Information processing method, information processing device, and computer program

Abstract

This document provides an information processing method that can clearly present query-related sections in documents concerning equipment and machinery. [Solution] The information processing method according to the first aspect of the present disclosure acquires a query relating to equipment, uses a language model that uses the acquired query and a group of files including multiple split files obtained by dividing a document file relating to the equipment to output file information relating to the split file corresponding to the query, and outputs the split file corresponding to the query after highlighting the part related to the query.

Description

Technical Field

[0001] This technology relates to an information processing method, a computer program, and an information processing apparatus.

Background Art

[0002] Conventionally, an apparatus for outputting search results of documents has been proposed. For example, the document search apparatus described in Patent Document 1 performs a search of the structured document data based on the search phrase.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in Patent Document 1, in a document related to facility equipment, consideration has not been given to presenting clearly the part related to the query.

[0005] The present disclosure has been made in view of such circumstances, and an object thereof is to provide an information processing method or the like capable of presenting clearly the part related to the query in a document related to facility equipment.

Means for Solving the Problems

[0006] The information processing method according to the first aspect of the present disclosure acquires a query related to facility equipment, and uses a language model that uses the acquired query and a file group including a plurality of divided files obtained by dividing a document file related to the facility equipment, outputs file information related to the divided file corresponding to the query, and outputs the divided file corresponding to the query with the part related to the query emphasized.

[0007] The information processing method relating to the second aspect of this disclosure outputs a split file and also outputs a link to access a document file containing the split file.

[0008] The information processing method according to a third aspect of the present disclosure includes, in which the file information includes a link for accessing a segmented file corresponding to the query and a summary of the segmented file summarized by the language model.

[0009] The information processing method according to a fourth aspect of this disclosure inputs a prompt to the language model, which includes an acquired query, an instruction to identify a split file corresponding to the query, an instruction to output an identifier for identifying each split file, an instruction to highlight the portion of the identified split file related to the query, and an instruction to generate a summary of the identified split file.

[0010] The information processing method according to the fifth aspect of this disclosure further includes, in which the file information contains thumbnails of figures, tables, or flowcharts related to the acquired query, which are included in the segmented files corresponding to the query.

[0011] The information processing method according to the sixth aspect of this disclosure, upon receiving a thumbnail selection, outputs a split file containing a figure, table, or flowchart corresponding to the selected thumbnail, with the portion of the figure, table, or flowchart related to the query highlighted.

[0012] The information processing method according to the seventh aspect of this disclosure calculates the similarity between the acquired query and the split files included in the file group, and outputs multiple file information for multiple split files in order of similarity.

[0013] The information processing method according to the eighth aspect of this disclosure obtains a query that includes the model name of the equipment, and inputs the obtained query into the language model.

[0014] The information processing method according to the ninth aspect of this disclosure obtains a query that includes the model name of the equipment and an abnormal code output from the equipment, and inputs the obtained query into the language model.

[0015] An information processing device according to an aspect of the present disclosure includes an arithmetic unit, which acquires a query relating to equipment, and uses a language model that uses the acquired query and a group of files including multiple split files obtained by dividing a document file relating to the equipment to output file information relating to the split file corresponding to the query, and outputs the split file corresponding to the query, with the portion related to the query highlighted.

[0016] The computer program according to the embodiment of this disclosure acquires a query relating to equipment, uses a language model that uses the acquired query and a group of files containing multiple split files obtained by dividing a document file relating to the equipment to output file information relating to the split file corresponding to the query, and outputs the split file corresponding to the query, with the portion related to the query highlighted. [Effects of the Invention]

[0017] In an information processing method according to one embodiment of this disclosure, it is possible to present the portion of a document relating to equipment that is relevant to the query in an easily understandable manner. [Brief explanation of the drawing]

[0018] [Figure 1] This is a schematic diagram showing an example configuration of a document search system according to Embodiment 1. [Figure 2] This is a block diagram showing an example configuration of an information processing device and a language model server. [Figure 3] This is a block diagram showing an example configuration of a terminal device. [Figure 4] This is an explanatory diagram illustrating an example of a second language model. [Figure 5] This is an explanatory diagram showing an example of a synonym database. [Figure 6]It is an explanatory diagram showing an example of an ideal question DB. [Figure 7] It is a flowchart showing the processing procedure of the synonym extraction process. [Figure 8] It is an explanatory diagram showing an example of a third language model. [Figure 9] It is an explanatory diagram showing the extraction of metadata of a diagram or image by a third language model. [Figure 10] It is an explanatory diagram showing the extraction of metadata of a table by a third language model. [Figure 11] It is an explanatory diagram showing the extraction of metadata of a flowchart by a third language model. [Figure 12] It is an explanatory diagram showing the extraction of text within a table and the table structure by a third language model. [Figure 13] It is an explanatory diagram showing the extraction of text within a flowchart and the relationship between each node by a third language model. [Figure 14] It is an explanatory diagram showing an example of a split file. [Figure 15] It is a flowchart showing the processing procedure of the description content registration process. [Figure 16] It is an explanatory diagram showing an example of a first language model. [Figure 17] It is an explanatory diagram showing an example of a search module. [Figure 18] It is an explanatory diagram showing an example of a fourth language model. [Figure 19] It is an explanatory diagram showing an example of a question and answer screen. [Figure 20] It is an explanatory diagram showing an example of a split file display screen. [Figure 21] It is an explanatory diagram showing an example of a manual file display screen. [Figure 22] It is a flowchart showing the processing procedure of the question and answer process. [Figure 23] It is an explanatory diagram showing an example of a question and answer screen according to Embodiment 2. [Figure 24] It is an explanatory diagram showing an example of a fourth language model according to Embodiment 3. [Figure 25]This is an explanatory diagram showing an example of a question answering screen according to Embodiment 3. [Modes for carrying out the invention]

[0019] (Embodiment 1) Figure 1 is a schematic diagram showing an example configuration of a document search system S according to Embodiment 1. The document search system S includes an information processing device 1, a terminal device 2, and a language model server 3. The information processing device 1 obtains a query (search query) for searching for split files obtained by dividing a manual file (document file) of a manual or instruction manual related to equipment by inputting a question obtained from the terminal device 2 into the language model of the language model server 3. The information processing device 1 also outputs file information related to the split files obtained by dividing a manual file (document file) related to equipment by inputting a prompt into the language model of the language model server 3, highlighting the parts related to the obtained search query, and outputting it to the terminal device 2. The terminal device 2 transmits the question entered by the user to the information processing device 1 and displays the file information related to the split file output from the information processing device 1. The information processing device 1 communicates with the terminal device 2 or the language model server 3 via the network N.

[0020] Figure 2 is a block diagram showing an example configuration of an information processing device 1 and a language model server 3. The information processing device 1 is, for example, a server computer and comprises a processing unit (arithmetic unit) 11, a storage unit 12, and a communication unit 13. The processing unit 11 is composed of a CPU (Central Processing Unit), MPU (Micro Processing Unit), GPU (Graphical Processing Unit), or quantum processor, and performs various control processing, arithmetic processing, etc., by reading and executing a program P (program product) and a database pre-stored in the storage unit 12. A database server or the like may be provided outside the information processing device 1, and the database may be read from the database server or the like. Furthermore, the information processing device 1 may have its functions realized by multiple server devices or computers. Also, the information processing device 1 may correspond to a node on a blockchain.

[0021] The storage unit 12 of the information processing device 1 is, for example, a volatile memory and a non-volatile memory. The storage unit 12 stores a program P, a synonym DB (database) 121, an ideal question DB 122, and a split file DB 123. The program P may be provided to the information processing device 1 using a storage medium 12a that is readable by a computer. The storage medium 12a is, for example, a portable memory. Examples of portable memory include CD-ROMs, USB (Universal Serial Bus) memory, SD cards, microSD cards, or CompactFlash® memory. If the storage medium 12a is a portable memory, the processing element of the processing unit 11 may read the program P from the storage medium 12a using a reading device (not shown). The read program P is written to the storage unit 12. Furthermore, the program P may be provided to the information processing device 1 by the communication unit 13 communicating with an external device. Details of the synonym DB 121, ideal question DB 122, and split file DB 123 will be described later. Furthermore, some or all of the synonym DB121, ideal question DB122, and split file DB123 may be stored on an external server different from the information processing device 1. Also, the synonym DB121, ideal question DB122, and split file DB123 do not need to be stored on the same device or server; they may be stored on different devices or servers and read by the information processing device 1.

[0022] The communication unit 13 is a communication module or communication interface for communicating with the terminal device 2 by wire or wireless connection, and is, for example, a wide-area wireless communication module such as LTE®, 4G, or 5G. The processing unit 11 communicates with the terminal device 2 and the language model server 3 via the communication unit 13, for example, through an external network N such as the Internet.

[0023] Language model server 3 stores the first language model M1, the second language model M2, the third language model M3, and the fourth language model M4. Details of the first language model M1, the second language model M2, the third language model M3, and the fourth language model M4 will be described later. Note that the first language model M1, the second language model M2, the third language model M3, and the fourth language model M4 may each be stored in different language model servers. Also, some or all of the first language model M1, the second language model M2, the third language model M3, or the fourth language model M4 may be implemented using the same language model. Furthermore, the first language model M1, the second language model M2, the third language model M3, and the fourth language model M4 may be stored in the information processing device 1 or the terminal device 2.

[0024] Figure 3 is a block diagram showing an example configuration of terminal device 2. Terminal device 2 is, for example, a smartphone, a tablet terminal, or a personal computer. In this embodiment, the example of terminal device 2 being a smartphone will be described. Terminal device 2 comprises a terminal processing unit 21, a storage unit 22, a communication unit 23, an input unit 24, and a display unit 25. The terminal processing unit 21 is composed of a CPU or MPU, and performs various control processing, arithmetic processing, etc. Terminal device 2 may perform all or part of the processing performed by the information processing device 1 or the language model server 3.

[0025] The memory unit 22 stores an application program Pa for an application that receives input from the user regarding equipment and presents the user with recommended air conditioning equipment. The application program Pa is provided to the terminal device 2, for example, using a storage medium 22a. The terminal processing unit 21 of the terminal device 2 may also obtain the application program Pa using the internet and store it in the memory unit 22.

[0026] The communication unit 23 is a communication module or communication interface for communicating with the information processing device 1 wirelessly. The terminal processing unit 21 communicates with the information processing device 1 via the communication unit 23 and through an external network N.

[0027] The input unit 24 receives input from the user regarding questions about the equipment. In this embodiment, the terminal device 2 is a smartphone, and the input unit 24 and the display unit 25 are integrated by a touch panel. The input unit 24 receives input from the user regarding questions about equipment, including home appliances such as controllers, remote controls, air conditioners, refrigerators, washing machines, televisions, or vacuum cleaners, as well as bathroom equipment, kitchen equipment, or other residential equipment, for example, by touch input on the touch panel. The input unit 24 may also be configured with a microphone to receive input from the user regarding questions about the equipment by voice input. In the following, the case where the equipment is an air conditioner will be explained as an example.

[0028] The display unit 25 displays an input reception screen for receiving questions about the equipment, and a file screen for displaying the split files received from the information processing device 1.

[0029] Figure 4 is an explanatory diagram showing an example of the second language model M2. The second language model M2 is composed of an LLM (Large Language Model) such as GPT (Generative Pretrained Transformer: registered trademark). The second language model M2 is input with a synonym extraction prompt that includes commands, output format constraints, a first document, and a second document. The commands included in the synonym extraction prompt input to the second language model M2 include a command to create a synonym dictionary that associates formal terms with commonly used abbreviations and nicknames (hereinafter also referred to as slang) for technical terms related to air conditioners (Please create a synonym dictionary that associates formal terms with commonly used abbreviations and nicknames for technical terms related to air conditioners), a command to extract formal terms from the first document (Please extract formal terms from the first document), and a command to extract slang from the second document (Please extract commonly used abbreviations and nicknames from the second document).

[0030] The output format constraint indicates a constraint that pairs of formal terms and colloquial terms should be output. The first document is a formal document written in formal language, such as an air conditioner manual or instruction manual. The second language model M2 receives the content of the formal first document. The second document is a colloquial document written in colloquial language, such as questions about air conditioners, records of answers to those questions, minutes of meetings about air conditioners, or service news created by field workers. The second language model M2 receives the content of the colloquial second document. The second language model M2 may also receive URL links to data from the first or second document.

[0031] When the synonym extraction prompt described above is input to the second language model M2, it outputs pairs of words (synonyms) that have the same meaning or refer to the same thing from among the formal terms listed in the first document and the abbreviations listed in the second document. Examples of pairs of formal terms and colloquial terms include pairs such as 7-segment and 7-segment, filling and charging, dew-forming and dew-forming, indoor unit and indoor machine, and refrigerant system diagram and piping system diagram. The processing unit 11 of the information processing device 1 stores the pairs of formal terms and colloquial terms output by the second language model M2 in the synonym DB 121 of the storage unit 12.

[0032] Figure 5 is an explanatory diagram showing an example of the synonym database 121. The synonym database 121 (synonym collection) stores pairs of formal terms and slang terms output by the second language model M2. The management items of the synonym database 121 include a slang field and a formal term field. The slang field stores slang terms extracted by the second language model M2 from the second document. The formal term field stores formal terms related to air conditioners extracted by the second language model M2 from the first document. Note that the synonym database 121 may have multiple slang fields, and multiple slang terms may be stored for a single formal term.

[0033] Figure 6 is an explanatory diagram showing an example of the Ideal Question DB 122. The Ideal Question DB 122 (question set) stores the word transformation rules used when converting user questions into search queries by the first language model, which will be described later. User questions are transformed based on the target words and keyword sets stored in the Ideal Question DB 122, thereby converting them into search queries that enable more accurate searching of segmented files by the search module 125. In other words, the Ideal Question DB 122 stores transformation rules for adding or correcting information to user questions and transforming them into ideal questions to be input into the search module 125. The management items of the Ideal Question DB 122 include a target word field and a transformation field. The target word field stores the words that will be transformed when the user question is converted into a search query. The words to be transformed include words with insufficient information, words that do not conform to the writing style of formal documents, or words that are colloquial expressions. The transformation field stores a keyword set containing one or more keywords after the transformation of the target words included in the question. The converted keyword set includes multiple keywords with supplemented information, keywords that conform to the formatting of formal documents, or keywords with literary expressions. The information stored in the Ideal Question DB122 is, for example, information that has been registered in advance by manual input. The Ideal Question DB122 may also store pairs of target words output by inputting the first and second documents into the second language model M2 and the converted keyword set. The information stored in the Ideal Question DB122, along with the information stored in the Synonym DB121, is input into the first language model M1, which will be described later.

[0034] The Ideal Question DB122 may store examples of questions. The processing unit 11 of the information processing device 1 may store questions previously entered and evaluations of the answers output to those questions (for example, an evaluation by a user of whether the answer matches what is required as an answer to the question), and store questions with high evaluation values ​​in the Ideal Question DB122. In addition, questions entered by specific users (for example, users with long work experience) or ideal questions created in advance may be stored in the Ideal Question DB122. Furthermore, questions previously entered regardless of the evaluation of the answer or the user who entered it may be stored in the Ideal Question DB122.

[0035] Figure 7 is a flowchart showing the processing procedure for synonym extraction. The processing unit 11 of the information processing device 1 acquires the instructions and output format constraints to be included in the synonym extraction prompt (S1). The processing unit 11 may acquire these instructions and output format constraints by reading them from the storage unit 12 of the information processing device 1 in advance, or it may acquire the instructions and output format constraints input by the terminal device 2, for example. The processing unit 11 acquires the first document and the second document (S2). The processing unit 11 acquires the first document and the second document sent from the terminal device 2, for example. The processing unit 11 may also acquire the first document and the second document read from an external database server or storage unit 12.

[0036] The processing unit 11 of the information processing device 1 inputs a synonym extraction prompt, including the acquired instruction, output format constraints, first document, and second document, to the second language model M2 (S3). The processing unit 11 acquires the pairs of formal and colloquial synonyms output by the second language model M2 (S4). The processing unit 11 stores the pairs of formal and colloquial synonyms in the synonym DB 121 of the storage unit 12 (S5), and terminates the process.

[0037] Figure 8 is an explanatory diagram showing an example of the third language model M3. The third language model M3 is composed of an LLM such as GPT. When a data extraction prompt and a manual file related to an air conditioner are input to the third language model M3, it performs various processes according to the instructions contained in the data extraction prompt. The processes performed by the third language model M3 include extracting metadata items that will become specified tags (such as model name, part name, or error code) from the manual file, extracting metadata that distinguishes a figure from a figure contained in the manual file, extracting metadata, text within a table, and table structure from a table contained in the manual file, extracting metadata, text within a flowchart, and simultaneous node relationships from a flowchart contained in the manual file, converting images or figures contained in the manual file to text, or dividing the manual file into units such as chapters, sections, or subsections and generating divided files.

[0038] The data extraction prompt input to the third language model M3 in this embodiment includes multiple commands. The commands included in the data extraction prompt include commands to extract items (metadata) of the air conditioner model name, part name, and error code that will serve as tags (extract the items of the air conditioner model name, part name, and error code from the text, figures, tables, or flowcharts in the manual file), commands to extract text and table structure within a table (if a table is included, extract the text and table structure within the table and output the text within the table faithfully to the table structure), commands to extract text and the relationships between each node within a flowchart (if a flowchart is included, extract the text and the relationships between each node within the flowchart and output the text within the flowchart faithfully to the relationships between the nodes), and commands to split the manual file into units and output them (split the manual file into units such as sections, chapters, or subsections and output the split files).

[0039] The third language model M3, upon receiving a data extraction prompt and a manual file (input data), outputs multiple split files (document files), metadata extracted from each split file, and the text contained in each split file. Each split file is assigned a manual ID from the manual file on which the split file originated, and a split ID (identifier) ​​that distinguishes each split file. The processing unit 11 of the information processing device 1 stores each split file, manual ID, split ID, metadata, and text in the split file DB123.

[0040] Figure 9 is an explanatory diagram illustrating the extraction of metadata from a figure or image by the third language model M3. When a figure or image representing, for example, an air conditioner part is input to the third language model M3, the third language model M3 extracts the words that indicate the name of the part in the figure or image as the part name and outputs it as metadata. In the example shown in Figure 9, the item for the part is extracted, and "thermistor" and "electronic expansion valve" are output as metadata. Note that the third language model M3 may also be a conversion model composed of a VLM (Vision and Language Model) that recognizes the figure or image when it is input and converts what the figure or image represents into text.

[0041] Figure 10 is an explanatory diagram illustrating the extraction of table metadata by the third language model M3. When a diagram showing the specifications of an air conditioner is input to the third language model M3, for example, the third language model M3 extracts the strings that indicate the model name in the table as the model name and outputs them as metadata. In the example shown in Figure 10, the strings stored in each field of the record indicating the "system name" in the table are extracted, and metadata such as "REYP224F" and "REYP280F" are output.

[0042] Figure 11 is an explanatory diagram illustrating the extraction of flowchart metadata by the third language model M3. When a flowchart showing, for example, a method for diagnosing and correcting an air conditioner malfunction is input to the third language model M3, the third language model M3 extracts the corresponding error code when using the flowchart and outputs it as metadata. In the example shown in Figure 11, the error code written in the upper left corner of the page containing the flowchart is extracted, and "A0" is output as metadata.

[0043] Figure 12 is an explanatory diagram illustrating the extraction of text and table structure within a table by the third language model M3. For example, if a table listing trial operation modes (settings) during inspection is input to the third language model M3, the third language model M3 extracts the text and table structure contained in each field of the table, integrates the extracted texts to reflect the table structure, and outputs a text explaining the contents of the table. In the example shown in Figure 12, the text contained in each field and the structure in which the contents of each trial operation mode are stored in the same record field are extracted, and a text explaining the contents of each trial operation mode is output. Note that the data extraction prompt input to the third language model M3 may include commands to output the table contents in structured text such as JSON notation, array notation, or Markdown notation. In this case, the third language model M3 outputs the input table contents in structured text.

[0044] Figure 13 is an explanatory diagram illustrating the extraction of text and the relationships between nodes in a flowchart by the third language model M3. When a flowchart showing diagnostic or corrective procedures for an error occurring in an air conditioner is input to the third language model M3, the model extracts the text and relationships between nodes contained within each node of the flowchart, integrates the extracted texts reflecting the relationships between nodes, and outputs text explaining the contents of the flowchart. In the example shown in Figure 13, the text indicating the diagnostic branch, confirmation items, and corrective actions when the error code "A0" is displayed on the remote control, as well as the arrow lines between nodes and the relationships between nodes indicated by the diagnostic branch conditions (YES or NO), are extracted, and text explaining the contents of the flowchart is output. Note that the data extraction prompt input to the third language model M3 may include commands to output the contents of the flowchart in structured text such as JSON notation, array notation, or Markdown notation. In this case, the third language model M3 outputs the contents of the input flowchart in structured text.

[0045] Figure 14 is an explanatory diagram showing an example of a split file DB123. The split file DB123 stores information related to the split files output by the third language model M3. The management items (fields) of the split file DB123 include the manual ID field, split ID field, split file name field, item field, metadata field, description type field, and description content field.

[0046] The manual ID field in the split file DB123 stores the manual ID that identifies the document (manual) file input to the third language model M3 when outputting the split file, i.e., the manual file that served as the basis for the split file. The split ID field stores the split ID that identifies the split file. The split file name field stores the name that indicates the split file. The split file name includes, for example, the document name of the source document of the split file, and the titles of units such as chapters, sections, or articles contained in the split file. The item field stores the metadata items (either model name, part name, or error code) extracted from the split file. Note that multiple items may be extracted for a single split file and stored in the item field. The metadata field stores the metadata (tags) extracted from the split file. The description type field stores the type of description method (figure, table, flowchart, or other) described in the split file. Note that "other" in the description type refers to, for example, plain text. If plain text is described in the split file, the third language model M3 extracts and outputs the plain text. The content field stores the content of the split files, which is extracted and output by the third language model M3 from the split files. The storage unit 12 of the information processing device 1 may store the split files in a file format such as PDF. In this case, the split file DB123 may store links for accessing the split files.

[0047] Figure 15 is a flowchart showing the processing steps for registering the contents. The processing unit 11 of the information processing device 1 obtains the command to be included in the data extraction prompt (S11). The processing unit 11 may obtain the command by reading it from the storage unit 12 of the information processing device 1 in advance, or it may obtain the command that was entered in the terminal device 2, for example. The processing unit 11 obtains the manual file (S12). The processing unit 11 obtains the manual file sent from the terminal device 2, for example. The processing unit 11 may also obtain the manual file read from an external database server or the storage unit 12.

[0048] The processing unit 11 of the information processing device 1 inputs the acquired command and a data extraction prompt including the manual file to the third language model M3 (S13). The processing unit 11 acquires the split files, metadata, and contents of the split files output by the third language model M3 (S14). The processing unit 11 stores the manual ID of the manual file input to the third language model M3, the split ID assigned to each split file, the metadata items extracted from the split files, the metadata extracted from the split files, and the contents of the split files into the split file DB123 (S15), and then terminates the process.

[0049] Figure 16 is an explanatory diagram showing an example of the first language model M1. The first language model M1 is composed of an LLM such as GPT. When a user's question about an air conditioner is input to the first language model M1, it refers to the synonym DB 121 and the ideal question DB 122 and outputs a search query to be input to the search module 125, which will be described later.

[0050] The question conversion prompt input to the first language model M1 includes an instruction, a synonym DB121, an ideal question DB122, and a question. The instruction in the question conversion prompt input to the first language model M1 includes an instruction to convert the question into a search query (Convert the user's question into a search query suitable for input to the search module.), an instruction to convert slang into formal terms by referring to the synonym DB121 (Refer to the synonym DB and convert slang into formal terms that are synonyms.), and an instruction to convert the question phrases into keywords or keyword sets by referring to the ideal question DB122 (Refer to the ideal question DB and convert the question phrases to resolve any information discrepancies.).

[0051] In the example shown in Figure 16, the question "Model: REYP224F Question: Coolant Recovery" is input to the first language model M1. At this time, the first language model M1 converts "coolant" in the question to "refrigerant" based on the synonym DB121 and outputs "Model: REYP224F Question: Refrigerant Recovery" as a search query. Note that the first language model M1 may output multiple different search queries.

[0052] Figure 17 is an explanatory diagram showing an example of the search module 125. The search module 125 receives a search query output by the first language model M1 as input. The search module 125 identifies multiple split files that contain content matching the input search query, for example, by calculating the similarity between the search query and the content stored in the content fields of the split file DB123. The search module 125 outputs the manual ID and split ID of the identified split files in descending order of similarity to the search query. The search module 125 may also identify multiple split files that contain content matching the input search query using methods such as pattern matching. Furthermore, the process of searching for split files from the split file DB123 based on the search query may be performed using the fourth language model M4.

[0053] Figure 18 is an explanatory diagram showing an example of the fourth language model M4. The fourth language model M4 is composed of an LLM such as GPT. When a search query related to air conditioners and a split file related to the manual ID and split ID output by the search module 125 are input to the fourth language model M4, it extracts the split files that are highly relevant to the search query and outputs a summary of the contents of the extracted split files. In addition to the search query, or in addition to it, a user's question may also be input to the fourth language model M4.

[0054] The summary prompt input to the fourth language model M4 includes commands to select (identify) up to a predetermined number (e.g., 5) of split files that are particularly relevant to the search query from the split files output by the first language model M1 (Please select up to 5 split files that you have determined to be highly relevant to the search query.), commands to output the manual ID and split ID (identifier) ​​of the selected split files (Please output the manual ID and split ID of the selected split files.), commands to output a summary of the contents of the selected split files (Please output a summary of the contents of the selected split files. The summary should use words contained in the contents of the split files and be 50 characters or less.), commands to output the reason why no split files highly relevant to the search query can be selected if none can be selected (Please output the reason if no split files are determined to be highly relevant to the search query.), and commands to extract and highlight the parts (evidence phrases) from the contents of the split files that were determined to be highly relevant to the search query (Please extract the words from the contents of the split files that were determined to be highly relevant to the search query.). The summary prompt also includes output examples that specify the output format.

[0055] When the fourth language model M4 receives a summary prompt containing an instruction, a search query, and multiple split files (input data, file group) output by the search module 125, the fourth language model M4 outputs the manual ID and split ID of the split files that are particularly relevant to the search query, a summary of the contents of the split files, and supporting keywords. If the fourth language model M4 cannot select a split file that is particularly relevant to the search query, it outputs the reason why it cannot be selected. The reason for not being able to select a file may include, for example, that the split file does not contain a section that answers the question, or that the question is unclear.

[0056] Figure 19 is an explanatory diagram showing an example of a question-answering screen. The terminal processing unit 21 of the terminal device 2 displays the question entered by the user in the input unit 24 on the display unit 25 and simultaneously transmits it to the information processing device 1. Subsequently, the terminal processing unit 21 displays the manual ID, document name, split ID, split file name (link), summary, and the words that formed the basis of the judgment for each of the multiple split files obtained from the information processing device 1. The split file name is displayed as a link (displayed by an underline) for displaying (accessing) the split file. If a split file name link is selected, the terminal processing unit 21 displays the split file display screen (see Figure 20) on the display unit 25. The split file name link and summary displayed on the question-answering screen constitute the file information. For example, on the question-answering screen, the terminal processing unit 21 sequentially displays the entered question and the information obtained from the information processing device 1 in a chat format.

[0057] Figure 20 is an explanatory diagram showing an example of a split file display screen. The terminal processing unit 21 of terminal device 2 displays the split file related to the split file name (link) selected in the question answering screen on the split file display screen. In the split file displayed on the split file display screen, the supporting terms are highlighted to emphasize them. In addition, in the split file display screen, groups of words or sentences containing supporting terms may be highlighted. Furthermore, the method of emphasizing supporting terms is not limited to highlighting; the color used to display the supporting terms may be different from other words, or the supporting terms may be underlined. If the supporting terms are in an image, figure, table, or flowchart of the split file, the supporting terms (parts related to the search query) in the image, figure, table, or flowchart may be highlighted to emphasize them.

[0058] Furthermore, the split file display screen also displays a link to access the manual file associated with the manual ID of the split file (the full file is here). If the link to the manual file is selected, the terminal processing unit 21 of terminal device 2 displays the manual file display screen (see Figure 21).

[0059] Figure 21 is an explanatory diagram showing an example of a manual file display screen. The manual file display screen displays the manual file corresponding to the manual ID of the split file displayed on the split file display screen, that is, the manual file that formed the basis of the split file. In addition, the relevant terms may be highlighted in the manual file displayed on the manual file display screen. Since the manual file consists of multiple pages, each page is displayed in a small size on the manual file display screen, and a page selection field that accepts the selection of the page to be viewed and a page display field that displays the selected page are displayed.

[0060] Figure 22 is a flowchart showing the processing procedure for question answering. The terminal processing unit 21 of the terminal device 2 displays the question answering screen on the display unit 25 (S21). The terminal processing unit 21 receives a question from the user on the input unit 24 (S22). The terminal processing unit 21 transmits the input question to the information processing device 1 (S23).

[0061] The processing unit 11 of the information processing device 1 acquires a question about the air conditioner entered in the terminal device 2 (S24). The processing unit 11 inputs the acquired question, the question conversion prompt, and the synonym DB 121 and ideal question DB 122 stored in the storage unit 12 to the first language model M1 of the language model server 3 (S25) and acquires a search query (S26). The processing unit 11 inputs the search query to the search module 125 (S27) and acquires the manual ID and split ID of multiple split files (S28).

[0062] The processing unit 11 of the information processing device 1 inputs the acquired manual ID and segmentation ID, multiple segmented files (file groups), search queries, and summary prompts into the fourth language model. (S29) The manual ID and split ID of the highly relevant split files, a summary of the split files, and the supporting keywords that formed the basis for determining that the split files are highly relevant to the search query are output (S30). The processing unit 11 sends the manual file related to the output manual ID, the split files related to the document ID and split ID, the summary of the split files, and the words that formed the basis for the determination to the terminal device 2 (S31), and terminates the process.

[0063] The terminal processing unit 21 of terminal device 2 obtains the manual file related to the output manual ID, the split files related to the document ID and split ID, a summary of the split files, and the words that formed the basis of the decision from the information processing device 1 (S32). The terminal processing unit 21 displays the document ID, split ID, link to the split file (split file name), summary, and basis words on the question answer screen (S33). The terminal processing unit 21 accepts the selection of the split file link (S34) and displays a split file display screen on the display unit 25, highlighting the basis words (S35). The terminal processing unit 21 accepts the selection of the manual file link (S36), displays the manual file display screen on the display unit 25 (S37), and terminates the process.

[0064] According to the configuration and processing of Embodiment 1, the processing unit 11 of the information processing device 1 can search for the split files containing the answers to the questions with high accuracy by using a search query converted from the acquired question by referring to the synonym DB 121 and the ideal question DB 122. Furthermore, by presenting the user with split files obtained by dividing the manual file into multiple parts and highlighting the relevant keywords in the split files, it is possible to present the user with the parts related to the search query in an easy-to-understand manner.

[0065] (Embodiment 2) Figure 23 is an explanatory diagram showing an example of a question answering screen according to Embodiment 2. In the question answering screen according to Embodiment 2, a model name input field, an error code input field, and a question content input field are provided. The model name is entered in the model name input field. When characters are entered in the model name input field, a suggestion of model names containing the entered characters is displayed. The model name suggestions are displayed based on the model names of the metadata stored in the segmented file DB123 stored in the storage unit 12 of the information processing device 1. The terminal processing unit 21 receives the selection of a model name to be entered in the model name input field from the suggestions and transmits the selected model name to the information processing device 1.

[0066] The error code input field is used to input error codes, for example, those displayed on the air conditioner's remote control or a computer connected to the air conditioner. If characters are entered in the error code input field, a suggestion of error codes containing the entered characters will be displayed. The error code suggestions are displayed based on the error codes in the metadata stored in the split file DB123, which is stored in the storage unit 12 of the information processing device 1. The terminal processing unit 21 accepts the selection of an error code to be entered in the error code input field from the suggestions and sends the selected error code to the information processing device 1 as the error code.

[0067] The user enters their question into the question input field. If a word is entered into the question input field, a set of keyword suggestions containing the entered word or a formal term that is a synonym of the entered word will be displayed. The question suggestions are displayed based on the synonym DB121 and ideal question DB122 stored in the storage unit 12 of the information processing device 1. Specifically, the terminal processing unit 21 refers to the synonym DB121 and converts the entered word to a formal term if it is slang. The terminal processing unit 21 also refers to the ideal question DB122 and displays multiple keyword sets containing the entered word or the converted formal term as suggestions. The terminal processing unit 21 accepts the selection of the question content to be entered into the question input field from the suggestions and sends the selected question content to the information processing device 1.

[0068] The terminal processing unit 21 of terminal device 2 transmits the model name, error code, and question content entered on the question answering screen to the information processing device 1. The processing unit 11 of the information processing device 1 inputs the acquired model name, error code, and question content as a search query into the search module 125, identifies multiple split files containing matching content as answers to the search query based on the content fields of the split file DB 123, and outputs the manual ID and split ID of the identified split files. Subsequent processing is the same as in Embodiment 1.

[0069] According to the configuration and processing of Embodiment 2, it becomes possible to guide the user to input an appropriate model name or question content through suggestions, thereby improving the accuracy of the search for split files by the search module 125.

[0070] (Embodiment 3) Figure 24 is an explanatory diagram showing an example of the fourth language model M4 according to Embodiment 3. The summary prompt input to the fourth language model M4 according to Embodiment 3 includes a command to output thumbnails of images, figures, tables, or flowcharts contained in the selected split file and related to the search query (if the split file contains images, figures, tables, or flowcharts, please output these thumbnails).

[0071] The fourth language model M4 according to Embodiment 3, when the above-described prompt is entered, extracts the image, figure, table, or flowchart portion from the split file and outputs it as a thumbnail. In the example shown in Figure 24, the figure included in the split file is output as a thumbnail.

[0072] Figure 25 is an explanatory diagram showing an example of a question-answering screen according to Embodiment 3. In the question-answering screen according to Embodiment 3, thumbnails of images, figures, tables, or flowcharts contained in each split file are displayed. If a split file does not contain any images, figures, tables, or flowcharts, the thumbnail for that split file is not displayed. When a thumbnail is selected, a split file display screen (see Figure 20) is displayed, showing the split file containing the image, figure, table, or flowchart corresponding to the thumbnail.

[0073] According to the configuration and processing of Embodiment 3, a thumbnail is displayed along with a summary of the split file on the question answering screen, making it easier for the user to select the split file they want to view.

[0074] The embodiments disclosed herein should be considered in all respects as illustrative and not restrictive. The technical features described in each embodiment can be combined with each other, and the scope of the present invention is intended to include all modifications within the claims and scope equivalent to the claims. Furthermore, the independent and dependent claims described in the claims can be combined with each other in any combination, regardless of the form of reference. In addition, the claims use a multi-claim format in which claims refer to two or more other claims (multi-claim format), but are not limited to this. They may also be described using a multi-claim format in which at least one multi-claim refers to another multi-claim (multi-multi-claim format). [Explanation of Symbols]

[0075] 1: Information Processing Device 2: Terminal device 3: Language Model Server 11: Processing Section 12: Storage section 12a:Storage medium 13: Communications Department 21: Terminal Processing Unit 22: Storage section 22a:Storage medium 23: Communications Department 24: Input section 25:Display section 125: Search Module 121:Synonym DB 122: Ideal question DB 123: Split File DB M1: First Language Model M2: Second Language Model M3: Third Language Model M4: Fourth Language Model N: Network P: Program Pa: App Program S: Document search system

Claims

1. Retrieve queries related to equipment and machinery. Using a language model that utilizes the acquired query and a group of files containing multiple split files obtained by dividing the document file related to the equipment, file information related to the split file corresponding to the query is output. The system outputs a split file corresponding to the query, highlighting the parts related to the query. Information processing methods.

2. The program outputs split files, along with a link to access the document file containing those split files. The information processing method according to claim 1.

3. The file information includes a link to access the segmented file corresponding to the query, and a summary of the segmented file summarized by the language model. The information processing method according to claim 1 or 2.

4. The language model is input a prompt that includes the retrieved query, a command to identify the split files corresponding to the query, a command to output an identifier for each split file, a command to highlight the portion of the identified split file related to the query, and a command to generate a summary of the identified split files. The information processing method according to claim 3.

5. The aforementioned file information further includes thumbnails of figures, tables, or flowcharts related to the retrieved query, which are contained in the split files corresponding to the query. The information processing method according to claim 3.

6. When a thumbnail selection is accepted, a split file containing a figure, table, or flowchart corresponding to the accepted thumbnail is output, with the portion of the figure, table, or flowchart related to the query highlighted. The information processing method according to claim 5.

7. The similarity between the retrieved query and the segmented files included in the aforementioned file group is calculated. Output multiple file information files related to multiple split files, sorted by similarity. The information processing method according to claim 1 or 2.

8. A query containing the model name of the aforementioned equipment is obtained. The retrieved query is input into the language model. The information processing method according to claim 1 or 2.

9. A query is obtained that includes the model name of the equipment and the error code output from the equipment. The retrieved query is input into the language model. The information processing method according to claim 1 or 2.

10. Equipped with a calculation unit, The aforementioned arithmetic unit, Retrieve queries related to equipment and machinery. Using a language model that utilizes the acquired query and a group of files containing multiple split files obtained by dividing the document file related to the equipment, file information related to the split file corresponding to the query is output. The system outputs a split file corresponding to the query, highlighting the parts related to the query. Information processing device.

11. Retrieve queries related to equipment and machinery. Using a language model that utilizes the acquired query and a group of files containing multiple split files obtained by dividing the document file related to the equipment, file information related to the split file corresponding to the query is output. The system outputs a split file corresponding to the query, highlighting the parts related to the query. A computer program that causes a computer to perform a process.

Images