by the operator. (S52) In response to this, the correlation degree is calculated for each of the other question and answer pairs with respect to the selected question and answer pair 2, and the other multiple question and answer pairs are sorted in descending order of correlation degree. (S53) Next, it is assumed that question and answer pair 4 is selected by the operator. (S54) In response to this, the correlation degree is calculated for each of the other question and answer pairs with respect to the selected question and answer pair 4, and the other multiple question and answer pairs are sorted in descending order of correlation degree. This allows the operator to intuitively visually recognize other question and answer pairs that have a closer correlation distance to the question and answer pair 2 that he or she has selected.

[0058] FIG. 6 is an explanatory diagram showing the question and answer pairs in ascending order of correlation degree in the registration pair selection unit of the present invention. (S61) First, it is assumed that question and answer pair 2 is selected by the operator. (S62) In response to this, the degree of correlation is calculated for each of the other question and answer pairs with respect to the selected question and answer pair 2, and the other multiple question and answer pairs are rearranged in order of "lowest correlation degree." (S63) Next, assume that question and answer pair 4 is selected by the operator. (S64) In response to this, the correlation degree is calculated for each of the other question and answer pairs with respect to the selected question and answer pair 4, and the other multiple question and answer pairs are sorted in order of "lowest correlation degree." This allows the operator to intuitively visually recognize question and answer pairs that have a greater distance in correlation than the question and answer pair 2 that he or she has selected.

[0059] [Question and Answer Pair Storage 101] The question and answer pair storage 101 registers the "question and answer pairs" selected by the operator. The question and answer pair storage 101 may also register the "cited passage" of the document along with the selected question and answer pair. When questions and answers are generated for the same document based on a plurality of different user attributes, the question and answer pair storage 101 registers a question and answer pair for each user attribute. By including the question and answer pair storage 101, the information processing device 1 of the present invention can instantly indicate the most suitable answer sentence to the desired question sentence selected by the user.

[0060] <Operational Phase> FIG. 7 is a functional configuration diagram of the information processing device of the present invention in the operation stage.

[0061] 7, the information processing device 1 of the present invention, in its operational stage, further includes a search keyword acquisition unit 121, a user attribute acquisition unit 122, a question-and-answer pair search unit 123, a question statement display unit 124, a desired question statement selection unit 125, an answer statement display unit 126, and a citation section display unit 127. These functional components may be realized by executing a program that causes a computer installed in the information processing device to function. The processing flow of these functional components can also be understood as a question-and-answer method of the information processing device.

[0062] [Search keyword acquisition unit 121] The search keyword acquisition unit 121 acquires a "search keyword" from the user. The search keyword received from the user terminal 4 is a "word" or "sentence" that the user himself / herself considers to be related to the question. Of course, the user's speech may be picked up by a microphone and converted into text by voice recognition. According to the present invention, it is not essential that the user inputs a search keyword, but rather the question statement display unit 124 (described later) may display all or some of the questions to the user.

[0063] [User attribute acquisition unit 122] The user attribute acquisition unit 122 acquires "user attributes" from the user. The user attributes received from the user terminal 4 represent the attributes of the user who asks the question. As mentioned above, it may be an age group such as "teens" or "over 70s." This allows the user to view questions appropriate for their own age group and obtain answers appropriate for that age group. It should be noted that, for example, a plurality of user attributes may be displayed on the user interface, allowing the user to select from among them.

[0064] [Question and Answer Pair Search Section 123] The question and answer pair search unit 123 refers to the question and answer pair storage 101 and searches for "plurality of question and answer pairs (question sentences and answer sentences)" that are related to the search keyword under a predetermined condition. Here, the predetermined condition may be based on the "degree of correlation" calculated for each question and answer pair for the search keyword. Then, the degree of correlation may be sorted in descending order to derive a predetermined number of question and answer pairs. Furthermore, if a user attribute is specified, only a plurality of question and answer pairs based on the specified user attribute are searched for. For example, a hash tag of the user attribute may be attached to each question and answer pair.

[0065] [Question Civilization Section 124] The question statement presentation unit 124 transmits to the user terminal 4 "plurality of question statements" based on the plurality of question and answer pairs registered in the question and answer pair storage 101, and presents the "plurality of question statements" to the user. The plurality of question statements presented to the user are based on the plurality of question and answer pairs searched by the question and answer pair search unit 123.

[0066] [Desired Question Sentence Selection Unit 125] The desired question selection unit 125 allows the user to select one "desired question" from among a plurality of questions displayed on the user terminal 4. The desired question has already been associated with an answer from among the plurality of question-and-answer pairs searched for by the question-and-answer pair search unit 123.

[0067] [Answer Civilization Section 126] The answer statement presentation unit 126 transmits an "answer statement" that forms a question-and-answer pair to the desired question statement selected by the user to the user terminal 4 and presents it to the user. A feature of the present invention is that, in response to a user's desired question, the answer can be instantly output from among the question and answer pairs already searched for by the question and answer pair search unit 123, without the need for inference by the generation AI server.

[0068] In another embodiment, the answer statement display unit 126 allows the user to select and display the "quote" of the document for the answer statement along with the answer statement. For example, a "document quote display button" is displayed to the user along with the answer statement. The user taps the "document quote display button" displayed on the user interface to request the display of the quote.

[0069] [Citation 127] When the user selects to display the quoted portion of the document, the citation indicating unit 127 indicates the quoted portion of the document. The page number and line number of the document are assigned to the question-and-answer pair corresponding to the answer sentence as the quoted portion. The citation indicator 127 displays a document based on the response text on a user interface. If the document is, for example, a PDF file, the PDF application is launched and the document is displayed. Next, the citation indicator 127 displays the page numbers and line numbers of the quoted parts of the document. This allows the user to visually recognize the answer to the desired question sentence that he or she has selected, as well as the citations of the documents that serve as the basis for the answer.

[0070] FIG. 8 is a first explanatory diagram of a user interface for the information processing device of the present invention.

[0071] According to FIG. 8(a), it is assumed that the search keyword "telephone" is input to the display screen of the user interface of the user terminal 4. 7, question and answer pairs 1, 2, 4, 9 (excluding 6) whose correlation with the search keyword is equal to or greater than a predetermined threshold are searched for, and questions 1, 2, 4, 9 based on these pairs are displayed. Questions 1, 2, 4, 9 in question and answer pairs 1, 2, 4, 9 are then sorted in descending order of correlation and displayed to the user. Question 1: "I would like to inquire with my telephone provider" Question 2: "I would like to inquire about telephone charges" Question 4: "My smartphone broke." Question 9: "I can't make a call" Here, suppose the user selects question sentence 4, "My smartphone broke."

[0072] According to FIG. 8(b), answer sentence 4 is clearly displayed to the user from question and answer pair 4 corresponding to question sentence 4 "My smartphone broke down." At this time, the cited section of the document that forms the basis of Answer 4, "AAA.pdf (page 5)", can also be displayed. Furthermore, the question statement revealing unit 124 refers to the question and answer pair storage 101 to search for other question and answer pairs 1, 2, and 9 that are similar in accordance with a predetermined condition to the question and answer pair 4 of the desired question statement 4 selected by the desired question statement selecting unit 125. Then, the question statement revealing unit 124 further reveals the questions 1, 2, and 9 in the other question and answer pairs 1, 2, and 9 to the user. At this time, the question statement revealing unit 124 calculates the "degree of correlation" of each of the other question and answer pairs 1, 2, and 9 with question and answer pair 4 of the desired question statement 4. Then, as a predetermined condition, the unit 124 further searches for a plurality of other question and answer pairs 1, 2, and 9 whose degree of correlation with question and answer pair 4 of the desired question statement is equal to or greater than a predetermined threshold. Then, the unit 124 sorts the questions 1, 2, and 9 in the other plurality of question and answer pairs 1, 2, and 9 in descending order of degree of correlation and reveals them to the user. Question 9: "I can't make a call" Question 2: "I would like to inquire about telephone charges" Question 1: "I would like to inquire with my telephone provider" Here, it is assumed that the user selects question sentence 9 "I can't make a call."

[0073] According to FIG. 8(c), answer sentence 9 is clearly displayed to the user from question and answer pair 9 corresponding to question sentence 9 "I can't make a call." At this time, the cited section of the document that forms the basis of Answer 9, "AAA.pdf (page 3)," can also be displayed. The question statement revealing unit 124 also refers to the question and answer pair storage 101 to search for a plurality of other question and answer pairs 1, 2 that are similar in accordance with a predetermined condition to the question and answer pair 9 of the desired question statement 9 selected by the desired question statement selecting unit 125. Then, the question statement revealing unit 124 further reveals the question statements 1, 2 in the other plurality of question and answer pairs 1, 2 to the user. At this time, the question sentence revealing unit 124 calculates the "degree of correlation" of each of the other question and answer pairs 1, 2 with the question and answer pair 9 of the desired question sentence 9. Then, as a predetermined condition, the unit 124 further searches for a plurality of other question and answer pairs 1, 2 whose degree of correlation with the question and answer pair 9 of the desired question sentence is equal to or greater than a predetermined threshold. Then, the unit 124 sorts the question sentences 1, 2 in the other plurality of question and answer pairs 1, 2 in descending order of degree of correlation and reveals the sorted results to the user. Question 1: "I would like to inquire with my telephone provider" Question 2: "I would like to inquire about telephone charges"

[0074] FIG. 9 is a second explanatory diagram of a user interface for the information processing device of the present invention.

[0075] According to FIG. 9(a), it is assumed that the user has selected (tapped) the citation "AAA.pdf (page 3)" of the document that is the basis of answer sentence 9. 9(b), the citation indication unit 127 searches the document storage 100 for document AAA.pdf based on answer sentence 9, transmits the document to the user terminal 4, and indicates it to the user. The user can view answer sentence 9 to question sentence 9, "I can't make a call," and can also see page 3 in document AAA.pdf, which is the citation.

[0076] FIG. 10 is a diagram illustrating the function of collecting documents into the document storage.

[0077] <Storage in document storage> According to FIG. 10, the operator can specify a file path of the local storage to the information processing device 1. By specifying a file path, the operator can collectively upload multiple documents under that path to the information processing device 1 and store them in the document storage 10. Alternatively, the operator can display indexes of multiple files under a specific folder on the user terminal 4 and select some or all of the indexes of those files. For example, multiple files can be collectively uploaded and registered to the information processing device 1 at the same time by using "cut and paste" or "drag and drop."

[0078] The operator may also specify to the information processing device 1 the URL (Uniform Resource Locator) of a website on the Internet as the location of the document. 10, the information processing device 1 includes a scraping unit. When a URL or a file path is specified by an operator, the scraping unit references a website on the Internet. Then, the scraping unit downloads a document from the URL or the file path and stores it in a document storage.

[0079] Furthermore, the operator can specify documents by adding a wildcard "*" to the URL or file path they specify. In this case, the scraping unit will download multiple documents simultaneously and store them in the document storage.

[0080] Furthermore, when a URL or file path is specified by the operator, it is also preferable to periodically crawl that URL or file path, download updated documents (site crawling), and update the document storage.

[0081] <Adding tags to documents> In another embodiment, a persona attribute designated by an operator may be assigned as a hash tag for each document stored in the document storage. This identifies organizational attributes or personal attributes based on the document creator. The user can assign a hash tag after registering a document, or can assign a hash tag simultaneously with registering a document. The persona attribute as a hash tag is used as the persona attribute in the registration stage described above when generating questions and answers.

[0082] As described above in detail, the information processing device, program, and method of the present invention reduce the burden on the user of inputting a question as much as possible, and can instantly display the most appropriate answer to the question.

[0083] With respect to the various embodiments of the present invention described above, various changes, modifications, and omissions that fall within the scope of the technical spirit and aspects of the present invention may be easily made by those skilled in the art. The above description is merely illustrative and is not intended to be limiting in any way. The present invention is limited only by the claims and their equivalents. [Explanation of symbols]

[0084] 1. Information processing equipment 100 document storage 101 Questions and Answers Paired Storage 111 Question generation section 112 Answer sentence generation section 113 Registered Pair Explained 114 Registration Pair Selection Unit 121 Search Keyword Acquisition Unit 122 User attribute acquisition unit 123 Question and Answer Pair Search 124 Question Civilization Department 125 Desired Question Selection Section 126 Answer Civilization Department 127 Citation section 2 Generative AI Server 3 Operator Control Terminal 4. User terminal

Claims

1. In an information processing device that searches for an answer to a user's question, At the registration stage, a question generation means for transmitting a first prompt including a document to be answered and an instruction for generating a question from the document to a language model, and for receiving a plurality of questions; an answer generation means for transmitting a second prompt including the document, the question, and an instruction to generate an answer to the question and to generate a citation of the document in the answer to the language model, and for receiving the answer; a registered pair indicating means for indicating to an operator a plurality of question and answer pairs each consisting of the question sentence, the answer sentence, and the quoted portion; a registration pair selection means for allowing an operator to select a plurality of question and answer pairs to be registered; a question and answer pair storage for registering the selected question and answer pairs; and In the operational stage, A search keyword acquisition means for acquiring a search keyword from a user; a question and answer pair search means for searching a plurality of question and answer pairs based on the correlation degree of each question and answer pair with respect to the search keyword by referring to the question and answer pair storage; a question statement indicating means for arranging a plurality of question statements based on the plurality of question and answer pairs in descending order of the correlation degree and indicating the plurality of question statements to a user; a desired question selection means for allowing a user to select a desired question; an answer statement indicating means for indicating to a user the answer statement and the quoted portion that form a question-and-answer pair for the desired question statement; An information processing device comprising:

2. In the registration stage, for each user attribute designated in advance by the operator, the question generation means further includes a user attribute in the first prompt and instructs the user to generate a question taking the user attribute into consideration as an instruction; the answer sentence generation means further includes a user attribute in the second prompt and instructs the user to generate an answer sentence taking the user attribute into consideration as an instruction sentence; The question and answer pair storage registers the question and answer pairs for each user attribute, In the operational stage, The system further includes an attribute acquisition means for acquiring a user attribute of the user, The question statement indicating means indicates to the user only a plurality of questions based on a plurality of question and answer pairs based on the user attribute from the question and answer pair storage.

2. The information processing apparatus according to claim 1, wherein:

3. At the registration stage, the operator specifies persona attributes based on the creator of the document to be answered in advance, the question generation means further includes a persona attribute in the first prompt and instructs the user to generate a question taking the persona attribute into consideration as an instruction; The answer sentence generating means further includes a persona attribute in the second prompt and instructs the user to generate an answer sentence that takes the persona attribute into consideration as an instruction sentence.

2. The information processing apparatus according to claim 1, wherein:

4. Regarding the registration stage, the registered pair indicating means, when the operator selects one question-and-answer pair, calculates a correlation degree for each of the other question-and-answer pairs with the selected question-and-answer pair, and indicates to the operator the other multiple question-and-answer pairs in order of highest or lowest correlation degree; The registration pair selection means allows the operator to select a plurality of question and answer pairs to be registered.

2. The information processing apparatus according to claim 1, wherein:

5. Regarding the registration stage, The question generation means instructs the user to identify a plurality of topics from the context of the document to be answered and to generate a question for each topic.

2. The information processing apparatus according to claim 1, wherein:

6. The language model used in the question generation means and the language model used in the answer generation means may be the same or different.

2. The information processing apparatus according to claim 1, wherein:

7. In the operational stage, The question statement specifying means refers to the question and answer pair storage, searches for other multiple question and answer pairs that are similar to the question and answer pair of the desired question statement selected by the desired question statement selecting means under a predetermined condition, and further specifies the question statement in the other multiple question and answer pairs.

2. The information processing apparatus according to claim 1, wherein:

8. In the operational stage, The question statement indicating means calculates a correlation degree for each of the other question and answer pairs with respect to the question and answer pair of the desired question statement, and further searches for a plurality of other question and answer pairs whose correlation degree with the question and answer pair of the desired question statement is equal to or greater than a predetermined threshold as a predetermined condition, and indicates to the user the question statements in the other plurality of question and answer pairs in descending order of correlation degree.

2. The information processing apparatus according to claim 1, wherein:

9. the information processing device is a question and answer server that is accessed from terminals operated by an operator and a user, The language model is a generative AI server based on a pre-trained large-scale language model.

2. The information processing apparatus according to claim 1, wherein:

10. A program that causes a computer installed in a device to search for answers to user questions to function, At the registration stage, a question generation means for transmitting a first prompt including a document to be answered and an instruction for generating a question from the document to a language model, and for receiving a plurality of questions; an answer generation means for transmitting a second prompt including the document, the question, and an instruction to generate an answer to the question and to generate a citation of the document in the answer to the language model, and for receiving the answer; a registered pair indicating means for indicating to an operator a plurality of question and answer pairs each consisting of the question sentence, the answer sentence, and the quoted portion; a registration pair selection means for allowing an operator to select a plurality of question and answer pairs to be registered; a question and answer pair storage for registering the selected question and answer pairs; and make it work, In the operational stage, A search keyword acquisition means for acquiring a search keyword from a user; a question and answer pair search means for searching a plurality of question and answer pairs based on the correlation degree of each question and answer pair with respect to the search keyword by referring to the question and answer pair storage; a question statement indicating means for arranging a plurality of question statements based on the plurality of question and answer pairs in descending order of the correlation degree and indicating the plurality of question statements to a user; a desired question selection means for allowing a user to select a desired question; an answer statement indicating means for indicating to a user the answer statement and the quoted portion that form a question-and-answer pair for the desired question statement; A program that causes a computer to function.

11. A question answering method for a device that searches for an answer to a user's question, comprising: The device is At the registration stage, a first step of sending a first prompt including a document to be answered and an instruction for generating a question from the document to a language model, and receiving a plurality of question sentences; a second step of sending a second prompt including the document, the question, and an instruction to generate an answer to the question and a citation of the document in the answer to the language model, and receiving the answer; a third step of presenting to an operator a plurality of question-and-answer pairs each consisting of the question, the answer, and the quoted passage; a fourth step of having an operator select a plurality of question-answer pairs to be registered; a fifth step of registering the selected question and answer pairs in a question and answer pair storage; Run In the operational stage, a sixth step of obtaining search keywords from the user; a seventh step of searching for a plurality of question and answer pairs based on the correlation degree of each question and answer pair with respect to the search keyword by referring to the question and answer pair storage; an eighth step of presenting to the user a plurality of question sentences based on the plurality of question-answer pairs in descending order of the correlation degree; a ninth step of having the user select a desired question; a tenth step of indicating to the user the answer sentences and citations that form question-answer pairs for the desired question sentence; A question answering method comprising:

Images