Information processing system, information processing method, and program

Abstract

To provide an AI system, method and program capable of acquiring an appropriate answer to a user's question from a generation information processor.SOLUTION: The prompt generation system includes a document information acquisition unit configured to acquire document data from a business entity device, a title specifying unit configured to specify a character string satisfying a predetermined title condition as a title of a sentence included in the document data, a body specifying unit configured to specify a body of the sentence based on the specified title, a storage unit configured to store the specified title and the body in association with each other as divided data in a document database, a question information acquisition unit configured to acquire question information indicating question content from a user, and a processor configured to execute a process including: And a transmission / reception part for transmitting the question information and the first divided information to the generation AI in association with each other so that answer information to the question information with reference to the first divided information can be acquired from the generation AI.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing system, an information processing method, and a program. [Background technology]

[0002] A caisson management system that outputs answer information to questions from users has been disclosed (Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7497538 Summary of the Invention [Problem to be solved by the invention]

[0004] The caisson management system described in Patent Document 1 acquires structure information indicating the state of the caisson structure and question information indicating a question. The caisson management system associates and stores the structure information with history information indicating work history. The caisson management system acquires similar structure information that is similar to the structure information related to the question information from the history information. The caisson management system inputs the question information and the similar structure information into a trained model and acquires answer information indicating an answer to the question information regarding pneumatic caisson construction methods from the trained model. Here, the trained model may be a large-scale language model, or may use a RAG (Retrieval-Augmented Generation) framework. This makes it possible to acquire answers to users' questions.

[0005] However, when using the RAG framework, the caisson management system described in Patent Document 1 has difficulty creating appropriately delimited external information (for example, information stored in a specific company's database) related to the question information (for example, external information delimited only by the number of characters).As a result, the caisson management system is unable to input appropriate prompts to the generation AI, which causes the problem of not being able to obtain an appropriate answer from the generation AI.

[0006] Therefore, in order to solve the above problems, the present invention aims to provide a system that can obtain appropriate answers to user questions from a generation AI device. [Means for solving the problem]

[0007] An information processing system according to one embodiment of the present invention includes a title identification unit that identifies a character string that satisfies a predetermined title condition from among character strings contained in document data related to a predetermined business stored in the business database of the predetermined business as the title of a sentence contained in the document data; a body identification unit that identifies the body of the sentence based on the title identified by the title identification unit; a memory unit that associates the title identified by the title identification unit with the body and stores them as split data in a document database; a question information acquisition unit that acquires question information indicating the content of a question from a user; a split data identification unit that identifies first split data from among the multiple split data stored in the document database, the first split data including a character string related to the question content of the question information; and a transmission unit that associates the question information with the first split data and transmits it to the generation AI so that answer data to the question information based on the first split data can be acquired from the generation AI.

[0008] An information processing method according to one embodiment of the present invention includes a computer executing the following steps: identifying, from among strings contained in document data relating to a specified business stored in the business database of the specified business, a string that meets specified title conditions as the title of a sentence contained in the document data; identifying the body of the sentence based on the identified title; associating the identified title with the body and storing them as split data in a document database; obtaining question information indicating the content of the question from a user; identifying, from among the multiple split data stored in the document database, a first split data that includes a string related to the question content of the question information; and associating the question information with the first split data and sending it to the generation AI so that answer data to the question information based on the first split data can be obtained from the generation AI.

[0009] A program according to one embodiment of the present invention causes a computer to perform the following steps: identify, from among strings contained in document data related to a specified business stored in the business database of the specified business, a string that meets specified title conditions as the title of a sentence contained in the document data; identify the body of the sentence based on the identified title; associate the identified title and the body and store them in a document database as split data; obtain question information indicating the content of the question from a user; identify, from among the multiple split data stored in the document database, a first split data that includes a string related to the question content of the question information; and associate the question information with the first split data and send it to the generation AI so that answer data to the question information based on the first split data can be obtained from the generation AI. [Effects of the Invention]

[0010] According to the present invention, a system can be provided that can obtain appropriate answers to user questions from a generation AI device. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 illustrates an overview of a prompt generation system. [Figure 2] 1 is a database showing an example of a document database 102a. [Figure 3] 10 is a database showing an example of a condition database 102b. [Figure 4] 10 is a flowchart showing the processing steps of the prompt generation system. [Figure 5] FIG. 2 is a diagram illustrating an example of document data. [Figure 6] FIG. 2 illustrates an example of a hardware configuration of a computer. DETAILED DESCRIPTION OF THE INVENTION

[0012] A prompt generation system 10 according to one embodiment of the present invention will be described in detail below with reference to the accompanying drawings. However, the embodiment described below is merely illustrative and is not intended to exclude various modifications or applications of techniques not explicitly described below. In other words, the present invention can be implemented by various modifications or combinations of the embodiments without departing from the spirit of the invention. Furthermore, in the following description of the drawings, identical or similar parts are denoted by identical or similar reference numerals.

[0013] In this embodiment, the terms "unit," "device," and "system" do not simply refer to physical means, but also include cases where the functions of the "unit," "device," or "system" are realized by software. Furthermore, the functions of one "unit," "device," or "system" may be realized by two or more physical means or devices, and the functions of two or more "units," "devices," or "systems" may be realized by a single physical means or device. Furthermore, the various functions described below of each of the multiple devices that make up prompt generation system 10 may be configured to be executed by other devices in the multiple devices.

[0014] === Overview of Prompt Generation System 10 === <<Configuration Overview>> An overview of the prompt generation system 10 will be described with reference to Fig. 1. Fig. 1 is a diagram showing an overview of the prompt generation system 10.

[0015] The prompt generation system 10 is a system that enables users to obtain appropriate answers from a large-scale language model (hereinafter referred to as "generative AI") that has a huge data set based on a deep learning model by outputting appropriate prompts to the generative AI.

[0016] The prompt generation system 10 includes, for example, a prompt generation device 100, a business operator device 200, and a user terminal 300.

[0017] The prompt generation device 100 generates appropriate prompts for user questions by generating appropriately divided data based on data indicating documents (hereinafter referred to as "document data") stored in the storage unit 210 of the business operator device 200. The prompt generation device 100 outputs the generated prompts to a generation AI (Artificial Intelligence).

[0018] Document data is, for example, data such as regulations specific to a particular business (for example, rules and regulations such as work regulations, various manuals) stored in a database of the business.

[0019] The business operator device 200 is a device of a business operator such as a company.

[0020] A generation AI is a text generation AI that is trained to extract appropriate information from specified data and generate appropriate sentences when a prompt containing information indicating the user's question (hereinafter referred to as "question information") is input, such as ChatGPT or Google Bard.

[0021] In this way, the prompt generation system 10 appropriately divides data indicating internal documents of at least one business to generate divided data, and outputs a prompt that associates the appropriately divided divided data with question information to the generation AI. This allows the prompt generation system 10 to obtain from the generation AI an appropriate answer desired by, for example, a user belonging to the business. In other words, the prompt generation system 10 is a system that appropriately implements the so-called RAG (Retrieval-Augmented Generation) framework.

[0022] The prompt generation device 100 and the business operator device 200 may be, for example, a cloud computer, a server computer, a personal computer (e.g., a desktop, laptop, tablet, etc.), a media computer platform (e.g., a cable or satellite set-top box, a digital video recorder), a handheld computer device (e.g., a PDA, an email client, etc.), or any other type of computer or communication platform. Note that at least a portion of the processing in the prompt generation device 100 and the business operator device 200 may be implemented by one or more computers (for example, but not limited to, cloud computing configured by one or more computers).

[0023] The user terminal 300 is a device that receives user operation input and displays various types of information.

[0024] The user terminal 300 may be, for example, a smartphone, a mobile phone (feature phone), a personal computer (e.g., a desktop, laptop, tablet, etc.), a media computing platform (e.g., a cable or satellite set-top box, a digital video recorder), a handheld computing device (e.g., a personal digital assistant (PDA), an email client, etc.), a wearable device (e.g., a glasses-type device, a watch-type device, etc.), or another type of computer or communication platform.

[0025] <<Processing Overview>> Referring to FIG. 1, an overview of the processing of the prompt generation system 10 will be described.

[0026] In step S10, the prompt generation device 100 acquires document data from the business device 200.

[0027] In step S11, the prompt generation device 100 divides the document data into well-defined data segments using a well-known processing technique.

[0028] Here, the divided data with good divisions is, for example, a collection of data divided into paragraphs or lines.

[0029] In step S12, the prompt generation device 100 identifies the title of the sentence included in the segment data. The prompt generation device 100 also identifies the body of the sentence related to the title. The process of identifying the title and body will be described in detail later.

[0030] A title is, for example, a character string that is the title of a sentence included in document data, such as a chapter, section, paragraph, or article. Specifically, the title may be, for example, "Section 1" or "Article 1," or may be "Section 1" or "Article 1" with a character string indicating the title of the sentence added, such as "Section 1 AAA" or "(BBB) ​​Article 1."

[0031] The main text is the text that is subordinate to the title. For example, if the title is "Section 1 AAA," the main text is the text that indicates the content of "Section 1."

[0032] The prompt generation device 100 then identifies a level of sentence structure for each title. The process of identifying a level of document structure will be described in detail later.

[0033] The level of the text structure is an index showing the superior-subordinate relationship of the text structure. For example, if a "chapter" is at a higher level, then a "section" subordinate to the "chapter" is at a lower level.

[0034] In the following description, as an example, a higher level title will be referred to as a "higher level title" and a lower level title subordinate to the higher level title will be referred to as a "subordinate title."

[0035] As an example, the "higher title" is "Chapter X AAA" and the subordinate title is "Section Y BBB." In other words, the following description will be given assuming that the document data has a sentence structure including a higher title "Chapter" and a subordinate title "Section" that is subordinate to the "Chapter."

[0036] As another example, divided data including a main text subordinate to a "higher title" is referred to as a "higher divided data," and divided data including a main text subordinate to a "subordinate title" is referred to as a "subordinate divided data."

[0037] The prompt generation device 100 stores segmented data including a title and a body for each level in the storage unit 102. At this time, the prompt generation device 100 converts the segmented data into vector information and stores it in the storage unit 102, for example.

[0038] In step S13, the prompt generation device 100 acquires question information indicating the content of the user's question from the user terminal 300. The prompt generation device 100 converts the question information into vector information. The prompt generation device 100 extracts lower-level subordinate divided data indicating vector information similar to the vector information of the question information from the storage unit 102. The prompt generation device 100 identifies the higher-level divided data to which the subordinate divided data belongs.

[0039] In this way, the prompt generation device 100 stores appropriately divided segmented data that includes a title and a body of text subordinate to the title, and can determine the similarity between segmented data that is meaningful as a sentence and question information. This enables the prompt generation device 100 to extract appropriate segmented data to send to the generation AI, thereby improving the accuracy of the answer obtained from the generation AI.

[0040] In step S14, the prompt generation device 100 generates a prompt by associating the question information with the higher-level divided data, and transmits the prompt to the generation AI.

[0041] A prompt is a sentence to be input to the generation AI, that is, data indicating the command to be given to the generation AI.

[0042] In this way, the prompt generation device 100 identifies dependent divided data similar to the question information and transmits to the generation AI the higher-level divided data to which the dependent divided data is subordinate. This allows the prompt generation device 100 to transmit to the generation AI a wider range of sentences than the dependent divided data, which are likely to include sentences related to the question information, thereby improving the accuracy of the answers obtained from the generation AI.

[0043] In step S15, the prompt generation device 100 acquires, from the generation AI, data indicating the answer text generated by the generation AI (hereinafter referred to as "answer data").

[0044] In step S16, the prompt generation device 100 transmits the answer data to the user terminal 300.

[0045] This allows the prompt generation system 10 to appropriately acquire information desired by the user from the generation AI.

[0046] ===Prompt Generator 100=== As shown in FIG. 1, the prompt generation device 100 includes a document acquisition unit 101, a memory unit 102, a document division unit 103, a title identification unit 104, a body identification unit 105, a divided data generation unit 106, a question information acquisition unit 107, a divided data identification unit 108, a prompt generation unit 109, and a transmission / reception unit 110.

[0047] The document acquisition unit 101 acquires document data from the business entity device 200 .

[0048] The storage unit 102 includes, for example, a document database D102a.

[0049] The document database D102a will be described with reference to Fig. 2. Fig. 2 shows an example of the document database D102a. The document database D102a is a database that stores document data and higher-level divided data and lower-level divided data included in the document data in association with each other. The document database D102a is a database for generating prompts to be sent to the generation AI.

[0050] The document database D102a includes items such as [Business ID], [Document ID], [Document Data], [Higher Segment Data], and [Subordinate Segment Data]. [Business ID] stores identification information that can uniquely identify a business. [Document ID] stores identification information that can uniquely identify document data. [Document Data] stores document data. [Higher Segment Data] stores higher-level segment data identified by the title identification unit 104 and the body identification unit 105. [Subordinate Segment Data] stores subordinate segment data (e.g., sentences included in "Section 1") that is subordinate to higher-level segment data (e.g., sentences included in "Chapter 1") identified by the title identification unit 104 and the body identification unit 105.

[0051] The condition database D102b will be described with reference to Fig. 3. Fig. 3 shows an example of the condition database D102b. The condition database D102b is a database that stores conditions for character strings that indicate titles (hereinafter referred to as "title conditions"). The document database D102a is a database for identifying character strings that can become titles (hereinafter referred to as "title candidates") and titles.

[0052] The condition database D102b includes items such as [Title Condition ID] and [Title Condition]. [Title Condition ID] stores identification information that can uniquely identify a title condition. [Title Condition] stores information that indicates a title condition.

[0053] The title conditions include, for example, conditions regarding the coordinates of character strings included in the divided data obtained by dividing document data (hereinafter referred to as "coordinate conditions") and conditions regarding the characters of the character strings (hereinafter referred to as "character string conditions").

[0054] The coordinate conditions are, for example, title conditions related to title IDs "001" and "002" in Fig. 3. Specifically, the coordinate conditions are conditions related to, for example, the vertical width of a line of character strings (for example, "line height" in Fig. 3, which may be 150% or more compared to the average vertical width of character strings included in the document data), the position of the left edge of the character string on the paper (for example, "left edge position" in Fig. 3, which may be to the left of the average position of character strings included in the document data), the vertical position of the line (for example, "vertical position" in Fig. 3, which may be above the average vertical position of character strings included in the document data), etc.

[0055] The character string conditions are, for example, title conditions related to title IDs "010" to "016" in Fig. 3, and are, for example, conditions related to combinations of letters and numbers. Specifically, the character string conditions include conditions related to combinations of letters and numbers, such as "(number) Chapter Title," "(number) Section Title," "(number) Article Title," "(number)-(number)-(number) Title," and "(number), (number), (number) Title."

[0056] By using the document database D102a, the prompt generation device 100 can appropriately extract the divided data to be included in the prompt for obtaining appropriate answer data from the generation AI.

[0057] The document divider 103 analyzes the document structure of the document data using a known processing technique and divides the document data. The document divider 103 may, for example, execute a process of dividing the document data by a predetermined number of characters, or may execute a process of dividing the document data by a format such as a paragraph, a line, a character, or a table.

[0058] The title identifying unit 104 identifies the title included in the document data. The title identifying unit 104 may identify the title of the text based on the data obtained by dividing the document data by the document dividing unit 103.

[0059] For example, the title identification unit 104 identifies title candidates, which are character strings that can become titles, from among the character strings included in the document data based on title conditions, which are conditions for character strings that indicate a title, and identifies the title based on the title candidates.

[0060] First, the title identification unit 104 identifies, as title candidates, character strings that satisfy a coordinate condition from among character strings included in the document data. Specifically, the title identification unit 104 identifies, as title candidates, character strings that are located to the left of the average of character strings in the document data among character strings in the data divided by the document division unit 103.

[0061] Next, if the number of title candidates among the plurality of title candidates that satisfy the character string condition is equal to or greater than a predetermined percentage (threshold) of the number of the plurality of title candidates, the title identification unit 104 identifies the title candidates that satisfy the character string condition as the title. Specifically, for example, if the title candidates that satisfy the character string conditions of "(number) Chapter Title" and "(number) Section Title" account for 10% or more of the total title candidates, the title identification unit 104 identifies the title candidates that satisfy the character string condition (here, title candidates including "(number) Chapter" or "(number) Section") as the title.

[0062] On the other hand, when the number of title candidates among the plurality of title candidates that satisfy the character string condition is less than a predetermined ratio (threshold) of the number of the plurality of title candidates, the title identification unit 104 may identify the character string of the title candidate that satisfies the coordinate condition as the title. Specifically, when the title candidates that satisfy the character string condition are less than 10% of the total title candidates, the title identification unit 104 identifies the title candidate (for example, a title candidate such as a vertically long character string or a character string that is shifted to the left) as the title.

[0063] In this way, the prompt generation system 10 identifies frequently appearing title candidates as titles, making it possible to identify titles that appropriately capture the document structure of the document data. This allows the prompt generation system 10 to appropriately identify the main text and generate well-delimited divided data, making it possible to obtain appropriate answers from the generation AI.

[0064] Next, the title identification unit 104 identifies the level of the identified title in the document structure. Here, the title identification unit 104 identifies the level of the title based on, for example, a character string condition. Specifically, the title identification unit 104 identifies the level based on the character string condition in the order in which titles appear in the document data, starting from the top of the document indicated by the document data. For example, if the title that appears first matches a first character string condition (e.g., "Chapter (number).<Title>" with condition ID "012" in FIG. 3) among the character string conditions, the title identification unit 104 identifies the title as a first-level title (hereinafter referred to as a "higher title"). Next, if the title that appears next after the higher-level title matches a second character string condition different from the first character string condition (e.g., "Section (number).<Title>" with condition ID "015" in FIG. 3), the title identification unit 104 identifies the title as a second-level title (hereinafter referred to as a "subordinate title") that is lower than the first level.

[0065] In this way, by appropriately analyzing the document structure of the document data, the prompt generation system 10 is able to generate a prompt that includes an appropriate range of split data that is similar to the question information, thereby enabling the generation AI to obtain an appropriate answer.

[0066] The main text identifying unit 105 identifies the main text of the sentence related to the title included in the document data, based on the title identified by the title identifying unit 104. Specifically, the main text identifying unit 105 identifies the main text between the first subordinate title and the second subordinate title that appears after the first subordinate title as the main text related to the first subordinate title.

[0067] The divided data generation unit 106 associates the title identified by the title identification unit 104 with the body text identified by the body text identification unit 105 to generate divided data according to the title level. The divided data generation unit 106 stores the divided data in the document database D102a. Specifically, the divided data generation unit 106 associates, for example, a higher-level title with a body text subordinate to the higher-level title to generate higher-level divided data. The divided data generation unit 106 also associates, for example, a subordinate title with a body text subordinate to the subordinate title to generate subordinate divided data. The divided data generation unit 106 associates, for example, the subordinate divided data with the higher-level divided data and stores the data in the document database D102a.

[0068] The question information acquisition unit 107 acquires, from the user terminal 300, question information indicating the content of the user's question.

[0069] The divided data identification unit 108 refers to the document database D102a to identify divided data related to the question information. Specifically, the divided data identification unit 108 identifies, for example, vector information indicating the text content of the dependent divided data that is similar to the vector information indicating the question information. The divided data identification unit 108 then identifies higher-level divided data to which the dependent divided data corresponding to the identified vector information is subordinate.

[0070] In this way, the prompt generation device 100 can identify appropriate segmented data by performing a similarity search on dependent segmented data that has a sentence structure that is appropriately divided into sentence blocks. However, if the segmented data to be compared with the question information is not appropriately divided, making the meaning of the sentence unclear, a problem arises in that segmented data similar to the question information cannot be appropriately identified. The prompt generation device 100 can solve this problem.

[0071] Furthermore, because the prompt generation device 100 performs structural analysis by dividing each piece of segmented data into levels, it can transmit to the generation AI a wider range of sentences from higher-level segmented data, including segmented data written near the identified subordinate segmented data. This enables the prompt generation device 100 to obtain an appropriate answer from the generation AI.

[0072] The prompt generation unit 109 generates a prompt by associating question information acquired from the user terminal 300 with higher-level segmented data that is highly relevant to the question information. Specifically, the prompt generation unit 109 first identifies dependent segmented data that is similar to the question information based on vector information of the question information. The prompt generation unit 109 then identifies higher-level segmented data to which the identified dependent segmented data is subordinate. The prompt generation unit 109 then generates a prompt by associating the question information with the higher-level segmented data. In other words, the prompt generation unit 109 generates a prompt that includes data that is highly relevant to the question information from among data included in the document data of the business. Furthermore, because the higher-level segmented data included in the prompt is segmented data in which sentences are appropriately divided, the prompt generation device 100 can acquire appropriate answer data from the generation AI.

[0073] The transmitting / receiving unit 110 transmits a prompt that associates the question information with the higher-level divided data to the generation AI so that answer data similar to the content of the question information and based on divided data included in the document data can be acquired. The transmitting / receiving unit 110 receives the answer data from the generation AI. The transmitting / receiving unit 110 transmits the answer data to the user terminal 300.

[0074] <<First Modification>> The title identification unit 104 may identify the title by identifying the table of contents in the document data. In this case, the title identification unit 104 identifies that a table of contents is included in a predetermined page of the document data, for example, based on a condition indicating the table of contents (hereinafter referred to as a "table of contents condition").

[0075] The table of contents condition may be a first table of contents condition indicating that a specified page of the document data contains a string indicating a table of contents (e.g., "Table of Contents"), or may be a second table of contents condition indicating that numbers are arranged vertically on one side of the horizontal direction of the specified page.

[0076] For example, when the title identification unit 104 identifies a table of contents, if the first title in the character strings included in the table of contents matches the first character string condition described above, it identifies that title as a higher-level title. Next, if the title that appears next to the higher-level title matches the second character string condition described above, the title identification unit 104 identifies that title as a second-level subordinate title.

[0077] On the other hand, if the table of contents is not specified, the title specifying unit 104 specifies the level based on the character string condition in the order in which the titles appear in the document data, as described above, to specify the higher title and subordinate titles.

[0078] In this way, the prompt generation device 100 can reduce the amount of processing required to identify a title by identifying the table of contents and then identifying the title.

[0079] <<Second Modification>> The divided data generation unit 106 may generate the higher-level divided data and the subordinate divided data so that the size does not exceed a predetermined size. Specifically, the divided data generation unit 106 determines whether the number of characters in the main text associated with each of the higher-level divided data and the subordinate title is equal to or greater than a predetermined number of characters. If the number of characters is less than the predetermined number, the divided data generation unit 106 does not adjust the size of the divided data. On the other hand, if the number of characters is equal to or greater than the predetermined number, the divided data generation unit 106 adjusts the size of the divided data.

[0080] For example, when adjusting the size of the dependent divided data, if the dependent divided data includes multiple paragraphs, the divided data generation unit 106 may divide the dependent divided data at the end of a predetermined paragraph so that the number of characters is within a predetermined range. Also, if the dependent divided data includes multiple titles that are lower in rank than the dependent title, the dependent divided data may be divided at the end of a predetermined title among the multiple titles so that the number of characters is within a predetermined range.

[0081] For example, when adjusting the size of higher-level split data, the split data generation unit 106 may separate the higher-level split data at the end of a specified dependent split data so that it fits within a specified number of characters (a number of characters greater than the specified number of characters for the dependent split data).

[0082] In this way, the prompt generation device 100 divides the higher-level split data and subordinate split data into split data of an appropriate size with good divisions, thereby enabling the generation AI to generate appropriate prompts that will enable it to generate highly accurate answer data.

[0083] <<Third Modification>> In the above, the prompt generation device 100 has been described as being configured to include the document divider 103, but it need not include the document divider 103. In this case, the prompt generation device 100 may identify titles for the document data in the title identification unit 104, and identify the sentences between the titles as the main text in the main text identification unit 105, and generate divided data in which the titles and the main text are associated with each other.

[0084] ===Business equipment 200=== 1, the following describes the configuration of the business operator device 200. The business operator device 200 is, for example, a device that stores document data and transmits the document data to the prompt generation device 100.

[0085] 1, the provider device 200 includes, for example, a storage unit 210, a transmission / reception unit 220, and a processing unit 230. Each functional unit is implemented, for example, by the processor 1001 reading out a program stored in the memory 1002. The storage unit 210 stores various types of information. The transmission / reception unit 220 transmits and receives various types of information to and from the prompt generation device 100. The processing unit 230 executes various types of processing.

[0086] ===User terminal 300=== The configuration of the user terminal 300 will be described with reference to Fig. 1. As shown in Fig. 1, the user terminal 300 includes functional units, such as a storage unit 310, a transmission / reception unit 320, and a display processing unit 330. Each functional unit is implemented, for example, by the processor 1001 reading out a program stored in the memory 1002. The storage unit 310 stores various types of information. The transmission / reception unit 320 transmits and receives various types of information to and from the prompt generation device 100. The display processing unit 330 displays various types of information obtained from the prompt generation device 100 on the display unit.

[0087] ===Processing Procedure=== The processing procedure of the prompt generation system 10 will be described with reference to Figures 4 and 5. Figure 4 is a flowchart showing the processing procedure of the prompt generation system 10. Figure 5 is a diagram showing an example of document data.

[0088] In step S100, the prompt generation device 100 acquires document data from the business operator device 200. The prompt generation device 100 stores the document data in the document database D102a.

[0089] In step S101, the prompt generation device 100 refers to the document database D102a and determines whether or not the document data includes a table of contents based on the table of contents condition.

[0090] If it is determined that a table of contents is included (step S101: Yes), then in step S102, the prompt generation device 100 identifies the character string included in the table of contents as the title. Here, it is assumed that the document data does not include a table of contents, as shown in FIG.

[0091] If it is determined that a table of contents is not included (step S101: No), in step S103, the prompt generation device 100 refers to the condition database D102b and identifies title candidates from the character strings included in the document data based on the coordinate conditions.

[0092] In step S104, the prompt generation device 100 refers to the condition database D102b and determines whether the proportion of the identified title candidates that satisfy the character conditions is equal to or greater than a predetermined threshold.

[0093] If it is less than the predetermined threshold (step S104: less than threshold), then in step S105 the prompt generation device 100 identifies the title candidate as the title.

[0094] If the result is equal to or greater than the predetermined threshold (step S104: equal to or greater than threshold), in step S106 the prompt generation device 100 identifies as title candidates those whose proportion of satisfying the character conditions is determined to be equal to or greater than the threshold. In this example, it is assumed that "Chapter (number) Title" and "Section (number) Title" are identified as titles.

[0095] In step S107, the prompt generation device 100 identifies the level of the document structure for each of the identified titles. As shown in Fig. 5, the prompt generation device 100 identifies the titles as higher-level titles and subordinate titles subordinate to the higher-level titles.

[0096] In step S108, the prompt generation device 100 identifies the body text (for example, "Body Text" in FIG. 5) associated with each title.

[0097] In step S109, the prompt generation device 100 generates higher-level segmented data (for example, "B10" in FIG. 5) and subordinate segmented data (for example, "B20" to "B22" in FIG. 5) based on the title, body, and title level, and stores them in the document database D102a. This allows the prompt generation device 100 to generate segmented data for appropriately divided blocks of sentences.

[0098] In step S 110 , the prompt generation device 100 acquires question information from the user terminal 300 .

[0099] In step S111, the prompt generation device 100 refers to the document database D102a and identifies dependent divided data items that are similar to the text content of the question information, based on the vector information of the question information.

[0100] In step S112, the prompt generation device 100 refers to the document database D102a to identify the higher-level divided data to which the identified subordinate divided data belongs.

[0101] In step S113, the prompt generation device 100 generates a prompt by associating the question information with the higher-level divided data, and transmits the prompt to the generation AI.

[0102] In step S114, the prompt generation device 100 obtains answer data from the generation AI.

[0103] In step S 115 , the prompt generation device 100 transmits the answer data to the user terminal 300 .

[0104] In step S116, the user terminal 300 displays the response data on the display unit.

[0105] In this way, the prompt generation system 10 identifies split data similar to the question information using split data (e.g., dependent split data) in which sentences are appropriately divided, and therefore, since appropriate split data can be identified, it becomes possible to obtain appropriate answer data from the generation AI.

[0106] Furthermore, the prompt generation system 10 does not send only the divided data similar to the question information to the generation AI, but also sends a prompt including the divided data (higher divided data) before and after the divided data similar to the question information, because there is a high possibility that information related to the question information is included before and after the divided data (subordinate divided data).This also enables the prompt generation system 10 to obtain appropriate answer data from the generation AI.

[0107] ===Hardware Configuration=== An example of the hardware configuration when the prompt generation device 100, the provider device 200, and the user terminal 300 are implemented by a computer will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of the hardware configuration of a computer.

[0108] As shown in FIG. 6, the computer 1000 includes a processor 1001, a memory 1002, a storage device 1003, an input I / F unit 1004, a data I / F unit 1005, a communication I / F unit 1006, and a display unit 1007.

[0109] The processor 1001 is a control unit that controls various processes in the computer 1000 by executing programs stored in the memory 1002 .

[0110] The memory 1002 is a storage medium such as a RAM (Random Access Memory), etc. The memory 1002 temporarily stores the program code of the program executed by the processor 1001 and data required when the program is executed.

[0111] The storage device 1003 is a non-volatile storage medium such as a hard disk drive (HDD), flash memory, etc. The storage device 1003 stores an operating system and various programs for realizing the above-mentioned components.

[0112] The input I / F unit 1004 is a device for receiving input from a user. Specific examples of the input I / F unit 1004 include a keyboard, a mouse, a touch panel, various sensors, and a wearable device. The input I / F unit 1004 may be connected to the computer 1000 via an interface such as a USB (Universal Serial Bus).

[0113] The data I / F unit 1005 is a device for inputting data from outside the computer 1000. A specific example of the data I / F unit 1005 is a drive device for reading data stored in various storage media. The data I / F unit 1005 may be provided outside the computer 1000. In this case, the data I / F unit 1005 is connected to the computer 1000 via an interface such as a USB.

[0114] The communication I / F unit 1006 is a device for performing data communication via the Internet N, either wired or wirelessly, with devices external to the computer 1000. The communication I / F unit 1006 may be provided external to the computer 1000. In this case, the communication I / F unit 1006 is connected to the computer 1000 via an interface such as a USB.

[0115] The display unit 1007 is a device for displaying various types of information. Specific examples of the display unit 1007 include a liquid crystal display, an organic EL (Electro-Luminescence) display, and a display of a wearable device. The display unit 1007 may be provided outside the computer 1000. In this case, the display unit 1007 is connected to the computer 1000 via, for example, a display cable. Furthermore, when a touch panel is used as the input I / F unit 1004, the display unit 1007 can be configured as an integrated unit with the input I / F unit 1004.

[0116] ===Summary=== <1> The prompt generation system 10 of this embodiment includes a title identification unit 104 that identifies, as the title of a sentence included in the document data, a character string that satisfies a predetermined title condition from among character strings included in document data related to a predetermined business stored in the business database of the predetermined business; a body identification unit 105 that identifies the body of the sentence based on the title identified by the title identification unit 104 (e.g., a subordinate title or a superordinate title); and a document database D102a that associates the title identified by the title identification unit 104 with the body of the sentence and stores the associated data as divided data (e.g., superordinate divided data or subordinate divided data). The prompt generation system 10 includes a memory unit 102 for storing the question information, a question information acquisition unit 107 for acquiring question information indicating the content of the question from the user, a divided data identification unit 108 for identifying subordinate divided data or higher-level divided data (first divided data) that includes a character string related to the question content of the question information from among the multiple divided data stored in the document database D102a, and a transmission / reception unit 110 (transmission unit) for associating the question information with the subordinate divided data or higher-level divided data (first divided data) and transmitting the associated data to the generation AI so that answer data to the question information can be acquired from the generation AI using the subordinate divided data or higher-level divided data (first divided data). This allows the prompt generation system 10 to appropriately acquire information desired by the user from the generation AI.

[0117] <2> Furthermore, in the prompt generation system 10 of this embodiment, the title identification unit 104 identifies, from among the character strings included in the document data, character strings that satisfy a coordinate condition, which is a condition related to the coordinates of the character string among the title conditions, as title candidates that can become the title; and if the number of title candidates among the multiple title candidates that satisfy a character string condition, which is a condition related to the characters included in the character string among the title conditions, is equal to or greater than a predetermined ratio of the number of multiple title candidates, the title candidate that satisfies the character string condition is identified as the title. <1> The prompt generation system according to the present invention is capable of generating appropriately divided data, thereby enabling the generation AI to provide an appropriate response.

[0118] <3> Furthermore, in the prompt generation system 10 of this embodiment, if the number of title candidates among the plurality of title candidates that satisfy the character string condition is less than a predetermined ratio of the number of the plurality of title candidates, the title identification unit 104 identifies the character string of the title candidate that satisfies the coordinate condition as a title (e.g., a subordinate title or a superior title). <2> The prompt generation system according to the present invention is capable of generating appropriately divided data, thereby enabling the generation AI to provide an appropriate response.

[0119] <4> Furthermore, in the prompt generation system 10 of this embodiment, the title identification unit 104 identifies, in the order in which the titles identified by the title identification unit 104 appear in the document data, titles that match a first character string condition, which is a condition regarding characters contained in a character string, among the title conditions, as higher-level titles that are titles at a first level in the document structure of the document data, and identifies titles that appear after the higher-level titles and match a second character string condition different from the first character string condition as titles at a second level lower than the first level in the document structure of the document data. the storage unit associates the superordinate title with the subordinate title and stores them in a document database; the divided data identification unit 108 identifies, based on the document database, the divided data including the subordinate title, which includes a character string related to the question content of the question information, as subordinate divided data (second divided data); the transmission / reception unit 110 (transmission unit) transmits the superordinate divided data (first divided data) including the subordinate divided data (second divided data) and the superordinate title related to the subordinate divided data (second divided data), and the question information, to the generation AI; <1> from <3> The prompt generation system according to any one of the preceding claims, wherein the prompt generation system 10 can appropriately analyze the document structure of the document data, thereby enabling the generation AI to obtain an appropriate answer by generating a prompt that includes an appropriate range of segmented data that is similar to the question information.

[0120] <5> Furthermore, in the prompt generation system 10 of this embodiment, if a specific page of document data satisfies a first table of contents condition, which is a condition indicating that a character string indicating a table of contents is included, or if the specific page satisfies a second table of contents condition, which indicates that numbers are arranged vertically on one of the left and right sides of the page, the title identification unit 104 identifies the specific page as a table of contents that includes a title, and identifies, from among the character strings included in the identified table of contents, a character string that satisfies a character string condition, which is a condition regarding characters included in the character string of the title condition, as the title included in the document data. <1> from <4> 1. A prompt generation system according to claim 1, wherein the prompt generation system 10 identifies the title by identifying the table of contents, thereby reducing the amount of processing required to identify the title. [Explanation of symbols]

[0121] 10...prompt generation system, 100...prompt generation device, 101...document acquisition unit, 102...memory unit, 103...document division unit, 104...title identification unit, 105...body identification unit, 106...divided data generation unit, 107...question information acquisition unit, 108...divided data identification unit, 109...prompt generation unit, 110...transmission / reception unit, 200...operator device, 300...user terminal.

Claims

1. a title specification unit that specifies, from among character strings included in document data related to a predetermined business stored in a business database of the predetermined business, a character string that satisfies a predetermined title condition as the title of a sentence included in the document data; a body identifying unit that identifies a body of the text based on the title identified by the title identifying unit; a storage unit that associates the title identified by the title identification unit with the body text and stores them as divided data in a document database; a question information acquisition unit that acquires question information indicating the content of a question from a user; a divided data identifying unit that identifies a first divided data item that includes a character string related to a question content of the question information from among the plurality of divided data items stored in the document database; a transmitting unit that associates the question information with the first divided data and transmits the first divided data to the generating AI so that answer data to the question information can be acquired from the generating AI by referring to the first divided data; An information processing system comprising:

2. The title specification unit Identifying, from among character strings included in the document data, character strings that satisfy a coordinate condition, which is a condition related to the coordinates of the character string among the title conditions, as title candidates that can become titles; If the number of the title candidates that satisfy a character string condition, which is a condition related to characters contained in a character string among the title conditions, is equal to or greater than a predetermined ratio of the number of the title candidates, the title candidate that satisfies the character string condition is identified as the title. The information processing system according to claim 1 .

3. when the number of the title candidates that satisfy the character string condition among the plurality of title candidates is less than a predetermined ratio of the number of the plurality of title candidates, the title identification unit identifies the character string of the title candidate that satisfies the coordinate condition as the title; The information processing system according to claim 2 .

4. The title specification unit In the document data, the titles identified by the title identification portion appear in the order in which they appear. Identifying the title that matches a first character string condition, which is a condition related to characters included in a character string, among the title conditions, as a higher-level title that is a title at a first level in the document structure of the document data; Identifying the title that appears after the superordinate title and that matches a second character string condition different from the first character string condition as a subordinate title, which is a title at a second level lower than the first level in the document structure of the document data; the storage unit associates the higher-level title with the subordinate title and stores them in the document database; the divided data identifying unit identifies, based on the document database, the divided data including the subordinate title, which includes a character string related to the question content of the question information, as a second divided data; The transmission unit transmits the first divided data including the second divided data and the higher-level title related to the second divided data, and the question information to the generation AI. The information processing system according to claim 1 .

5. The title specification unit If a predetermined page of the document data satisfies a first table of contents condition, which is a condition indicating that a character string indicating a table of contents is included, or if a second table of contents condition, which indicates that numbers are arranged vertically on one of the left and right sides of the page, is satisfied, the predetermined page is identified as a table of contents including the title; Among the character strings included in the identified table of contents, a character string that satisfies a character string condition, which is a condition related to characters included in the character string, among the title conditions, is identified as the title included in the document data. The information processing system according to claim 1 .

6. The computer Identifying a character string that satisfies a predetermined title condition from character strings included in document data related to a predetermined business stored in a business database of the predetermined business as a title of a sentence included in the document data; Identifying the body of the text based on the identified title; storing the identified title and the identified body text in a document database as divided data in association with each other; acquiring question information indicating question content from a user; Identifying a first piece of divided data that includes a character string related to the question content of the question information from among the plurality of pieces of divided data stored in the document database; transmitting the question information and the first divided data to the generation AI in association with each other so that answer data to the question information, which is based on the first divided data, can be acquired from the generation AI; An information processing method that performs the above.

7. On the computer, Identifying a character string that satisfies a predetermined title condition from character strings included in document data related to a predetermined business stored in a business database of the predetermined business as a title of a sentence included in the document data; Identifying the body of the text based on the identified title; storing the identified title and the identified body text in a document database as divided data in association with each other; acquiring question information indicating question content from a user; Identifying a first piece of divided data that includes a character string related to the question content of the question information from among the plurality of pieces of divided data stored in the document database; transmitting the question information and the first divided data to the generation AI in association with each other so that answer data to the question information, which is based on the first divided data, can be acquired from the generation AI; A program that executes the following.

Images