Program, information processing method, and information processing device

Abstract

To support more efficient acquisition of useful information.SOLUTION: A program causes a computer to: acquire a prompt related to a product or a component of the product; acquire search results that are obtained by searching a database storing the product or the component browsable by an authorized object person using a query based on the prompt, and that include supplementary information about subjects included in the query; and generate a modified prompt for input into a trained model, which has been trained to output information corresponding to the input information, and including all or a part of the supplementary information on the basis of the prompt and the search results.SELECTED DRAWING: Figure 1

Description

[Technical field]

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

[0002] Conventionally, a data generation method has been disclosed in which a prompt is input to a language model and information is obtained from the language model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2023-18624 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the above techniques, prompts for obtaining useful information more efficiently have not been sufficiently considered.

[0005] The present invention has been made in consideration of the above circumstances, and one of its objects is to provide a program, an information processing method, and an information processing device that can assist in obtaining useful information more efficiently. [Means for solving the problem]

[0006] An embodiment of the present invention is a program that causes a computer to obtain a prompt related to a product or a part of the product, use a query based on the prompt to search a database that stores relevant information related to the product or part that can be viewed by authorized subjects to obtain search results including supplemental information related to a subject included in the query, and generate a modification prompt including some or all of the supplemental information based on the prompt and the search results, the modification prompt being input into a trained model that has been trained to output information corresponding to the input information. Effect of the Invention

[0007] In this embodiment, it is possible to assist in obtaining useful information more efficiently. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 illustrates an example of a configuration of an information processing system 1 including a program. [Diagram 2] FIG. 2 illustrates an example of a functional configuration of a management device 100. [Diagram 3] FIG. 2 illustrates an example of a functional configuration of a search device 200. [Figure 4] FIG. 2 is a conceptual diagram of information stored in a search DB 280. [Diagram 5] FIG. 13 is a diagram showing an example of an interface screen IF1 provided by a search DB 280. [Figure 6] FIG. 13 is a diagram illustrating an example of a screen relating to properties. [Figure 7] FIG. 13 is a diagram showing an example of an issue management screen IF4. [Figure 8] FIG. 11 is a diagram showing an example of identified content. [Figure 9] FIG. 13 is a diagram illustrating an example of information included in a correction prompt. [Figure 10] The interface screen IF6 is a diagram showing an example of processed information. [Figure 11]13 is a flowchart showing an example of the flow of processing executed by the management device 100. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, embodiments of a program, an information processing method, and an information processing device of the present invention will be described with reference to the drawings.

[0010] [Overall configuration] 1 is a diagram showing an example of a configuration of an information processing system 1 including a program. The information processing system 1 includes, for example, a user terminal device 10, a language model 50, a management device 100, and a search device 200. The management device 100 communicates with the user terminal device 10, the language model 50, and the search device 200.

[0011] In this embodiment, the management device 100 is described as performing various processes in cooperation with the search device 200 and the language model 50, but a part or all of the functional configuration of the management device 100 may be provided in the user terminal device 10. The user terminal device 10 may perform various processes in cooperation with the search device 200 and the language model 50.

[0012] [Processing Overview] An outline of the process will be explained below, and details will be given later. (1) The user terminal device 10 transmits a prompt to the management device 100 based on a user operation. (2) The management device 100 generates a query from the prompt. (3) The management device 100 transmits the generated query to the search device 200 . (4) The search device 200 searches the search DB 280 (database) based on the transmitted query, and transmits the search results to the management device 100. (5) The management device 100 generates a revision prompt based on the prompt provided by the user and the results of the search. (6) The administration device 100 sends the correction prompt to the language model 50, requesting it to provide an answer to the prompt. (7) The language model 50 provides responses to the correction prompts to the management device 100 . (8) The management device 100 processes the provided answer and generates information to be provided to the user. (9) The management device 100 provides the provision information to the user terminal device 10 . As a result, a response to the prompt is displayed on the display unit of the user terminal device 10.

[0013] [User terminal device] The user terminal device 10 is, for example, a computer device such as a personal computer, a smartphone, or a tablet terminal. A user operates the user terminal device 10 to receive information provided from the search device 200 and to register information. The user operates the user terminal device 10 to obtain an answer to a prompt input from the language model 50 via the management device 100.

[0014] [Language model] The language model 50 is a model in which learning data is trained to output information that answers an input question or request. When a question is input, the language model 50 outputs an answer to the question. The language model 50 is a large language model (LLM) such as GPT-3, GPT-3.5, GPT-4, BERT, PaLM, or Llama.

[0015] [Management device] 2 is a diagram showing an example of a functional configuration of the management device 100. When a prompt is acquired, the management device 100 searches information accumulated in the search device 200 to acquire information highly related to the prompt, and realizes a framework called RAG (Retrieval-augmented Generation) that generates a processed prompt using the acquired information. The management device 100 includes, for example, an information processing unit 110, a query generating unit 120, a prompt generating unit 130, an information processing unit 140, and a storage unit 180. The information processing unit 110, the query generating unit 120, the prompt generating unit 130, and the information processing unit 140 are realized by, for example, a hardware processor such as a CPU (Central Processing Unit) executing a program (software) 182 stored in the storage unit 180. Some or all of these components may be realized by hardware (including circuitry) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or GPU (Graphics Processing Unit), or may be realized by cooperation of software and hardware. The program may be stored in advance in a storage device (a storage device having a non-transient storage medium) such as HDD (Hard Disk Drive) or flash memory, or may be stored in a removable storage medium (non-transient storage medium) such as DVD or CD-ROM, and may be installed by mounting the storage medium in a drive device. The storage unit 180 is realized by a storage device such as HDD, flash memory, or RAM (Random Access Memory). The storage unit 180 stores, for example, the above-mentioned program 182 and information provided by the user terminal device 10 (for example, information such as questions and operation contents input in the past).

[0016] The information processing unit 110 acquires a prompt related to a product or a part of the product. The information processing unit 110 performs natural language processing to analyze the prompt. The information processing unit 110 performs, for example, morphological analysis, syntactic analysis, semantic analysis, and context analysis using a corpus stored in a storage device to recognize the meaning of words and sentences included in the prompt. The information processing unit 110 refers to the corpus to identify words close to the meaning of a target word or words similar to the target word. Note that these processes may be performed using a learning model prepared in advance, or may be performed by requesting the processes from another device.

[0017] The query generation unit 120 generates a query using the result of processing by the information processing unit 110. For example, the query generation unit 120 identifies words of high importance from among the words included in the prompt using a predetermined rule or algorithm. The query generation unit 120 requests the search DB 280 to perform a search using the query.

[0018] The prompt generation unit 130 generates a correction prompt based on the result of the search of the search DB 280 and the prompt provided by the user. The information processing unit 140 acquires an answer to the correction prompt from the language model 50, processes the acquired answer, and provides the processed answer to the user terminal device 10. Details of the processing by the prompt generation unit 130 and the information processing unit 140 will be described later.

[0019] [Search device] 3 is a diagram showing an example of a functional configuration of the search device 200. The search device 200 includes, for example, an information management unit 210, an information search unit 220, and a search DB 280. The information management unit 210 and the information search unit 220 are realized by a hardware processor such as a CPU executing a program (software). Some or all of these components may be realized by hardware (including circuitry) such as an LSI, ASIC, FPGA, or GPU, or may be realized by a combination of software and hardware. The program may be stored in advance in a storage device such as an HDD or flash memory (a storage device having a non-transient storage medium), or may be stored in a removable storage medium such as a DVD or CD-ROM (a non-transient storage medium), and may be installed by mounting the storage medium in a drive device.

[0020] The information management unit 210 acquires information provided from the management device 100 or the user terminal device 10, and provides the results of processing by the search device 200 to the management device 100 or the user terminal device 10.

[0021] The information search unit 220 searches for information in the search DB 280 in response to a request from the management device 100 or the user terminal device 10. The information search unit 220 performs a search based on, for example, a query or keyword provided from the management device 100 or the user terminal device 10, or a specified condition.

[0022] The search DB280 is a database that stores various information. FIG. 4 is a conceptual diagram of information stored in the search DB280. In the search DB280, information is hierarchically organized, and each piece of information is stored in association with each other. For example, a first lower layer is associated with a higher layer, and a second lower layer is associated with the first lower layer. For example, the higher layer is a project, and the lower layer is an item related to the project. The project is a finished product, the items in the first lower layer are assemblies, and the items in the second lower layer are parts. In addition to the first lower layer and the second lower layer, a lower layer may be provided. The hierarchical structure of projects and items is sometimes called a "Bill Of Materials (BOM) structure."

[0023] Information other than projects and items is associated with a set of layers or each layer. The associated information is files and properties related to a set of layers or each layer. A file or property may be associated with one product or part, or with multiple products or parts. Files include drawings, cost calculations, test data, CAD data, specifications, requirement definition documents, and quality standards documents. Files are managed by version.

[0024] Properties are associated with projects, assemblies, and parts, for example, properties that are set by the user (see Figure 6 below).

[0025] In addition to the above, the search DB 280 manages various information such as change history information, data tables, specifications, product information on the product on which the target part will be installed, defect information, defect countermeasure information or guidelines, standards (design requirements, standards, in-house standards), and related product or part information. The search DB 280 manages information on products, parts, etc. as master information. Various specifications of products, parts, etc. are associated and managed in the master information.

[0026] [Interface screen provided by Search DB280] (BOM screen) Here, a description will be given of information that the search DB 280 provides to the user terminal device 10. The search DB 280 may provide information to the user terminal device 10 directly, or may provide information via the management device 100.

[0027] 5 is a diagram showing an example of an interface screen IF1 provided by the search DB 280. The interface screen IF1 is a screen that is displayed after a specific project is specified on a screen that displays a list of multiple items (assemblies, parts, etc.). The interface screen IF1 is a BOM screen that hierarchically displays parts that make up a specific item (e.g., a product) that is a specific project. When the above-mentioned master information is updated, it is reflected in the information on the BOM screen.

[0028] Interface screen IF1 includes information such as the model number of the specified item, category, phase related to the concept or manufacturing process (drawing release, prototyping, mass production, etc.), cost, creation date of the BOM screen, update date (for example, the last update date), creator, and person in charge (project owner). It also includes information on the parts that make up the specified item. Part information is displayed with each part as one recorder. For example, information such as the part's item name, model number, ID, phase, category, link to CAD data / drawing, and standard unit price is displayed as a record.

[0029] (Properties screen) When "Properties" is selected on the interface screen IF1, a properties screen (not shown) is displayed. The properties screen IF2 in FIG. 6 shows a part of the properties screen. The properties screen is a screen for managing the attributes of an item. The attributes are design, procurement, manufacturing, and other attributes. The design attributes are various information related to design, such as the person who inspected the drawing, the person who approved it, the material of the item, the production method, and weight. The procurement attributes are various information related to procurement, such as matters related to estimates, the supplier, the price, the estimated delivery date, the budget, the manufacturer product code (MPN), and the minimum order quantity (MOQ). The manufacturing attributes are various information related to manufacturing, such as the manufacturing record and the inventory management category. The user can assign tags, which are keywords used in searches, to each attribute item. The user can cause the search DB 280 to narrow down searches and information using property information and tags.

[0030] The information displayed on the above property screen can be customized by the user. For example, the user can customize the property setting screen by operating the property setting screen. The property setting screen IF3 in FIG. 6 is a diagram showing a part of the property setting screen. In the property setting screen IF3, the user can set the information to be displayed on the property screen IF2 by setting the property name, data type, unit, description, etc.

[0031] Furthermore, if the user changes the information, the search DB 280 stores the changes as a change history, allowing the user to perform cost analysis, search for past drawings, and the like by referring to the information before and after the change.

[0032] As described above, the user can set the properties. Generally, customizing properties requires development costs, but the search DB 280 provides an environment in which settings can be made without coding, reducing the burden on the user. Furthermore, it is possible to add not only the physical attributes of parts (weight and material), but also various information related to product development management, such as cost and inventory quantity, improving user convenience.

[0033] (Issue management screen) When the user performs a predetermined operation, an issue management screen is displayed. Fig. 7 is a diagram showing an example of the issue management screen IF4. The issue management screen IF4 is a screen on which, for example, issues and solutions to the issues are displayed as a set. Each issue is associated with an item (item, assembly, part). Each issue is associated with what item the issue is related to.

[0034] The issue management screen IF4 displays an overview of the issue, issues, measures, as well as items such as tasks, assignees, labels (complaints, improvement requests, requests, etc.), status, related projects, related items, files, etc. Items displayed on the issue management screen IF4 are associated with issues.

[0035] The issue management screen IF4 further displays a question (prompt) input by the user and an answer to the question. The answer is, for example, an answer generated by the management device 100 using an answer in the language model 50. After the answer is displayed, an input field for inputting a question is further displayed below it.

[0036] This allows the user to easily ask questions to solve the issue or questions related to the issue and check the answers.

[0037] In the above example, the questions and answers have been described as being displayed on the issue management screen IF4, but instead of (or in addition to) this, they may be displayed on other interface screens such as the interface screen IF1, the property screen IF2, the property setting screen IF3, etc. For example, the field for entering a question may be displayed on any screen.

[0038] [Providing answers to users] The management device 100 executes the program 182 to obtain a prompt related to a product or a part of a product, and obtains a search result including supplemental information related to a target included in the obtained query by searching a search DB 280 (database) in which relevant information related to the product or part that can be viewed by an authorized subject is stored, using a query based on the prompt. The management device 100 generates a correction prompt for input to a language model 50 (trained model), which includes a part or all of the supplemental information, based on the prompt and the search result. The management device 100 provides information based on output information obtained by inputting the correction prompt to the language model 50 to the user (user terminal device 10) who input the prompt. Details of this process will be described below.

[0039] (1) to (9) below correspond to (1) to (9) in Figure 1. (1) When the management device 100 acquires a prompt from the user terminal device 10, (2) it generates a query from the prompt. The query generation unit 120 generates a query from the prompt using a predetermined rule or algorithm. For example, the query generation unit 120 extracts one or more characteristic words from the prompt and determines the extracted words as the query. The characteristic words are words (words with similar meanings) that match or match a dictionary prepared in advance, or nouns. The generation of the query is not limited to the above method, and other methods may be used. The query generation unit 120 may generate queries of multiple or broad concepts so that information related to the user's question can be searched without omission. The information search unit 220 may include compatible products or substitutes for products or parts included in the prompt, related products (parent products, child products) or parts (parent parts, child parts), etc. in the query.

[0040] The above query may be generated as follows. The query generating unit 120 may, for example, specify a category of a preset keyword and include the specified category in the query. For example, a correspondence between a keyword and a category is prepared in advance. For example, it is assumed that a prompt "Tell me the points to be careful about when designing using this motor" is acquired. For example, the keyword "motor" is associated with an item classification. For example, the keyword "points to be careful about" is associated with quality regulations and know-how. The query generating unit 120 generates, for example, a query including the keywords "motor" and "points to be careful about" included in the prompt as described above, the item classification, quality regulations, know-how, or model number of the motor that are categories specified from the keywords, a query including a plurality of these words, or a query combining these. Note that, when a user is viewing an interface screen of a motor with a specific model number, the model number of the motor can be specified from the interface screen by viewing it. In this way, the query generating unit 120 can generate a query suitable for a search from information obtained based on information included in the prompt and content that the user is viewing.

[0041] In the above example, it has been described that a correspondence between keywords and categories is prepared, but the management device 100 may have a learned model and generate a query using the learned model. The learned model is a model that outputs a query corresponding to a prompt when a prompt is input. The learned model is a model generated by learning the learning data. The learning data is, for example, information in which data, which is one or more words or sentences, is associated with a category that is the correct answer data of the data. For example, the learning data is information in which a category or meaning in the design of a machine or the like corresponding to the word is associated with the word. For example, as described above, the learning data is information in which the meaning, category, etc. of an adjective or modifier is associated, such as a combination of a motor and an item classification, or a combination of a caution point and a quality regulation or know-how. By learning the learning data, the learned model outputs, for example, a query including an item classification when "motor" is input, and outputs, for example, a query including a quality regulation or know-how when "attention point" is input. In this way, the query generation unit 120 can easily generate a query by using the learned model.

[0042] (3) When a query is provided, the information search unit 220 of the search device 200 (4) searches the search DB 280 using the query and transmits the results to the management device 100. The information search unit 220 identifies content that includes a target word that matches or is similar to the query, and identifies content that includes the target word in the identified content, or content surrounding the content that includes the target word. The search may target all information in the search DB 280, or may target only a portion of the information. For example, documents showing problems and solutions in the search DB 280, quality standards, specifications, requirements definition documents, checklists, design standards, know-how lists, documents containing laws and regulations, documents related to safety standards, etc. may be searched. These are just examples, and other documents may be included, or some documents may be omitted. The identified content is provided to the management device 100. If the search results include information with the same name, or information that matches or is similar, the information search unit 220 may include the latest information with dates such as the update date or creation date of these data in the search results. For example, when there are specifications of old and new versions, the information in the specification of the new version is adopted with priority.

[0043] The information search unit 220 acquires information within a predetermined range of a part of a document that includes a part that matches or coincides with a query in the database of the search DB 280. FIG. 8 is a diagram showing an example of the specified content. For example, a target word is included in a predetermined page of a file such as a specification, a requirement definition document, or a quality standard document (for example, a file in any format such as a PDF or Word file). The information search unit 220 specifies all or a part of the pages before and after the target word in the page (P1, P3 in FIG. 8), the pages before and after the target word and the page (P1-P3 in FIG. 8), the sentences before and after the target word (the sentences in areas AR1, AR2 in FIG. 8), the sentences on the pages before and after the target word (areas AR3, AR4 in FIG. 8), and the like, as information to be provided to the management device 100. The information provided to the management device 100 is an example of "supplementary information."

[0044] The information search unit 220 may acquire, as supplemental information, information on a plurality of projects to which tag information (e.g., a property or a tag) indicating an attribute or classification corresponding to the query is assigned, from among a plurality of pieces of information (chunk information) corresponding to the respective projects. For example, information on the projects corresponding to the above-mentioned property or tag corresponding to the query is acquired as supplemental information, or a search result is acquired from information on the projects corresponding to the above-mentioned property or tag corresponding to the query.

[0045] The information in the search DB 280 may be vectorized based on a distributed representation technique using Word2Vec or the like. The information search unit 220 may generate a query vector by vectorizing a query based on the distributed representation, and obtain information on a vector similar to the generated query vector from the search DB 280. For example, the information search unit 220 may obtain information on a vector whose similarity to the query vector, such as cosine similarity, is equal to or greater than a threshold value from the search DB 280.

[0046] (5) The prompt generating unit 130 of the management device 100 generates a correction prompt based on the prompt provided by the user and the information provided by the search device 200. For example, the information processing unit 140 generates a correction prompt by adding the information provided by the search device 200 to the prompt provided to the user. The prompt generating unit 130 may generate a correction prompt that includes one or both of a request to provide information on an image related to the information provided by the search device 200 or the prompt, and a request to provide information on a URL related to the information provided by the search device 200 or the prompt. The correction prompt may include a request that specifies that a URL or an image tag must be included.

[0047] 9 is a diagram showing an example of information included in the correction prompt. The correction prompt includes, for example, a prompt provided to the user, information provided by the search device 200, information instructing that the URL specification is to be returned in {}, and information instructing that the image tag is to be returned in {}. The above URL or image tag is included in the correction prompt in the hope that it is a URL or image tag related to one or both of the prompt provided to the user and the information provided by the search device 200. The related URL or image tag is, for example, a URL or image tag corresponding to information or an image that explains, explains, or helps to understand the prompt provided to the user or the information provided by the search device 200.

[0048] The prompt generating unit 130 may include in the modification prompt a request to provide any information, such as a "machine drawing," a "table," a "flowchart," or a "system configuration diagram." For example, the prompt generating unit 130 may include in the modification prompt a request to provide information corresponding to information currently being viewed by the user. For example, when the user is viewing a machine drawing, the prompt generating unit 130 may generate a modification prompt including a request to provide the machine drawing.

[0049] When instructing to provide an image tag or the image itself, the prompt generating unit 130 may generate a correction prompt including a request to insert an image or an image tag between sentences. For example, the prompt generating unit 130 may generate a correction prompt such as "When generating an answer sentence, please insert an association diagram (actually a URL indicating the association diagram) between information A and information B." Information A and information B are sentences, information indicating a predetermined web page, and information indicating an image. In this case, when provided to the user, an attachment file related to the association diagram, a URL, or information displayed by a function of a predetermined application program may be inserted between information A and information B.

[0050] The prompt generating unit 130 may generate a correction prompt that includes compatible products or substitutes for the product or part included in the prompt, related products (parent product, child product) or parts (parent part, child part), etc. The search DB 280 manages product information and information on parts used in the products in a hierarchical structure, and further manages information on products or parts related to the target that is a product or part, or products or parts that substitute for the target. The prompt generating unit 130 may refer to the search DB 280 and generate a correction prompt that includes products or parts related to or substitutes for the product or part included in the prompt.

[0051] When a user inputs a prompt in content (e.g., a BOM screen) that includes information in a hierarchical structure, the prompt generating unit 130 obtains search results having supplemental information including information on an upper layer (e.g., a product or part that incorporates a part) or lower layer (a product or part included in a product or part) of the hierarchical structure that includes an object that is a product or part included in the query obtained by searching the search DB 280, and generates a correction prompt including information on the upper or lower layer of the hierarchical structure based on the prompt and the search results. Information on an upper layer of the hierarchical structure is information on a product or part that includes a product or part. Information on a lower layer of the hierarchical structure is information on a product or part that is included in a product or part.

[0052] (6) The management device 100 transmits the correction prompt to the language model 50, and (7) obtains an answer corresponding to the correction prompt from the language model 50. The interface screen IF5 in FIG. 10 is a diagram showing an example of an answer corresponding to the correction prompt. The answer includes information related to the prompt provided to the user or the information provided by the search device 200, related URLs, related image tags, and other information. When the correction prompt includes questions such as a question, an assignment, or a request for a proposal, the related information includes answers to these questions.

[0053] (8) The information processing unit 140 of the management device 100 processes the answer provided by the language model 50 to generate information to be provided to the user, and (9) provides the information to be provided to the user terminal device 10. The information processing unit 140 generates content that displays an image based on an image tag (information indicating the source of the image) provided from the language model 50, and provides the generated content to the user. For example, the information processing unit 140 obtains an image corresponding to the image tag included in the provided answer, and replaces the image tag with the obtained image (for example, a table, graph, drawing, etc.).

[0054] An interface screen IF6 in Fig. 10 is a diagram showing an example of processed information. As shown in an interface screen IF5, the information processing unit 140 generates information to be provided including an image, and provides the generated information to be provided to the user terminal device 10. The information processing unit 140 may provide the user with content including the information provided by the language model 50 and information on the location of information in the search DB 280 corresponding to the result of the search obtained using the query (link 1 in the figure), or may provide the user with the information in the search DB 280 corresponding to the result of the search obtained using the query included in the content.

[0055] The information processing unit 140 may emphasize a preset type of information in the information based on the output information output by the language model 50 more than other types of information. The preset type of information may be, for example, numerical data, a phrase (e.g., a part) whose relevance to the prompt is equal to or greater than a threshold, a solution to the problem of the prompt, or the like. The preset type of information may be a specific part in the drawing. The specific part is a part or corresponding part whose relevance to the prompt is equal to or greater than a threshold. Emphasis means making the information more noticeable than other information, and may involve changing the color or adding highlighting or underlining. Emphasis may involve changing the font or surrounding the information with a shape such as a circle or a square. The management device 100 executes natural language processing to analyze the relevance to the prompt, numerical data, and information written in the drawing, and identifies the part to be emphasized, etc.

[0056] The user can obtain answers to the prompt questions and information related to items, parts, etc. related to the questions. For example, the information provided by the language model 50 is information that is easier for the user to understand. Furthermore, since the provided information includes the information source, the user can obtain more detailed information and determine the credibility of the information by referring to the content of the information source. Furthermore, since the provided information includes images, the user can understand the information more intuitively.

[0057] [flowchart] When the management device 100 acquires a prompt related to a product in a hierarchical structure for a part without identifying a product, the management device 100 identifies the identification information of the product based on the hierarchical structure and the part. The management device 100 acquires a search result including supplementary information related to the target included in the query using a query based on the prompt and the identified product, and generates a correction prompt including some or all of the supplementary information based on the prompt and the search result. For example, this process is executed when a product corresponding to a part cannot be identified from the content being viewed by the user (when content in which a hierarchical structure of parts and products is not shown is being viewed). Note that the process is also performed based on the same concept when a product is included in the prompt.

[0058] 11 is a flowchart showing an example of the flow of processing executed by management device 100. This flowchart shows an example of processing when a user enters the following prompt (question) in a chat window while viewing the content of a certain part. The prompt is "What kind of packaging box is suitable for this product?"

[0059] When the management device 100 receives the prompt (step S100), it analyzes the meaning of the prompt (step S102). The management device 100 executes natural language processing to recognize that the user wants to know the type of the packaging box and that the packaging box is for "this product." Next, the management device 100 refers to the information of the BOM structure of the search DB 280 to identify the product associated with the part of the content that the user is viewing (step S104).

[0060] Next, the management device 100 generates a query using the identified product name, packaging box, type, and words similar thereto (step S106). Next, the management device 100 obtains the results of the search using the query from the search DB 280 (step S108).

[0061] Next, the management device 100 generates a correction prompt (step S110) and obtains an answer from the language model 50 (step S112). Next, the management device 100 generates content by adding a link to the search DB 280 to the answer (step S114), and provides the generated content to the user (step S116). The added link is the link used to generate the correction prompt. This completes the processing of one routine of this flowchart.

[0062] As described above, the management device 100 can utilize a database of viewed content and BOM structures to generate appropriate revision prompts without the user having to specify a product.

[0063] Furthermore, even for prompts such as those below, an appropriate correction prompt is created, and useful information is provided to the user. When the prompt is "Can you tell me about a XX screw that is made using the same production method and material as this part?", the management device 100 identifies the part based on the content of the part that the user is viewing. Furthermore, the management device 100 refers to the properties of the identified part and the properties of other parts to identify parts that meet the conditions (see FIG. 4, for example). The management device 100 obtains information to be included in the correction prompt from the file of the identified part, etc., and generates the correction prompt. This allows the user to easily obtain the desired information.

[0064] In addition, even if the prompt is as follows, an appropriate correction prompt is created and useful information is provided to the user. When the prompt includes an instruction such as "Tell me about points to be careful about," the management device 100 may identify related / adjacent parts (parent / child, other parts included in the same hierarchy) based on the content of the part being viewed by the user and the hierarchical structure, and generate a correction prompt using a query based on the identified related / adjacent parts (for example, a query including the model number of the related / adjacent parts). This is because, in the case of the above-mentioned prompt, it may be necessary to check for interference with related / adjacent parts rather than with a single part. In this case, the management device 100 can obtain information about related / adjacent parts from documents such as a collection of know-how in the search DB 280 using a query corresponding to the related / adjacent parts. The specific keyword may be any keyword related to "points to be careful about," "dimensions," "compatibility," "layout," etc., in addition to the above-mentioned "points to be careful about." As described above, when a specific keyword is included, the management device 100 can generate a correction prompt so that more substantial information can be obtained by including information about related / adjacent parts in the query.

[0065] In the above example, the first search range of the search DB 280 when a query based on related / adjacent parts is used may be different from the second search range of the search DB 280 when a query generated from a prompt input by the user is used. For example, the second search range may be wider than the first search range. For example, the first search range may be a predetermined search range (a predetermined range of documents). This generates a correction prompt that more appropriately reflects the intent of the user's prompt.

[0066] [Summary so far] It is known to retrieve information about product or service specifications or past designs from a database. If a user is not familiar with searching, it may be cumbersome or difficult for the user to retrieve the desired information. For example, the user may generate a prompt multiple times, or it may take time to generate the prompt. For this reason, it is desirable to retrieve the desired information efficiently.

[0067] In this embodiment, as described above, the management device 100, the search DB 280, and the language model 50 work together to provide the user with an appropriate answer to the user's question or request. Even if the user is not familiar with searches, the user can efficiently obtain the information he or she desires. For example, the management device 100 can efficiently obtain useful information for the user by generating a correction prompt using information from the search DB 280 to obtain useful information for the user from the prompt.

[0068] For example, assume that the user generates a prompt "I want to know the type and maximum total mass of cardboard boxes used for product A." The management device 100 generates a correction prompt that requests the provision of information, images, and URLs obtained from the search DB 280 using a query based on the above prompt. The management device 100 obtains information corresponding to the correction prompt from the language model 50, and provides the user terminal device 10 with content including information obtained by processing the obtained information. This content includes, for example, a natural answer that is easy for the user to understand, an image of a cardboard box, a URL containing detailed information, and a link destination of information related to the cardboard box in the search DB 280. The user can easily understand the answer and can easily check detailed information with high accuracy as necessary. In this way, the user can efficiently obtain useful information.

[0069] Furthermore, if a user is not familiar with searches, the prompt may be long or the prompt may be processed multiple times to obtain desired information. In this case, the processing load of each functional configuration may be large, and various costs such as usage fees may be high. On the other hand, if the prompt is shortened or the processing of the prompt multiple times is suppressed in order to suppress these, the user may not be able to obtain useful information. In contrast, in this embodiment, as described above, the management device 100 generates an appropriate correction prompt, so that useful information can be efficiently provided to the user while suppressing the processing load and cost.

[0070] In the above example, the management device 100 transmits the correction prompt to the language model 50. However, instead of (or in addition to) this, the correction prompt may be provided to the user terminal device 10 or another device, and the user or another device may transmit the correction prompt to the language model 50. The management device 100 can assist the user in obtaining useful information more efficiently.

[0071] [More examples of generating correction prompts] (A) Use of the information currently being viewed by the user The prompt generating unit 130 may generate a correction prompt based on information currently being viewed by a user who inputs a prompt when using the search DB 280. The query generating unit 120 may generate a query from information included in the content being viewed by the user, for example, and obtain a search result according to the generated query. For example, a query may be generated using information related to the information included in the content. The related information is, for example, information on compatible or substitute products of products or parts included in the content, related products (parent product, child product) or parts (parent part, child part), etc. For example, when a related product and a compatible or substitute product can be identified, the prompt generating unit 130 may generate a query using information on the compatible or substitute product. This is expected to enable the language model 50 to provide information that is more suited to the needs of the user.

[0072] (B) Use of the route used to arrive at the information currently being viewed by the user The prompt generating unit 130 may generate a correction prompt based on a path taken by the user to reach the information currently being viewed by the user when using the search DB 280. The query generating unit 120 generates a query from information included in the content that the user has followed to reach the content currently being viewed by the user, and obtains a search result corresponding to the generated query.

[0073] (Ba) When a user operates an operation button of a content of a predetermined project to reach a predetermined content (e.g., a content of a part), the query generating unit 120 determines information related to the predetermined project as the range of a search using the query. For example, when a user browses the content in the order of the project content, the assembly content, and the part content shown in FIG. 4 described above, it is predicted that the user wants to view the role of the part in the project, whether the part is used in a related project (e.g., a related finished product), the usage status of parts related to the part, etc. For this reason, the search range of the query generating unit 120 is determined as described above. As a result, it is expected that the language model 50 will provide information that is more suited to the needs of the user.

[0074] (Bb) When a user searches for parts and products from master information and is viewing content indicating the parts and products, the query generation unit 120 may set the search range of information in the search DB 280 to a wider range than in the case of (Ba) and perform a search using the query. For example, when a user is viewing content of parts obtained from master information by specifying dimensions, material, etc., it is predicted that the user wants to use the parts in other products and further wants to view alternatives and related parts. For this reason, the search range of the query generation unit 120 is set to a wider range as described above. As a result, it can be expected that the language model 50 will provide information that better matches the user's needs.

[0075] (C) Designation of Priority The prompt generating unit 130 may generate a correction prompt so as to give an answer to the language model 50 based on a priority order. For example, the prompt generating unit 130 may specify that a given piece of information in the correction prompt is given priority over other information. The term "prioritize" refers to specifying that the information provided by the language model 50 is given a higher priority, or requesting a more detailed answer than the answer to the other information. When a user operates an operation button for content of a given project to reach a given piece of content (e.g., a part content), a correction prompt may be generated so that information related to the project is given priority over information not related to the project. This is expected to allow the language model 50 to provide information that is more suited to the needs of the user.

[0076] (D) Use of usage history The prompt generating unit 130 may generate a correction prompt based on the prompt, the search result, and the target person's usage history regarding the use of the search DB 280. When multiple search results are obtained, the prompt generating unit 130 may generate a correction prompt by preferentially using the search result, which is information that the user has previously viewed, or may generate a correction prompt by preferentially using the search result, which is information that the user has not previously viewed. Priority means that more information obtained from the target information is included in the correction prompt than information obtained from other information, or when no search result is obtained from the target information, a search result is obtained from other information and included in the correction prompt, or a part of the target information or a part of the search result obtained from the target information is included in the correction prompt. The prompt generating unit 130 may generate a correction prompt by using the search result of information that has been viewed in the past and has a high viewing frequency (highest, the ranking of the viewing degree is a predetermined ranking) or information above a threshold. As a result, it is expected that information that is more suited to the needs of the user will be provided by the language model 50.

[0077] The prompt generator 130 may generate a correction prompt that includes all or part of a question previously asked by the user or another user related to the user. The other user may be a user who has asked the same question as the user's question or a similar question, or a user who belongs to the same department or group. The other user may be a user who has asked the same question as the user's question or a similar question and who belongs to the same department or group. This allows the user to simultaneously check information about past questions.

[0078] (E) Generating correction prompts with constraints The prompt generator 130 may generate a correction prompt including a request to impose a constraint on the amount of information output by the language model 50. The constraint on the amount of information may be, for example, the length of a sentence, the number of characters, or the number of tokens. The summary may be provided based on a user request. The request may be included in the prompt or a button provided on the interface screen for requesting a summary may be operated.

[0079] The summary may be automatically added as follows. The prompt generation unit 130 generates a correction prompt that requests a constraint if the same or similar question has been asked multiple times in the past, and generates a correction prompt that does not request a constraint if the same or similar question has not been asked in the past and is the first question. Instead of the same or similar question in the past, the same or similar product, part, issue, or problem may be targeted. When the same or similar question is asked multiple times, the amount of information may be limited according to the number of times. For example, the more the number of times, the more the amount of information may be restricted. For example, the more the same or similar question is asked, the more summarized information may be requested by the language model 50. This is expected to provide information that is more in line with the user's experience by the language model 50.

[0080] A request may be made so that the amount of information varies depending on the number of times. For example, a modified prompt may be generated so that more specific or detailed information is provided as the number of times increases. For example, a request may be made to provide a conceptual or abstract answer the first time, and a request may be made to provide a more specific or detailed answer as the same or similar question is asked. This is expected to enable the language model 50 to provide information that is more in line with the user's experience.

[0081] [Other examples of information provided to users] (F) Use of information regarding malfunctions When the issue, product, or part that is the subject of the user's question has had a predetermined number of past malfunctions in the past, the information processing unit 140 may add information about the malfunction to the information provided to the user in addition to the information provided from the language model 50. When the user's prompt includes a question about a malfunction, the information processing unit 140 may provide information about the malfunction together with the information provided from the language model 50. The search DB 280 manages correspondence information that associates the issue, product, or part included in the prompt with information about malfunctions that have occurred in the past for the target.

[0082] The information on the defect is, for example, information managed in the search DB 280 in association with cases, products, and parts. The information on the defect is, for example, a solution plan, a guideline, a report on the defect, and the like. The solution is, for example, the latest solution, a solution that has not caused a defect after being executed, a solution proposed by multiple other users, and the like. This is expected to enable the language model 50 to provide information that is more suited to the needs of the user.

[0083] As described above, the user can easily recognize information related to the defect. For example, the user can easily recognize solutions to the defect and past cases.

[0084] According to the embodiment described above, the management device 100 generates a correction prompt based on the prompt and the search results of the search device 200, thereby making it possible to assist in obtaining useful information more efficiently.

[0085] The above describes the form for carrying out the present invention using an embodiment, but the present invention is not limited to such an embodiment, and various modifications and substitutions can be made within the scope that does not deviate from the gist of the present invention. [Explanation of symbols]

[0086] 1. Information processing system, 10. User terminal device, 50. Language model, 100. Management device, 110. Information processing unit, 120. Query generation unit, 130. Prompt generation unit, 140. Information processing unit, 180. Memory unit, 182. Program, 200. Search device, 210. Information management unit, 220. Information search unit, 280. Search DB 280

Claims

1. On the computer, obtaining prompts related to a product or a part of said product; using a prompt-based query to obtain search results obtained by searching a database of relevant information related to the product or part that is viewable by an authorized subject, the search results including supplemental information related to the subject of the query; a modification prompt for input to a trained model that has been trained to output information corresponding to input information, the modification prompt including some or all of the supplemental information being generated based on the prompt and a result of the search; program.

2. On the computer, providing information based on output information obtained by inputting the correction prompt to the trained model to a user who input the prompt; The program according to claim 1.

3. On the computer, generating the modified prompt including, in addition to the supplemental information, one or both of a request to provide information about an image related to the supplemental information or the prompt and a request to provide information about a URL related to the supplemental information or the prompt; 3. The program according to claim 1 or 2.

4. On the computer, obtaining, as supplemental information, information within a predetermined range before and after a portion of a document that includes a portion that matches or coincides with the query in the database; generating a revision prompt including said supplemental information; 3. The program according to claim 1 or 2.

5. On the computer, generating the correction prompt including information requesting that information indicating the origin of the image be included in output information output by the trained model; generating content that displays the image based on the information indicating the source of the image, in response to the output information including information indicating the source of the image, and providing the generated content to a user; 3. The program according to claim 1 or 2.

6. On the computer, Providing a user with content including information based on output information output by the trained model and information on the location of information in a database corresponding to a search result obtained using the query; 3. The program according to claim 1 or 2.

7. the database includes a plurality of chunk information pieces obtained by chunking the related information, and tag information indicating an attribute or classification of the chunk information is assigned to each of the chunk information pieces; On the computer, acquiring, as supplemental information, a plurality of chunk information pieces to which tag information corresponding to the query is assigned, from among the plurality of chunk information pieces; 3. The program according to claim 1 or 2.

8. the database manages information on the products and information on the parts used in the products in a hierarchical structure; On the computer, In the content, when a user inputs a prompt, a search result having supplemental information including information on an upper or lower level of a hierarchical structure including an object that is a product or part included in the query obtained by searching the database is obtained; generating a modified prompt including information higher or lower in the hierarchical structure based on the prompt and the results of the search; 3. The program according to claim 1 or 2.

9. the information at a higher level in the hierarchical structure is information on a product or part that includes the product or part; The information at a lower level of the hierarchical structure is information on a product or part included in the product or part. The program according to claim 8.

10. the database manages information on the products and information on parts used in the products in a hierarchical structure, and further manages information on products or parts related to the target which is the product or the part, or information on products or parts which substitute for the target; On the computer, referencing said database to generate said modified prompt including a product or part related to or a substitute for said product or part included in said prompt; 3. The program according to claim 1 or 2.

11. On the computer, generating the modification prompt including, in addition to the supplemental information, a request to constrain an amount of information on output information output by the trained model; 3. The program according to claim 1 or 2.

12. On the computer, generating the modified prompt based on the prompt, the results of the search, and a usage history of the subject regarding usage of the database; 3. The program according to claim 1 or 2.

13. On the computer, generating the modified prompt based on one or both of information currently being viewed by a user who input the prompt in using the database and a route taken by the user to arrive at the information currently being viewed; 3. The program according to claim 1 or 2.

14. On the computer, Providing a user with content based on the output information output by the trained model, the content being based on the output information in which a preset type of information is emphasized more than other types of information.

3. The program according to claim 1 or 2.

15. On the computer, With reference to correspondence information in which the target, which is the product or the part, included in the prompt is associated with information regarding a malfunction of the target that has occurred in the past, If the target malfunction has occurred in the past for a threshold or more, information based on output information obtained by inputting the correction prompt into the trained model and information about the malfunction are provided to the user who input the prompt.

3. The program according to claim 1 or 2.

16. On the computer, With reference to correspondence information in which the target, which is the product or the part, included in the prompt is associated with information regarding a malfunction of the target that has occurred in the past, providing a solution or guideline, which is information about the defect, to a user who inputs the prompt, together with information based on output information obtained by inputting the correction prompt to the trained model; 3. The program according to claim 1 or 2.

17. On the computer, the database manages information on the products and information on the parts used in the products in a hierarchical structure; On the computer, when a prompt is received for a hierarchical product for the part that does not specify a product, determining an identity of the product based on the hierarchical structure and the part; obtaining results of a search using a query based on the prompt and the identified product, the results including supplemental information related to the subject matter included in the query; generating a revised prompt including some or all of the supplemental information based on the prompt and the results of the search; 3. The program according to claim 1 or 2.

18. The computer Obtaining prompts related to a product or part of said product; using a prompt-based query to obtain search results including supplemental information related to the subject of the query obtained by searching a database of relevant information related to the product or part that is available to authorized subjects; a modification prompt for input to a trained model that has been trained to output information corresponding to the input information, the modification prompt including some or all of the supplemental information being generated based on the prompt and a result of the search; Information processing methods.

19. Obtaining prompts related to a product or part of said product; using a prompt-based query to obtain search results including supplemental information related to the subject of the query obtained by searching a database of relevant information related to the product or part that is available to authorized subjects; a modification prompt for input to a trained model that has been trained to output information corresponding to the input information, the modification prompt including some or all of the supplemental information being generated based on the prompt and a result of the search; Information processing device.

Images