Information processing apparatus and information processing program

Abstract

To provide an information processing apparatus which can efficiently provide information about what a user wants to know.SOLUTION: A server device 2 being the information processing apparatus comprises: means which receives a question sentence for asking what a user wants to know, from a terminal device operated by the user; means which extracts a content including an expression constituting the received question sentence and expressions similar to the expression constituting the question sentence, from published information; means which generates an answer to the question sentence by inputting a first instruction for requesting a large-scale language model to generate the answer to the received question sentence by using the extracted content, the question sentence, and received content to the large-scale language model; and means which provides the generated answer to the terminal device.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to an information processing device and an information processing program. [Background technology]

[0002] Patent Document 1 below discloses a system for displaying newspaper articles. This system displays a list of the top 10 news items of the week selected from newspaper articles for the past week, and displays an article selected by the user from among them on a newspaper page display unit. In other words, the system of Patent Document 1 picks out the top 10 ranked articles from among the articles published in the newspaper and presents them to the user. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-178370 Summary of the Invention [Problem to be solved by the invention]

[0004] The system of Patent Document 1 provides users with articles picked up by the system, so if the information the user wants is not included in the picked articles, the user must search for that information themselves. However, if the task of manually searching for and compiling information about the desired topic from the vast number of articles published in newspapers were to be done, it would require a great deal of time and effort. Therefore, such a task would be a heavy burden on the worker.

[0005] The present invention has been made to solve the above-mentioned problems, and one of its objects is to provide an information processing device and an information processing program that can efficiently provide information about what a user wants to know. [Means for solving the problem]

[0006] An information processing device according to one embodiment of the present invention comprises: means for receiving a question from a terminal device operated by the user to ask what the user wants to know; means for extracting content from published information that includes wording that constitutes the received question and wording similar to the wording that constitutes the question; means for generating an answer to the question by inputting a first instruction to a large-scale language model that requests the large-scale language model to generate an answer to the input question using the extracted content, the question, and the input content; and means for providing the generated answer to the terminal device.

[0007] An information processing program according to another aspect of the present invention causes a computer to function as: means for receiving a question from a terminal device operated by the user to ask what the user wants to know; means for extracting content from published information that includes wording that constitutes the received question and wording similar to the wording that constitutes the question; means for generating an answer to the question by inputting a first instruction to the large-scale language model using the extracted content, the question, and the input content to request the large-scale language model to generate an answer to the input question; and means for providing the generated answer to the terminal device.

[0008] According to these aspects, a question is received from a user to ask what the user wants to know, content containing the words that make up the question and words similar to the words that make up the question is extracted from published information, and the extracted content, the question, and the first instruction are input into a large-scale language model, thereby generating an answer to the question and providing it to the user. [Effects of the Invention]

[0009] According to the present invention, it is possible to provide an information processing device and an information processing program that can efficiently provide information about desired information. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a diagram illustrating a configuration of an information processing system including a server device that is an information processing device according to an embodiment; [Figure 2] FIG. 2 is a diagram illustrating a configuration of a server device. [Figure 3] FIG. 2 is a diagram illustrating a configuration of a terminal device. [Figure 4] 10 is an example of a screen displayed on a terminal device. [Figure 5] 10 is an example of a screen displayed on a terminal device. [Figure 6] 10 is an example of a screen displayed on a terminal device. [Figure 7] 10 is a flowchart illustrating an example of an operation when providing an answer to a question. [Figure 8] 10 is an example of a screen displayed on a terminal device. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that the same elements are given the same reference numerals and redundant description will be omitted.

[0012] [Outline of information processing system] An information processing system including a server device which is an information processing device according to an embodiment will be described with reference to Fig. 1. The information processing system 1 includes a server device 2 which is an information processing device, and a terminal device 3, for example.

[0013] The information processing system 1 according to the embodiment is a system that realizes, for example, a function of generating an answer to a question entered by a user and providing the answer to the user. The question is a sentence that asks what the user wants to know. The answer is a response to the question, and is answer content that is newly generated based on existing content. The existing content is, among the content, articles such as news, columns, and essays that have been published in published newspapers (including electronic versions) and magazines. The users may include, for example, editors who edit articles and people in charge of articles.

[0014] The server device 2 is, for example, a computer device that executes processes to realize functions such as generating and providing answers to the above-mentioned questions. The terminal device 3 is, for example, a computer device used by a user who inputs questions.

[0015] The terminal device 3 may be, for example, a PC (personal computer), a notebook PC, a tablet terminal, a smartphone, or other terminal device.

[0016] The terminal device 3 may or may not have an application for providing a response (hereinafter also referred to as an "app") installed.

[0017] If an answer-providing application is not installed on the terminal device 3, it is preferable to exchange data between the terminal device 3 and the server device 2, for example, by accessing a specific site on the server device 2 and using a screen displayed on the terminal device 3 by an answer-providing web application running on the website.

[0018] On the other hand, if an application for providing answers is installed on the terminal device 3, it is preferable to exchange data between the terminal device 3 and the server device 2, for example, using a screen displayed on the terminal device 3 by executing the installed application.

[0019] In this embodiment, a case will be described in which an application for providing answers is not installed on the terminal device 3, but a Web application for providing answers is used.

[0020] The server device 2 and the terminal device 3 are configured to be able to communicate with each other via a network N. The network N may be, for example, the Internet, a LAN, a dedicated line, a telephone line, an in-house network, a mobile communication network, WiFi (Wireless Fidelity), Bluetooth (registered trademark), other communication lines, or a combination thereof, and may be wired or wireless.

[0021] [Server device configuration] As shown in FIG. 2, the server device 2 includes, as its physical configuration, a processor 21, a communication interface 22, and a storage device 23, for example.

[0022] The processor 21 is, for example, a CPU (Central Processing Unit) or an MPU (Micro Processing Unit), etc. The processor 21 executes a program 231 stored in the storage device 23, thereby functioning as a control unit 211 that realizes various functions of the server device 2.

[0023] The control unit 211 realizes functions such as displaying various screens on the terminal device 3 in accordance with requests received from the terminal device 3, receiving content entered on various screens from the terminal device 3, and executing predetermined processing based on information received from the terminal device 3 and transmitting the processing results to the terminal device 3.

[0024] The communication interface 22 connects to the network N and functions as a communication unit that communicates with other devices on the network N.

[0025] The storage device 23 is a computer-readable recording medium such as a disk drive or a semiconductor memory. The storage device 23 functions as a storage unit that stores a program 231 for implementing various functions of the server device 2, various data 232 used by the program 231, and a large language model (hereinafter also referred to as "LLM") 233. Note that the LLM 233 does not necessarily have to be stored in the server device 2, and may be stored in an external device or system.

[0026] The various data 232 includes, for example, data constituting existing content and information relating to existing content.

[0027] LLM233 is a generative AI model that generates answers to questions based on input prompts, etc. For example, GPT-3.5, GPT-3.5-16K, GPT-4, etc. can be used as LLM233.

[0028] [Terminal device configuration] As shown in FIG. 3, the terminal device 3 includes, as its physical configuration, a processor 31, a communication interface 32, a storage device 33, an input device 34, and a display device 35, for example.

[0029] The processor 31 is, for example, a CPU, an MPU, etc. The processor 31 executes a program 331 stored in the storage device 33, thereby functioning as a control unit 311 that realizes various functions of the terminal device 3.

[0030] The control unit 311 realizes, for example, functions such as sending various requests to the server device 2 in accordance with the operation instructions of the user operating the terminal device 3 and displaying various screens on the display device 35, sending contents entered into various screens to the server device 2 in accordance with the operation instructions of the user, and executing predetermined processing based on the information entered into various screens and displaying the processing results on the display device 35.

[0031] The communication interface 32 connects to the network N and functions as a communication unit that communicates with other devices on the network N.

[0032] The storage device 33 is a computer-readable recording medium such as a disk drive or a semiconductor memory. The storage device 33 functions as a storage unit that stores a program 331 for realizing various functions of the terminal device 3 and various data used by the program 331.

[0033] The input device 34 functions as an input unit that accepts input from a user. As the input device 34, for example, a touch panel, a pen tablet, a keyboard, a mouse, a microphone, etc. can be used.

[0034] The display device 35 functions as a display unit that displays images, screens, etc. As the display device 35, for example, an organic EL display, a liquid crystal display, etc. can be used.

[0035] [Functions of information processing systems] This section describes various functions of the information processing system 1. In the following explanation, a case will be described in which an answer to a question entered by a user such as an editor is generated based on existing content such as published electronic newspapers or articles published in newspaper pages (hereinafter also referred to as "published articles"), and the answer is provided to the user. Note that this explanation is merely an example, and the existing content is not limited to the example content.

[0036] The functions of the information processing system 1 are provided, for example, using a screen displayed on the terminal device 3 by a web application for providing answers on the server device 2.

[0037] [Function to vectorize published articles (preparation process)] As a preparatory process before executing the process of generating an answer to the question, the server device 2 extracts specific public articles from a database that stores public articles as electronic data, vectorizes the extracted public articles, and stores them in a search database. It is preferable to use, for example, news articles as the specific public articles. When using news articles, specific exceptional articles may be excluded from the extraction. Specific exceptional articles may include, for example, articles written outside the user's company, column articles other than editorials, sports articles, culture articles, interview articles, personnel articles, numerical table articles, articles containing "ask the," "the interviewer," or "contribution" in the text, etc.

[0038] A search period may also be set in association with each published article. When setting a search period, multiple patterns may be used. Examples of multiple patterns include a specific date range, the last month, the last three months, and the last 13 months.

[0039] [Function to accept questions] The server device 2 receives a question from the terminal device 3 operated by the user, asking what the user wants to know. At this time, search conditions for searching for published articles may also be received from the terminal device 3 operated by the user. The search conditions may include, for example, a target period for published articles, and this target period may be set to a specific date range, the last month, the last three months, the last 13 months, or the like.

[0040] The question and search conditions input by the user will be described with reference to Fig. 4. Fig. 4 is an example of a screen 4a displayed on the terminal device 3. This screen 4a is provided with, for example, an instruction input field 4b for inputting instructions to the server device 2, and a main field 4c for inputting a question and displaying an answer.

[0041] The instruction input field 4b illustratively includes a selection field 4ba for selecting the target period for public articles. As described above, the target period for public articles is an example of a search condition for searching for public articles. In the same figure, four options are provided as the target period for public articles, illustratively "Automatic Selection," "Last Month," "Last 3 Months," and "13 Months," and the user can select one of them. Selecting "Automatic Selection" enables the function of "expanding the target period for public articles and re-executing the process of generating an answer if an answer to a question is not output from the LLM 233," which will be described in the "Function for retrying answer generation" below.

[0042] Note that the instruction input field 4b is not limited to the selection field 4ba shown in the figure, and input fields and selection fields for specifying various instructions to the server device 2, including the instruction content for the LLM 233, can be provided as appropriate.

[0043] The main field 4c is provided with, for example, a question input field 4ca for inputting a question, a question button 4cb, and displays answers 4cc to the question. The answers 4cc will be described later.

[0044] The question input field 4ca is a field where the user can freely input a question to ask about something they want to know. In the figure, for example, "I would like to know about decarbonization efforts by Japanese companies" is input.

[0045] The question button 4cb is a button that the user presses when asking a question entered in the question input field 4ca. When the question button 4cb is pressed, the question and the target period of the public article are transmitted, and the question and the target period of the public article are accepted by the server device 2. The server device 2 identifies the public article to refer to based on the accepted target period and generates an answer to the accepted question.

[0046] [Function to extract reference content] The server device 2 extracts from the database public articles that contain the wording that constitutes the received question and wording similar to the wording that constitutes the question. The public articles to be extracted are narrowed down to public articles that are published within the target period of the public articles selected by the user. This will be explained in detail below.

[0047] First, the server device 2 vectorizes the question text and calculates the similarity between the vectorized question text and vectorized published articles stored in a search database. Next, the server device 2 extracts published articles with high similarity from the database storing published articles as content to be referenced when generating an answer (hereinafter also referred to as "reference content").

[0048] Whether the similarity is high or not can be determined, for example, by setting a predetermined threshold. Alternatively, a predetermined number of published articles may be extracted as reference content in descending order of similarity. For example, if the predetermined number is set to 10, published articles with the highest to tenth highest similarities may be extracted as reference content. In this case, published articles with a similarity below the predetermined threshold may be excluded from the extraction targets.

[0049] It is preferable to add copyright information to published articles and extract reference content by narrowing down the search to published articles whose copyrights are held by the user's company.

[0050] [Analyzes questions and selects published articles to extract] When extracting the reference content, the server device 2 analyzes the question sentence and uses the analysis results to select (filter) the published articles to be extracted. This will be specifically described below.

[0051] First, the server device 2 divides the question sentence into morphemes using, for example, natural language processing, and identifies proper nouns from among the divided morphemes. Next, the server device 2 acquires synonyms of the identified proper nouns based on the identified proper nouns. When acquiring synonyms, the server device 2 may input the identified proper nouns from the question sentence to the LLM 233, and acquire synonyms of the proper nouns from the LLM 233.

[0052] Next, when extracting the reference content, the server device 2 extracts published articles that include either the specified proper noun or the acquired synonyms. This allows the published articles selected by the specified proper noun and the acquired synonyms to be extracted as reference content.

[0053] Proper nouns have spelling variations, such as the existence of both full names and abbreviations, and whether or not to add a "-" (hyphen). Therefore, by extracting reference content using synonyms that encompass the spelling variations present in the proper noun along with the proper noun, it becomes possible to extract published articles related to the proper noun with fewer omissions. This makes it possible to improve the accuracy of the reference content.

[0054] [Function to generate answers to questions] The server device 2 inputs the reference content and the question to the LLM 233 and generates an answer to the question. At this time, if the reference content does not contain valid information, it is preferable to further instruct the LLM 233 not to output an answer to the question (seventh instruction).

[0055] Here, when the reference content and the question are input into the LLM233, an instruction (first instruction) is input to the LLM233 together with the input to request the LLM233 to generate an answer to the input question using the input reference content.

[0056] This instruction is incorporated in advance as a prompt to be input to the LLM 233 in the program executed by the server device 2 when the user presses the question button 4cb to request an answer to the question. In other words, this instruction is not input by the user but is stored in advance in the server device 2.

[0057] A specific example of an answer will be described with reference to Figure 4. An answer 4cc to the question is shown below the question input field 4ca and question button 4cb in the main field 4c in Figure 4. This answer 4cc is an answer generated in response to the question, "I would like to know about decarbonization efforts by Japanese companies."

[0058] [Ability to retry generating answers] When an answer to the question is not output from the LLM 233, the server device 2 executes (retries) the process of generating an answer to the question again. This will be specifically described below.

[0059] First, the server device 2 expands the target period of the published articles that are the search criteria when no answer to the question is output from the LLM 233. A specific example of expanding the target period is when the current target period is set to the most recent one month, and then the target period is set to the most recent three months.

[0060] Next, the server device 2 extracts reference content from the database for published articles within the expanded target period. Next, the server device 2 inputs the extracted reference content and the question into the LLM 233, thereby again executing the process of generating an answer to the question.

[0061] Here, when the process of generating an answer to a question is executed again, as when it was executed the first time, in addition to inputting the reference content and question extracted this time, an instruction (first instruction) is input to LLM233 to request LLM233 to generate an answer to the input question using the input reference content.

[0062] This instruction is incorporated in advance as a prompt to be input to the LLM 233 in a program executed by the server device 2 when an answer to the question is not output from the LLM 233. In other words, this instruction is not input by the user but is stored in advance in the server device 2.

[0063] Furthermore, if an answer to the question is not output from the LLM233 even after the target period is extended, the LLM233 may be made to modify the question and the process of generating an answer may be executed again using the modified question. To achieve this, for example, when the currently extracted reference content and question are input to the LLM233, it is preferable to input, together with the input, an instruction to request the LLM233 to modify the question so that the reference content includes valid information and to generate and output an answer using the modified question if an answer cannot be output next time.

[0064] [Function to generate related questions and answers] The server device 2 generates a question related to the wording included in the generated answer (hereinafter also referred to as a "related question"), extracts related content (described later) using the related question, and generates an answer to the related question (hereinafter also referred to as a "related answer") using the related content. This will be explained in detail below.

[0065] First, the server device 2 generates a related question based on the wording included in the generated answer. To achieve this, for example, when inputting the reference content and the question to the LLM 233, it is preferable to input, together with the input, an instruction (second instruction) to the LLM 233 to request the LLM 233 to generate a related question based on the wording included in the generated answer.

[0066] This instruction is incorporated in advance as a prompt to be input to the LLM 233 in a program executed by the server device 2 after the user presses the question button 4cb to request an answer to the question. In other words, this instruction is not input by the user but is stored in advance in the server device 2.

[0067] The related questions will be described in detail with reference to Figure 5. The main field 4c displayed on the screen 5a in Figure 5 is displayed, for example, by scrolling down the main field 4c displayed on the screen 4a in Figure 4. The main field 4c in Figure 5 shows examples of related questions 4cd, 4cf, and 4ch that were generated based on the wording included in the answer 4cc in Figure 4.

[0068] Next, the server device 2 extracts from the database content (hereinafter also referred to as "related content") that includes the words that make up the generated related question sentence and words similar to the words that make up the related question sentence. Note that the specific procedure for extracting related content is the same as the specific procedure for extracting reference content described above in the "Function for extracting referenced content" and "Function for analyzing question sentences and selecting published articles to extract."

[0069] Next, the server device 2 inputs an instruction (third instruction) to the LLM 233 to request the LLM 233 to generate an answer to the related question using the extracted related content, the related question, and the related content to be input, and generates an answer to the related question (hereinafter also referred to as a "related answer"). Note that the specific procedure for generating a related answer is the same as the specific procedure for generating an answer described above in the "Function for generating an answer to a question" and "Function for retrying answer generation." Also, the instructions are stored in advance in the server device 2, similar to each of the above-mentioned functions.

[0070] A specific example of related answers will be described with reference to Fig. 5. In the main column 4c of Fig. 5, related answers 4ce, 4cg, and 4ci generated based on related question sentences 4cd, 4cf, and 4ch are shown.

[0071] [Function to generate additional questions and answers] The server device 2 generates a question (hereinafter also referred to as an "additional question") inquiring about recent topics or news related to the words constituting the question, extracts additional content (described later) using the additional question, and generates an answer to the additional question (hereinafter also referred to as an "additional answer") using the additional content. A specific description will be given below.

[0072] First, the server device 2 generates an additional question based on the wording that constitutes the question. To achieve this, for example, when inputting the reference content and the question to the LLM 233, it is preferable to input, together with the input, an instruction (fourth instruction) to the LLM 233 to request the LLM 233 to generate an additional question that asks about recent topics or news based on the wording that constitutes the question.

[0073] This instruction is incorporated in advance as a prompt to be input to the LLM 233 in a program executed by the server device 2 after the user presses the question button 4cb to request an answer to the question. In other words, this instruction is not input by the user but is stored in advance in the server device 2.

[0074] Next, the server device 2 extracts content (hereinafter also referred to as "additional content") from the database that includes the wording that constitutes the generated additional question and wording similar to the wording that constitutes the additional question. Note that the specific procedure for extracting additional content is the same as the specific procedure for extracting reference content described above in the "Function for extracting referenced content" and "Function for analyzing question text and selecting published articles to extract."

[0075] Next, the server device 2 inputs an instruction (fifth instruction) to the LLM 233 to request the LLM 233 to generate an answer to the additional question using the extracted additional content, the additional question, and the additional content to be input, and generates an answer to the additional question (hereinafter also referred to as an "additional answer"). The specific procedure for generating the additional answer is the same as the specific procedure for generating an answer described above in the "Function for generating an answer to a question" and "Function for retrying answer generation". Also, the instructions are stored in advance in the server device 2, similar to each of the above-mentioned functions.

[0076] A specific example of an additional answer will be described with reference to Fig. 6. The main field 4c displayed on the screen 6a in Fig. 6 is displayed, for example, by scrolling further down the main field 4c displayed on the screen 5a in Fig. 5. The main field 4c in Fig. 6 shows an example of an additional answer 4cj to an additional question sentence generated based on the wording constituting the question sentence entered in the question input field 4ca in Fig. 4.

[0077] [Report generation function] The server device 2 generates a report by compiling the generated answer, related answer, and additional answer in a predetermined report format. This function may be realized by inputting the generated answer, related answer, and additional answer into the LLM 233.

[0078] Here, when the generated answers, related answers, and additional answers are input into LLM233, an instruction (sixth instruction) is input into LLM233 together with these inputs to request LLM233 to generate a report compiled in a specified report format using the answers, related answers, and additional answers to be input.

[0079] This instruction is incorporated in advance into the program executed by the server device 2 when the user requests the creation of a report, as a prompt to be input to the LLM 233. In other words, this instruction is not input by the user, but is stored in advance in the server device 2.

[0080] The specified report format is a display format that defines how answers, related answers, and additional answers are to be structured (arranged, format, etc.) and displayed, and it is preferable that the display format be set by the user as appropriate.

[0081] A specific example of a report will be described with reference to Fig. 6. The main column 4c in Fig. 6 shows an example of a portion of a summary report 4ck compiled in a predetermined report format. The summary report 4ck is generated based on the answer 4cc displayed in the main column 4c in Fig. 4, the related answers 4ce, 4cg, and 4ci displayed in the main column 4c in Fig. 5, and the additional answer 4cj displayed in the main column 4c in Fig. 6.

[0082] [Document download function] The server device 2 downloads the generated answers, related answers, additional answers, and reports to the terminal device 3 as editable document data.

[0083] [Function to display comparison results with published articles] The server device 2 uses natural language processing to provide an answer that visually distinguishes between morphemes that make up the generated answer and morphemes that differ from those of the published article, which is the reference content. For example, among the wording of the answer displayed in black, the wording that appears in the published article may be displayed in green, and the wording generated by the LLM 233 may be displayed in red. This allows the user to easily visually confirm the rephrasing and additional content created by the LLM 233.

[0084] Specifically, the server device 2 uses natural language processing to divide the generated answer into morphemes and determines whether the divided morphemes exist in the text of the published article used to generate the answer. When displaying the answer, the server device 2 may, for example, display in green the morphemes of the answer that exist in the text of the published article and display in red the morphemes of the answer that do not exist in the text of the published article.

[0085] At this time, users can specify specific parts of speech, such as particles, auxiliary verbs, or suffixes, from among the morphemes that make up the answer to exclude from comparison with the text of the published article. Morphemes excluded from comparison will remain in black. This allows users to visually identify parts of the answer that are expressed in red (parts independently generated by LLM233) that differ from the content of the published article used to generate the answer.

[0086] [Behavior when providing an answer to a question] An example of the operation when providing an answer to a question will be described with reference to FIG.

[0087] First, the user operates the terminal device 3 to select the target period for public articles included in the search criteria for public articles in the selection field 4ba of Figure 4, enter a question in the question input field 4ca, and press the question button 4cb.The server device 2 then accepts the entered search criteria and question from the terminal device 3 (step S101).

[0088] Next, the server device 2 extracts from the database as reference content public articles that satisfy the search conditions received in step S101 and that contain the wording that constitutes the question received in step S101 and wording similar to the wording that constitutes the question (step S102).

[0089] Here, when extracting the reference content, the server device 2 may analyze the question sentence and select published articles to extract using the analysis result (proper nouns included in the question sentence).

[0090] Next, the server device 2 inputs the reference content extracted in the above step S102 and the question received in the above step S101 to the LLM 233, and generates an answer to the question (step S103).

[0091] Here, if an answer to the question is not output from the LLM 233, the server device 2 may extend the target period of the published article and re-execute the process of generating an answer in steps S102 to S103. Furthermore, if an answer to the question is not output from the LLM 233 even after re-execution, the server device 2 may cause the LLM 233 to revise the question and re-execute the process of generating an answer in steps S102 to S103 using the revised question.

[0092] Next, the server device 2 generates a related question based on the wording included in the answer generated in step S103 (step S104).

[0093] Next, the server device 2 extracts, from the database as related content, published articles that include the words that make up the related question generated in step S104 and words similar to the words that make up the related question (step S105).

[0094] Next, the server device 2 inputs the related content extracted in the above step S105 and the related question generated in the above step S104 to the LLM 233, and generates related answers to the related question (step S106).

[0095] Next, the server device 2 generates an additional question asking about recent topics or news related to the wording constituting the question received in step S101 (step S107).

[0096] Next, the server device 2 extracts, from the database as additional content, published articles that include the wording that constitutes the additional question generated in step S107 and wording similar to the wording that constitutes the additional question (step S108).

[0097] Next, the server device 2 inputs the additional content extracted in the above step S108 and the additional question generated in the above step S107 to the LLM 233, and generates an additional answer to the additional question (step S109).

[0098] Next, the server device 2 compiles the answer generated in step S103, the related answer generated in step S106, and the additional answer generated in step S109 into a predetermined report format to generate a report (step S110).

[0099] Next, the server device 2 provides the answer generated in step S103, the related answer generated in step S106, the additional answer generated in step S109, and the report generated in step S110 to the user (step S111). Providing them to the user includes displaying them on the terminal device 3 and downloading them to the terminal device 3.

[0100] As described above, the information processing system 1 including the server device 2 can receive a question from a user to ask what the user wants to know, extract reference content from published articles that includes the wording that makes up the question and wording similar to the wording that makes up the question, and input instructions to the LLM233 to request the LLM233 to generate an answer to the input question using the extracted reference content, the question, and the input reference content, thereby generating an answer to the question and providing it to the user.

[0101] In addition, by generating a related question based on the wording contained in the generated answer, further extracting related content from the published article that includes the wording that constitutes the generated related question and wording similar to the wording that constitutes the related question, and inputting instructions to LLM233 to request LLM233 to generate a related answer to the related question using the extracted related content, the related question, and the related content to be input, further related answers to the related question can be generated and provided to the user.

[0102] In addition, an additional question can be generated that asks about recent topics or news related to the wording that makes up the question received from the user, and additional content including the wording that makes up the generated additional question and wording similar to the wording that makes up the additional question can be further extracted from the published article.By inputting instructions to LLM233 to request LLM233 to generate an additional answer to the additional question using the extracted additional content, the additional question, and the additional content to be input, an additional answer to the additional question can be further generated and provided to the user.

[0103] Furthermore, the generated answers, related answers and additional answers can be compiled in a predetermined report format and provided to the user.

[0104] Therefore, the information processing system 1 including the server device 2 can efficiently provide the user with information about what the user wants to know.

[0105] [Variations] The present invention is not limited to the above-described embodiment, and can be embodied in various other forms without departing from the spirit of the present invention. Therefore, the above-described embodiment is merely an example in all respects and should not be interpreted as being limiting. For example, the order of the above-described processing steps can be arbitrarily changed or executed in parallel as long as no contradiction occurs in the processing content.

[0106] In the "function for generating related questions and related answers" of the above-described embodiment, after generating an answer to a question, the related questions and related answers are automatically generated, but this is not limiting. For example, after generating an answer to a question, candidates for related questions may be displayed on the screen, and the user may be allowed to select a desired related question, and then the related answer may be generated.

[0107] The function of this modified example will be described in detail with reference to Fig. 8. Fig. 8 is an example of a screen 8a displayed on the terminal device 3. Similar to the screen 4a of Fig. 4 described in the embodiment, this screen 8a has an instruction input field 4b for inputting instructions to the server device 2 and a main field 4c for inputting a question and displaying the answer.

[0108] The instruction input field 4b has a selection field 4ba for selecting the target period of the published article as described in the embodiment, as well as a selection field 4bb for selecting the processing mode. In Fig. 8, an "automatic" mode and a "manual selection" mode are provided as processing modes, for example.

[0109] When "automatic" mode is selected, similar to the "function for generating related questions and related answers" described in the embodiment, an answer 4cc to the question is generated and displayed on the screen 8a, and subsequently related questions and related answers are automatically generated and displayed on the screen.

[0110] On the other hand, when the "manual selection" mode is selected, an answer 4cc to the question is generated and displayed on the screen 8a, and then candidate related questions 4cm are displayed on the screen 8a. Then, when the user selects a related question, a related answer to the selected related question is generated and displayed on the screen. Candidate related questions are generated in the same way as the related questions of the "function for generating related questions and related answers" described in the embodiment.

[0111] In Figure 8, six questions are displayed as related question candidates 4cm, and three questions are selected. The three selected questions are displayed in the display field 4cn that displays the selected related questions.

[0112] The Continue Processing button 4co is a button that the user presses when continuing processing using the selected related question sentence. When the Continue Processing button 4co is pressed, a related answer to the related question sentence is generated and displayed on the screen. The related answer is generated in the same manner as the "function for generating related question sentences and related answers" described in the embodiment.

[0113] In addition, the programs in the above-described embodiments can be installed or loaded onto a computer through various recording media (computer-readable recording media) such as optical disks such as CDs and DVDs, magnetic disks, and semiconductor memories, or by downloading them from a server device (medium) via a communication network, etc. [Explanation of symbols]

[0114] 1...information processing system, 2...server device, 3...terminal device, 21...processor, 22...communication interface, 23...storage device, 31...processor, 32...communication interface, 33...storage device, 34...input device, 35...display device, 211...control unit, 231...program, 232...data, 233...large scale language model (LLM), 311...control unit, 331...program

Claims

1. means for receiving a question from a terminal device operated by the user to ask what the user wants to know; A means for extracting content including a word constituting the received question and a word similar to the word constituting the question from published information; a means for generating an answer to the question sentence by inputting a first instruction to the large-scale language model, the first instruction requesting the large-scale language model to generate an answer to the question sentence using the extracted content, the question sentence, and the content to be input; means for providing the generated response to the terminal device; An information processing device comprising:

2. the answer generating means, when inputting the first instruction to the large-scale language model, further inputs a second instruction to request the large-scale language model to generate a related question sentence, which is a question sentence related to a wording included in the generated answer, based on a wording included in the answer, thereby further generating the related question sentence; The extracting means further extracts related content, which is content including a phrase constituting the generated related question sentence and a phrase similar to the phrase constituting the related question sentence, from the published information; the answer generating means further inputs a third instruction to the large-scale language model to request the large-scale language model to generate a related answer that is an answer to the related question sentence, using the extracted related content, the related question sentence, and the related content to be input, thereby further generating the related answer to the related question sentence; The providing means further provides the generated related answers.

2. The information processing device according to claim 1.

3. the answer generating means, when inputting the first instruction to the large-scale language model, further inputs a fourth instruction to request the large-scale language model to generate an additional question, the additional question being a question asking about a recent topic or news related to a word constituting the input question, thereby further generating the additional question; The extracting means further extracts additional content, which is content including a word that constitutes the generated additional question sentence and a word that is similar to the word that constitutes the additional question sentence, from the published information; the answer generating means further inputs a fifth instruction to the large-scale language model to request the large-scale language model to generate an additional answer that is an answer to the additional question sentence, using the extracted additional content, the additional question sentence, and the additional content to be input, thereby further generating the additional answer to the additional question sentence; The providing means further provides the generated additional answer.

3. The information processing device according to claim 2.

4. the providing means inputs a sixth instruction to the large-scale language model to request the large-scale language model to generate a report compiled in a predetermined report format using the generated answer, the related answer, and the additional answer, as well as the answer, the related answer, and the additional answer to be input, and provides the report generated using information included in the answer, the related answer, and the additional answer.

4. The information processing device according to claim 3.

5. The published information is vectorized, the extracting means vectorizes the question sentence and extracts the content from the published information based on the vectorized question sentence; 2. The information processing device according to claim 1.

6. When extracting the content, the extracting means identifies a proper noun from morphemes constituting the question sentence, acquires synonyms of the proper noun based on the identified proper noun, and narrows down and extracts the content containing the identified proper noun and the acquired synonyms.

2. The information processing device according to claim 1.

7. the means for generating an answer further inputs, when inputting the first instruction to the large-scale language model, a seventh instruction for requesting the large-scale language model not to output the answer if there is no valid information in the extracted content; When the answer generating means does not output the answer from the large-scale language model, the extracting means expands a target period of the published information to be searched, further extracts the content from the published information within the expanded target period, and causes the answer generating means to generate the answer again.

2. The information processing device according to claim 1.

8. the providing means downloads the generated answer to the terminal device as editable document data.

2. The information processing device according to claim 1.

9. The providing means compares morphemes constituting the answer generated using the extracted content with the text of the content used to generate the answer, and further provides the answer in a manner that allows visual identification of parts of the generated answer that are expressed in a manner different from the content used to generate the answer.

2. The information processing device according to claim 1.

10. The published information is information published in an electronic version of a newspaper or in a newspaper print version, and the content is an article; 2. The information processing device according to claim 1.

11. Computer, means for receiving a question from a terminal device operated by the user to ask what the user wants to know; A means for extracting content including the wording that constitutes the received question and wording similar to the wording that constitutes the question from published information; a means for generating an answer to the question sentence by inputting a first instruction to the large-scale language model, the first instruction requesting the large-scale language model to generate an answer to the question sentence using the extracted content, the question sentence, and the content to be input; means for providing the generated response to the terminal device; An information processing program that functions as a

Images