Information processing device, program, and method for receiving document-based questions.

The system addresses user burden and response latency in FAQ systems by pre-registering question-answer pairs and using display keywords for immediate, relevant answers.

JP7953844B1Active Publication Date: 2026-11-19DEFIDE CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
JP2026121387
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2026-06-29
Publication Date
2026-11-19
Estimated Expiration
2046-06-29

AI Technical Summary

Technical Problem

Existing FAQ systems using large language models impose a burden on users by requiring accurate question input and long waiting times for answers, and risk returning irrelevant responses due to free inference.

Method used

A system that pre-generates and registers question-answer pairs, allowing operators to select relevant pairs, infers display keywords, and enables users to select keywords for immediate display of appropriate answers.

Benefits of technology

Reduces user input burden and provides immediate, relevant answers by pre-registering question-answer pairs and using display keywords for quick retrieval.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007953844000001_ABST
    Figure 0007953844000001_ABST
Patent Text Reader

Abstract

The present invention provides an information processing device that minimizes the burden on users in inputting question texts and immediately displays the most appropriate answer to those questions. [Solution] The system includes: a question-answer pair generation means that sends a first prompt to a language model, which includes an instruction to generate question-answer pairs from multiple documents to be answered, and receives multiple question-answer pairs; a question-answer pair selection means that displays the multiple question-answer pairs on an operator terminal, allows the operator to select the question-answer pairs to be made public, and registers the selected question-answer pairs in a question-answer pair storage; and a display keyword generation means that sends a second prompt to a language model, which includes a group of question-answer pairs in the question-answer pair storage and an instruction to infer a predetermined number of display keywords that the user is likely to recall as search keys from the group of question-answer pairs, and receives multiple display keywords.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to technology of FAQ (Frequently Asked Question) systems that return response texts to question texts from a plurality of users.

Background Art

[0002] Conventionally, as a customer support FAQ system, there is a technology such as a chatbot incorporated into a website or application. Such a FAQ system can respond to various question texts regardless of time zone, and contributes to service cost reduction and customer satisfaction improvement.

[0003] Additionally, there is also technology of a FAQ system that infers response texts using Large Language Models (LLM) for natural language question texts from users (see, for example, Patent Document 1). According to this technology, it is determined whether the question text itself is suitable for input to the large language model to infer a response text. When it is determined that the question text is not suitable, a response text for digging deeper into the question text is presented to the user.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, when a generative AI server for large language models as in the technology of Patent Document 1 is used in the background of the FAQ system, several problems arise. From the user's perspective, they must input the question as accurately as possible in order to obtain the desired answer. When using large-scale language models, even if the user inputs a question, they often have to re-enter the question depending on the answer they receive, which places a burden on the user in terms of question input. Furthermore, from the user's perspective, even if they input an accurate question, there is a relatively long waiting time before the answer is returned while the large-scale language model performs its inference. Users expect the answer to be output immediately after they input their question. On the other hand, from the perspective of the operators, because the system is freely inferred by a large-scale language model, there is a risk that unexpected answers may be returned in response to a question. In particular, if a user enters a question that even the operators did not anticipate, there is a risk of sending an irrelevant answer.

[0006] Therefore, the present invention aims to provide an information processing device, program, and method that minimizes the burden on the user in inputting question texts and immediately displays the most appropriate answer text for those questions. [Means for solving the problem]

[0007] According to the present invention, in an information processing device that receives question sentences based on multiple documents to be answered, As part of the registration process, A question-answer pair generation means sends a first prompt to a language model that includes multiple documents to be answered and instructions to generate one or more question-answer pairs of question sentences and answer sentences from the multiple documents, and receives multiple question-answer pairs from the language model. A question-answer pair selection means displays the multiple question-answer pairs on the operator's terminal, allows the operator to select the question-answer pairs to be made public, and registers the selected question-answer pairs in the question-answer pair storage. The system infers a group of question-answer pairs from the question-answer pair storage, and a predetermined number of display keywords that a user is likely to recall as search keys from the group of question-answer pairs. In addition, for each inferred display keyword, the correlation between it and multiple question-answer pairs is inferred, and a predetermined number of question-answer pairs are linked in order of the highest correlation.A second prompt, including a "should" instruction, is sent to the language model, and the language model then... 、 Multiple display keywords Along with, Each display keyword A set of linked question-answer pairs. A means for receiving and displaying keywords and It has, In the operational phase, A display keyword selection means that displays a question input field and multiple display keywords on the user terminal, and when the user selects a display keyword, that display keyword is entered into the input field. A question-answer pair search means that searches for a predetermined number of question-answer pairs already associated with the display keyword entered in the input field of the question text, A question text expliciting means that explicitly displays multiple question texts to the user terminal based on multiple question-answer pairs that have been searched. It is characterized by having the following features.

[0012] According to another embodiment of the information processing apparatus of the present invention, In the operational phase, A means for selecting a desired question from among the multiple question sentences, A means for explicitly displaying the answer text of a question-answer pair for the desired question text to the user terminal, and It is also preferable to close it.

[0013] According to another embodiment of the information processing apparatus of the present invention, The language model used in the question-answer pair generation means and the language model used in the display keyword generation means may be the same or different. That is also desirable.

[0014] According to another embodiment of the information processing apparatus of the present invention, The information processing device is a question answering server accessed from terminals operated by operators and users. The language model is a generative AI server based on a pre-trained, large-scale language model. That is also desirable.

[0015] According to the present invention, in a program for causing a computer mounted in an apparatus that accepts question sentences based on a plurality of documents to be answered to function, as a registration step, a question-and-answer pair generation means for transmitting a first prompt including a plurality of documents to be answered and an instruction to generate one or more question-and-answer pairs of question sentences and answer sentences from the plurality of documents to a language model, and receiving a plurality of question-and-answer pairs from the language model, a question-and-answer pair selection means for displaying the plurality of question-and-answer pairs on an operator operation terminal, allowing the operator to select a question-and-answer pair to be made public, and registering the selected question-and-answer pair in a question-and-answer pair storage, inferring a group of question-and-answer pairs in the question-and-answer pair storage and a predetermined number of display keywords that the user is highly likely to recall as search keys from the group of question-and-answer pairs In addition, for each inferred display keyword, the correlation between it and multiple question-answer pairs is inferred, and a predetermined number of question-answer pairs are linked in order of the highest correlation. transmitting a second prompt including the instruction to do so to the language model, and from the language model 、 a plurality of display keywords Along with, for each display keyword A set of linked question-answer pairs. a display keyword generation means for receiving, and to make the computer work, as an operation step, a display keyword selection means for displaying an input field for question sentences and the plurality of display keywords on a user terminal, and inputting the display keyword into the input field when the user selects any display keyword, A question-answer pair search means that searches for a predetermined number of question-answer pairs already associated with the display keyword entered in the input field of the question text, A question text expliciting means that explicitly displays multiple question texts to the user terminal based on multiple question-answer pairs that have been searched. to cause the computer to function.

[0016] According to the present invention, in a question accepting method for an apparatus that accepts question sentences based on a plurality of documents to be answered, The device is as a registration step, A first step involves sending a first prompt to a language model that includes multiple documents to be answered and instructions that the language model should generate one or more question-answer pairs of question sentences and answer sentences from those multiple documents, and receiving multiple question-answer pairs from the language model. The second step involves displaying the multiple question-answer pairs on the operator's terminal, allowing the operator to select the question-answer pairs to be made public, and registering the selected question-answer pairs in the question-answer pair storage. The system infers a group of question-answer pairs from the question-answer pair storage, and a predetermined number of display keywords that a user is likely to recall as search keys from the group of question-answer pairs. In addition, for each inferred display keyword, the correlation between it and multiple question-answer pairs is inferred, and a predetermined number of question-answer pairs are linked in order of the highest correlation. A second prompt, including a "should" instruction, is sent to the language model, and the language model then... 、 Multiple display keywords Along with, Each display keyword A set of linked question-answer pairs. The third step is to receive, In the operational phase, A fourth step involves displaying a question input field and the multiple display keywords on the user's terminal, and when the user selects any of the display keywords, those keywords are entered into the input field. The fifth step involves searching for a predetermined number of question-answer pairs already associated with the display keywords entered in the question input field, The sixth step is to explicitly display multiple question sentences based on the multiple question-answer pairs found to the user terminal. It is characterized by performing the following. [Effects of the Invention]

[0017] According to the information processing device, program, and method of the present invention, the burden on the user to input a question is reduced as much as possible, and the optimal answer to that question can be immediately displayed. [Brief explanation of the drawing]

[0018] [Figure 1] This is a system configuration diagram including the information processing device according to the present invention. [Figure 2] This is a functional configuration diagram of the registration stage in the information processing device of the present invention. [Figure 3]This is an explanatory diagram of the question-answer pair generation unit in the information processing device of the present invention. [Figure 4] This is an explanatory diagram of the question-answer pair selection unit in the information processing device of the present invention. [Figure 5] This is a first explanatory diagram of the display keyword generation unit in the information processing device of the present invention. [Figure 6] This is a second explanatory diagram of the display keyword generation unit in the information processing device of the present invention. [Figure 7] This is an example of a user interface for registering desired keywords. [Figure 8] This is a functional configuration diagram of the information processing device of the present invention during its operational phase. [Figure 9] This is an explanatory diagram of the user interface for the information processing device of the present invention. [Figure 10] This is a diagram illustrating the function of collecting documents into document storage. [Modes for carrying out the invention]

[0019] Embodiments of the present invention will be described in detail below with reference to the drawings.

[0020] Figure 1 is a system configuration diagram including the information processing device according to the present invention.

[0021] According to the system in Figure 1, the information processing device 1 receives question statements based on multiple documents to be answered. Specifically, the information processing device 1 is a question answering server accessed from terminals operated by operators and users. The information processing device 1, acting as a front-end, communicates via a network between the operator terminal 3, which manages question and answer texts, and multiple user terminals 4, which send question texts and obtain answer texts. Furthermore, the information processing device 1 is equipped with a web interface that functions as a website server and communicates with operator terminals 3 and user terminals 4 on which browsers and applications are installed. In other words, the information processing device 1 is a question answering server accessed by operator terminals 3 and multiple user terminals 4. The operator terminal 3 manages question-and-answer pairs (question text and answer text) for the "document" to be answered, via a user interface unique to the present invention, and transmits them to the information processing device 1. The user terminal 4 also transmits "question texts" to the information processing device 1 and receives corresponding "answer texts" via a user interface unique to the present invention.

[0022] An operator may be an information operations manager of a specific organization, such as a company, organization, or business. An operator is responsible for managing "documents that can be answered" for users and for publishing FAQs based on those documents. For example, regarding documents related to the company's products and services, the operator may publish numerous "questions" to users and disclose "answers" to those questions. Furthermore, a user is anyone who requires a document-based FAQ, and may be an unspecified number of third parties or a specific number of people (for example, employees who use the company's internal systems).

[0023] According to the system shown in Figure 1, the information processing device 1 communicates via a network with a Generative Artificial Intelligence (AI) server 2, which acts as the backend for creating question and answer sentences. The generating AI server 2 is based on pre-trained large language models (LLMs). Large language models are a type of deep learning model that has been trained on a large amount of text data. Based on neural network architectures such as Transformers, they can understand context and generate natural language sentences. These are pre-trained with general linguistic knowledge from a large text corpus. Specifically, examples include OpenAI's ChatGPT®, Anthropic's Claude®, Google's Gemini®, and Meta's Llama®. These are pre-trained using general linguistic knowledge derived from large text corpora.

[0024] The information processing device 1 of the present invention generates and stores in storage "question and answer pairs" consisting of pre-defined question sentences and answer sentences for a "document" that is the subject of an answer. This allows for the immediate display of multiple candidate question sentences in response to user access. Furthermore, when a user selects a desired question sentence from among these, the answer sentence can be presented immediately. In this way, the burden on the user to input the question is reduced as much as possible, and the most appropriate answer to that question can be immediately displayed.

[0025] Figure 2 is a functional configuration diagram of the registration stage in the information processing device of the present invention.

[0026] As shown in Figure 2, the information processing device 1 of the present invention includes a document storage 100 and a question-answer pair storage 101. Furthermore, as a registration step for registering question sentences and answer sentences in the question-answer pair storage 101, it further includes a question-answer pair generation unit 111, a question-answer pair selection unit 112, a display keyword generation unit 113, a display keyword registration unit 114, and a desired keyword determination unit 115. These functional components may be realized by executing a program that makes the computer installed in the information processing device function. Moreover, the processing flow of these functional components can also be understood as a question-answering method of the information processing device.

[0027] The document storage 100 stores multiple "documents" that are the subject of a response, which are received from the operator terminal 3. The document may be based on a wide variety of file formats, but it must contain at least text. Examples of file formats include 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), and xml (markup language text).

[0028] Furthermore, if the document includes images, the text generated from those images using character recognition / reading may also be included in the document. Specifically, the character recognition software can be open-source PaddleOCR® or Tesseract OCR®, or cloud-based APIs (Application Programming Interfaces) from companies such as Google® and Amazon®. These technologies allow for highly accurate text recognition by inputting documents.

[0029] <Registration Stage> [Question and Answer Pair Generation Unit 111] The question-answer pair generation unit 111 sends the first prompt to the generation AI server 2a (language model).

[0030] Figure 3 is an explanatory diagram of the question-answer pair generation unit in the information processing device of the present invention.

[0031] According to Figure 3, the first prompt shall include the following: First prompt = One or more "documents" to which the response should be provided. From one or more documents, extract one or more question-and-answer pairs consisting of a question and an answer. The "instruction statement" to be generated Specifically, the first prompt might look like this: -------------------------------------------------------------------------------- Please generate multiple question-and-answer pairs (question and answer) for the user from the following document. #document ············ --------------------------------------------------------------------------------

[0032] In another first embodiment, the question-answer pair generation unit 111 may, as an instruction, identify multiple topics from the context of the document to be answered and instruct the unit to generate a question and an answer for each topic. -------------------------------------------------------------------------------- Based on the context of the following document, identify multiple topics and generate multiple question-and-answer pairs (question and answer) for each topic. #document ············ -------------------------------------------------------------------------------- In this way, by inferring topics from the document beforehand, it is possible to generate multiple questions and answers from the entire document without any omissions.

[0033] Another second embodiment may involve the operator pre-specifying "persona attributes" based on the creator of the document to be answered. The question-answer pair generation unit 111 further includes persona attributes in the first prompt and instructs the generation of a question that takes those persona attributes into consideration as an instruction. -------------------------------------------------------------------------------- #As someone with the persona attributes, please generate multiple question-and-answer pairs (question and answer) for the user from the following #document. #Persona Attribute:····· #document ············ -------------------------------------------------------------------------------- In this way, multiple question and answer statements can be generated depending on the perspective (persona attributes) of the document creator.

[0034] Persona attributes are those of the document creator, and may include attributes based on an organization or service, such as a company profile or IR (Investor Relations) information. For example, in the case of a document about telephone usage, there may be various perspectives such as that of a telephone company or a public institution, and questions can be generated according to each persona attribute.

[0035] According to Figure 3, for example, the multiple documents relate to telephone use and are explanatory documents for the user. The question-answer pair generation unit 111 sends a first prompt containing the document to be answered to the generation AI server 2a and receives multiple question-answer pairs (question and answer). -------------------------------------------------------------------------------- Question 1: "I want to contact the telephone company." Answer 1: "By phone, ...; by email, ..." Question 2: "I would like to inquire about phone charges." Answer 2: "The website for phone charges is..." ················· ················· Question 9: "I can't make a call." Answer 9: "For inquiries about malfunctions..." --------------------------------------------------------------------------------

[0036] In addition, as another embodiment, for example, if a document already has persona attributes assigned to it, the question and answer statements generated from that document may be automatically assigned the same persona attributes.

[0037] As mentioned above, all questions and answers for the documents to be answered can be automatically generated by the question-answer pair generation unit 111. In this invention, the question and answer are generated simultaneously by a single generation AI server. Of course, it is also possible to have a two-step process, first generating the question from the document to be answered, and then generating the answer from that question. This allows for the generation of the answer from a different perspective than the question, leading to an improvement in the quality of the answer.

[0038] Multiple question-answer pairs (question text and answer text) received are output to the question-answer pair selection unit 112.

[0039] Figure 4 is an explanatory diagram of the question-answer pair selection unit in the information processing device of the present invention.

[0040] [Question and Answer Pair Selection Section 112] The question-answer pair selection unit 112 displays multiple "question-answer pairs (question text and answer text)" to the operator terminal 3. This allows the operator to recognize what question text and answer text have been automatically generated for the document they specified. Next, the question-answer pair selection unit 112 allows the operator to select "multiple question-answer pairs" to be registered (published) from among the multiple question-answer pairs. This makes it possible to eliminate unnecessary question-answer pairs from the automatically generated question-answer pairs. The question-answer pair selection unit 112 then registers the question-answer pair selected by the operator in the question-answer pair storage 101.

[0041] [Question and Answer Pair Storage 101] The question-answer pair storage 101 registers the "question-answer pairs" selected by the operator. The information processing device 1 of the present invention, by including a question-answer pair storage 101, can immediately display the optimal answer sentence for a desired question sentence selected by the user.

[0042] [Display Keyword Generation Unit 113] The display keyword generation unit 113 sends a second prompt to the generation AI server 2b (language model) which includes a "group of question-answer pairs" from the question-answer pair storage 101 and an instruction sentence that infers a predetermined number of display keywords that the user is likely to recall as search keys from the group of question-answer pairs. Furthermore, the displayed keywords are not limited to single words; they may also be phrases or short sentences.

[0043] Figure 5 is a first explanatory diagram of the display keyword generation unit in the information processing device of the present invention.

[0044] According to Figure 5, the second prompt may be, for example, the following: ------------------------------------------------------------------------------- #Infer a predetermined number (e.g., 20) of display keywords that users are likely to recall as search keys from a group of question-answer pairs. #document ············ -------------------------------------------------------------------------------- In response, the display keyword generation unit 113 receives multiple "display keywords" from the generation AI server 2b.

[0045] Figure 6 is a second explanatory diagram of the display keyword generation unit in the information processing device of the present invention.

[0046] As shown in Figure 6, the display keyword generation unit 113 instructs the generation AI server 2b (language model) to, in the second prompt, further infer the degree of correlation between each inferred display keyword and multiple question-answer pairs, and to associate a predetermined number of question-answer pairs in order of the highest correlation.

[0047] According to Figure 6, the second prompt may be, for example, the following: ------------------------------------------------------------------------------- #Infer a predetermined number (e.g., 20) of display keywords that users are likely to recall as search keys from a group of question-answer pairs. At this point, calculate the correlation between the #display keyword and the #question-answer pair, and for each #display keyword, associate a predetermined number of question-answer pairs in descending order of correlation. #document ············ ------------------------------------------------------------------------------- Here, "correlation" refers to the degree of correlation between the pre-inferred display keywords and each question-answer pair. This is inferred by the generating AI server 2b and does not need to be stored by a separate algorithm on the information processing device 1 side. The generative AI server 2b, as a language model, may infer correlations using, for example, semantic ranking. Unlike keyword search, this uses techniques such as natural language processing and machine learning to understand the meaning, context, and search intent of the search query, and then ranks the texts (queries) in descending order of relevance. In particular, it inherits the advantages of fuzzy search and, through syntactic / semantic analysis, can rank search results higher if they are closer to the meaning of the search query.

[0048] According to Figure 5, the following display keywords were inferred. [Phone][Address][Charges][Fault] These display keywords are registered in the display keyword registration unit 114.

[0049] As shown in Figure 6, for each display keyword, the generating AI server 2b (language model) receives multiple inferred display keywords along with one or more question-answer pairs associated with each display keyword. As shown in Figure 6, each displayed word is associated with a question-answer pair. Of course, the association could also be done using only the identifier (ID) of the question-answer pair.

[0050] Furthermore, the generation AI server (language model) 2a used by the question-answer pair generation unit 111 and the generation AI server (language model) 2b used by the display keyword generation unit 113 may be the "same" or they may be "different". By using different generation AI servers, even with the same question-answer pair, display keywords based on different perspectives may be obtained.

[0051] [Desired Keyword Determination Unit 115] The desired keyword determination unit 115 prompts the operator to input the "desired keyword" to be added as a display keyword. Here, the "desired keyword" and the "keyword similar under the specified conditions" may or may not be included in the question-answer pair storage 101. If included, the system determines to "add" the desired keyword as a display keyword. If the desired keyword is not included, the system will determine "Do not add" it to the displayed keywords.

[0052] Figure 7 shows an example of a user interface for registering desired keywords.

[0053] As shown in Figure 7(a), the desired keyword determination unit 115 prompts the operator to input the desired keyword. At this time, the operator's terminal 3 also displays the registered "display keywords". The operator can input a "desired keyword" that is not displayed here. -------------------------------------------------------------------------------- Are there any additional keywords you would like to add? [Registration] Display Keywords [Phone][Address][Charges][Fault] --------------------------------------------------------------------------------

[0054] Suppose the operator enters the desired keyword [Register]. At that time, it is determined whether the desired keyword [Register] is included in the question and answer sentences of the question and answer pair storage 101. In this case, it is included in question sentence 6 "I want to change my registered address" stored in the question and answer pair storage 101. In this case, the desired keyword [Register] is registered in the display keyword registration unit 114. As a result, the display keyword [Register] is also explicitly displayed on the operator's terminal 3.

[0055] According to Figure 7(b), the following desired keywords are entered into the operator. -------------------------------------------------------------------------------- Are there any additional keywords you would like to add? [vegetables] Display Keywords [Phone][Address][Charges][Fault][Registration] --------------------------------------------------------------------------------

[0056] Assume the operator enters the desired keyword [vegetables]. The system determines whether the desired keyword [vegetables] is included in the question and answer sentences of the question-answer pair storage 101. In this case, it is not included in either the question or answer sentence. Therefore, the desired keyword [vegetables] is not registered in the display keyword registration unit 114.

[0057] <Operational Phase> Figure 8 is a diagram showing the functional configuration of the information processing device of the present invention during its operational phase.

[0058] According to Figure 8, the information processing device 1 further includes, in its operational phase, a display keyword selection unit 121, a question-answer pair search unit 122, a question text specification unit 123, a desired question text selection unit 124, and an answer text specification unit 125. These functional components may be realized by executing a program that enables the computer installed in the information processing device to function. Furthermore, the processing flow of these functional components can also be understood as a question-answering method of the information processing device.

[0059] Figure 9 is an explanatory diagram of the user interface for the information processing device of the present invention.

[0060] [Keyword Selection Section 121] The display keyword selection unit 121 displays an "input field" for the question text and multiple "display keywords" on the user terminal 4. When the user selects a display keyword, that keyword is entered into the input field.

[0061] According to Figure 9(a), it is displayed as follows: -------------------------------------------------------------------------------- Please enter your inquiry details. [ ] Display Keywords [Phone][Address][Charges][Fault][Registration] -------------------------------------------------------------------------------- Here, if the user selects the displayed keyword [phone], [phone] will be automatically entered into the input field.

[0062] Furthermore, according to the present invention, it is not necessary for the user to input search keywords; it is sufficient that the question text specification unit 123, described later, specifies all or some of the questions to the user.

[0063] [Question and Answer Pair Search Unit 122] The question-answer pair search unit 122 uses the "display keyword" entered in the "input field" of the question text as a search key, refers to the question-answer pair storage 101, and searches for multiple question-answer pairs that are similar based on the search key and predetermined conditions.

[0064] According to the example in Figure 5 mentioned above, the question-answer pair search unit 122 refers to the question-answer pair storage 101 and searches for "multiple question-answer pairs (question text and answer text)" related to the search keyword under predetermined conditions. Here, the predetermined conditions may be based on the "similarity" calculated for each question-answer pair for the search keyword. Then, the pairs may be sorted in descending order of similarity to derive a predetermined number of question-answer pairs. For example, the similarity between "telephone" and "call" is above a predetermined threshold, and "call" is also used as a search key. Furthermore, in another embodiment, as shown in the example in Figure 6 above, it is also preferable that the question sentences of question-answer pairs linked to the display keywords selected by the user are displayed in order of "correlation." Rather than a simple keyword search, the question sentences are displayed in order of the highest correlation inferred by the language model.

[0065] At this time, from the question sentences 1, 2, 4, 6, and 9 in Figure 4 mentioned above, question-answer pairs 1, 2, 4, and 9 (excluding only 6) whose similarity to the search keyword is above a predetermined threshold are searched.

[0066] In another embodiment, as shown in the example in Figure 6 above, the question-answer pair search unit 122 searches for "one or more question-answer pairs (question and answer)" associated with the display keyword entered in the question input field. Because they are already associated, the question-answer pairs can be derived immediately.

[0067] [Question Civilization Section 123] The question text display unit 123 displays multiple question texts to the user terminal 4 based on multiple question-answer pairs retrieved by the question-answer pair search unit 122. According to Figure 9(b), the search keyword [Telephone] is used as the search key, and questions 1, 2, 4, and 9 based on it are explicitly displayed. At this time, questions 1, 2, 4, and 9 in question-answer pairs 1, 2, 4, and 9 are sorted in descending order of similarity and displayed to the user. Question 1: "I want to contact the telephone company." Question 2: "I would like to inquire about phone charges." Question 4: "My smartphone is broken." Question 9: "I can't make a call." Here, the question text of the question-answer pair with the highest similarity (for example, question text 1) may be entered into the input field. The other question texts 2, 4, and 9 are listed and displayed below.

[0068] [Selection section for desired question 124] The desired question selection unit 124 prompts the user to select a "desired question" from among several question sentences. The desired question selection unit 124 prompts the user to select one "desired question" from among several question sentences explicitly displayed on the user terminal 4. An answer sentence has already been associated with that desired question sentence from among several question-answer pairs found by the question-answer pair search unit 122. According to Figure 9(b), the user selected question 9, "I cannot make a call."

[0069] [Answer Civilization Section 125] The response text display unit 125 displays the response text for the question-answer pair corresponding to the desired question text to the user terminal 4. The response text display unit 125 transmits the "response text" that forms a question-and-answer pair for the desired question text selected by the user to the user terminal 4 and displays it to the user. A key feature of the present invention is that, in response to a user's desired question, the question-answer pair search unit 122 can immediately output an answer from a previously searched question-answer pair without requiring inference by the generating AI server.

[0070] Figure 10 is a diagram illustrating the function of collecting documents into document storage.

[0071] <Storage in document storage> According to Figure 10, the operator can specify the file path of the local storage to the information processing device 1. The operator can specify a file path to upload multiple documents under that path to the information processing device 1 and store them in the document storage 100. Alternatively, the system may display indexes of multiple files under a specific folder on the user terminal 4 and allow the user to select some or all of those file indexes. For example, multiple files can be uploaded and registered to the information processing device 1 simultaneously using "cut and paste" or "drag and drop".

[0072] Alternatively, the operator may specify the URL of a website on the internet (Uniform Resource Locator) as the location of the document to the information processing device 1. According to Figure 10, the information processing device 1 has a scraping unit. When a URL or file path is specified by the operator, the scraping unit accesses a website on the internet. It then downloads a document from the URL or file path and stores it in document storage.

[0073] Furthermore, operators can specify documents by including the wildcard "*" in the URL or file path they specify. In this case, the scraping unit downloads multiple documents simultaneously and stores them in document storage.

[0074] Furthermore, if a URL or file path is specified by the operator, it is preferable to periodically crawl that URL or file path, download updated documents (site crawling), and update the document storage.

[0075] <Adding tags to documents> In another embodiment, each document stored in the document storage may be assigned a persona attribute specified by the operator as a hashtag. This identifies organizational or personal attributes based on the document creator. Users can assign hashtags after registering a document, or they can assign hashtags simultaneously with document registration. The persona attribute as a hashtag is used as the persona attribute in the registration stage described above when generating the question and answer text.

[0076] As described in detail above, the information processing device, program, and method of the present invention reduce the burden on the user in inputting question sentences as much as possible, and immediately display the most appropriate answer sentence for that question sentence.

[0077] Various changes, modifications, and omissions to the scope of the technical concept and viewpoint of the present invention can be readily made by those skilled in the art with respect to the various embodiments of the present invention described above. The above description is merely illustrative and is not intended to limit the present invention in any way. The present invention is limited only to what is limited by the claims and their equivalents. [Explanation of symbols]

[0078] 1. Information Processing Device 100 Document Storage 101 Question Answer Pair Storage 111 Question and Answer Pair Generation Unit 112 Question and Answer Pair Selection Section 113 Display Keyword Generation Unit 114 Display Keyword Registration Section 115 Desired Keyword Determination Unit 121 Display Keyword Selection Section 122 Question and Answer Pair Search Section 123 Question Civilization Department 124 Selection of Desired Question 125 Answer Civilization Department 2. Generative AI server, language model 3 Operator Operation Terminal 4. User terminals

Claims

1. In an information processing device that accepts questions based on multiple documents to be answered, As part of the registration process, A question-answer pair generation means sends a first prompt to a language model that includes multiple documents to be answered and an instruction that the language model should generate one or more question-answer pairs of question sentences and answer sentences from the multiple documents, and receives multiple question-answer pairs from the language model. A question-answer pair selection means displays the multiple question-answer pairs on the operator's terminal, allows the operator to select the question-answer pairs to be made public, and registers the selected question-answer pairs in the question-answer pair storage. A display keyword generation means sends a second prompt to a language model which includes a group of question-answer pairs in the question-answer pair storage, a predetermined number of display keywords that the user is likely to recall as search keys from the group of question-answer pairs, a correlation between each of the inferred display keywords and multiple question-answer pairs, and instructions to associate a predetermined number of question-answer pairs in order of decreasing correlation, and receives from the language model multiple display keywords along with a predetermined number of question-answer pairs associated with each display keyword. It has, In the operational phase, A display keyword selection means that displays a question input field and multiple display keywords on the user terminal, and when the user selects a display keyword, that display keyword is entered into the input field. A question-answer pair search means that searches for a predetermined number of question-answer pairs already associated with the display keyword entered in the input field of the question text, A question text expliciting means that explicitly displays multiple question texts to the user terminal based on multiple question-answer pairs that have been searched. An information processing device characterized by having the following features.

2. In the operational phase, A means for selecting a desired question from among the multiple question sentences, A means for explicitly displaying the answer text of a question-answer pair for the desired question text to the user terminal, and The information processing apparatus according to claim 1, characterized by having the following features.

3. The language model used in the question-answer pair generation means and the language model used in the display keyword generation means may be the same or different. The information processing apparatus according to feature 1.

4. The information processing device is a question answering server accessed from terminals operated by operators and users. The language model is a generative AI server based on a pre-trained, large-scale language model. The information processing apparatus according to feature 1.

5. In a program that operates a computer installed in a device that accepts questions based on multiple documents to be answered, As part of the registration process, A question-answer pair generation means sends a first prompt to a language model that includes multiple documents to be answered and an instruction that the language model should generate one or more question-answer pairs of question sentences and answer sentences from the multiple documents, and receives multiple question-answer pairs from the language model. A question-answer pair selection means displays the multiple question-answer pairs on the operator's terminal, allows the operator to select the question-answer pairs to be made public, and registers the selected question-answer pairs in the question-answer pair storage. A display keyword generation means sends a second prompt to a language model which includes a group of question-answer pairs in the question-answer pair storage, a predetermined number of display keywords that the user is likely to recall as search keys from the group of question-answer pairs, a correlation between each of the inferred display keywords and multiple question-answer pairs, and instructions to associate a predetermined number of question-answer pairs in order of decreasing correlation, and receives from the language model multiple display keywords along with a predetermined number of question-answer pairs associated with each display keyword. to make the computer work, In the operational phase, A display keyword selection means that displays a question input field and multiple display keywords on the user terminal, and when the user selects a display keyword, that display keyword is entered into the input field. A question-answer pair search means that searches for a predetermined number of question-answer pairs already associated with the display keyword entered in the input field of the question text, A question text expliciting means that explicitly displays multiple question texts to the user terminal based on multiple question-answer pairs that have been searched. A program characterized by its ability to make a computer function.

6. In a question reception method for a device that accepts questions based on multiple documents to be answered, The device is As part of the registration process, A first step involves sending a first prompt to a language model that includes multiple documents to be answered and instructions that the language model should generate one or more question-answer pairs consisting of a question and an answer from those multiple documents, and receiving multiple question-answer pairs from the language model. The second step involves displaying the multiple question-answer pairs on the operator's terminal, allowing the operator to select the question-answer pairs to be made public, and registering the selected question-answer pairs in the question-answer pair storage. A third step involves sending a second prompt to a language model which includes a group of question-answer pairs in the question-answer pair storage, a predetermined number of display keywords that the user is likely to recall as search keys from the group of question-answer pairs, a correlation between each of the inferred display keywords and multiple question-answer pairs, and instructions to associate a predetermined number of question-answer pairs in order of decreasing correlation, and receiving from the language model a set of multiple display keywords with a predetermined number of question-answer pairs associated with each display keyword. In the operational phase, A fourth step involves displaying a question input field and the multiple display keywords on the user's terminal, and when the user selects any of the display keywords, those display keywords are entered into the input field. The fifth step involves searching for a predetermined number of question-answer pairs already associated with the display keywords entered in the question input field, The sixth step is to explicitly display multiple question sentences based on the multiple question-answer pairs found to the user terminal. A method for receiving questions from a device, characterized by performing the following actions.

Citation Information

Patent Citations

  • Searcher device

    JP2016192067A

  • Program, method, information processor and system

    JP2025060386A

  • Non-transitory computer-readable storage medium, answering device, and method for retrieving document and answering to question sentence

    JP2026068676A

  • JPP7823852B