Information provision device, information provision method, and information provisionprogram

Abstract

An information provision device acquires problem information relating to a material development problem transmitted from a user terminal. The information provision device transmits a plurality of pieces of viewpoint information to the user terminal. The information providing device acquires selected viewpoint information from among the plurality of pieces of viewpoint information. The information provision device: inputs, to a large-scale language model, a prompt that is based on the selected viewpoint information and that is for generating decomposition result information that is the result of performing element decomposition, from the selected viewpoint, on the material development problem indicated by the problem information; and thereby acquires a plurality of pieces of decomposition result information outputted from the large-scale language model. The information provision device acquires, on the basis of the acquired decomposition result information, document information related to the decomposition result information from among a plurality of pieces of document information, and outputs the acquired document information.

Description

Information Providing Device, Information Providing Method, and Information Providing Program

[0001] The disclosed technology relates to an information providing device, an information providing method, and an information providing program.

[0002] Conventionally, research on the applicability of GPT-4, an example of a large language model, to chemical research is known. For example, when utilizing a large language model in material development, a technique using RAG (Retrieval Augmented Generation) is known (for example, Document 1 (Yuan Chiang, Elvis Hsieh, Chia-Hong Chou, Janosh Riebesell, "LLaMP: Large Language Model Made Powerful for High-fidelity Materials Knowledge Retrieval and Distillation" (arXiv:2401.17244))).

[0003] By the way, when a user attempts to conduct material development using a large language model, there may be cases where the answers desired by the user are not output from the large language model. For example, even if the user simply inputs the issues of material development into the large language model, it is often the case that useful answers desired by the user cannot be obtained.

[0004] The disclosed technology has been made in view of the above circumstances, and aims to provide an information providing device, an information processing method, and an information processing program that can provide more useful information to a user when the user attempts to conduct material development using a large language model.

[0005] A first aspect of this disclosure is an information providing device that acquires problem information, which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmits to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints that represent a standard viewpoint for decomposing the problem used to examine the material development problem represented by the problem information; acquires the viewpoint information selected by the user operating the user terminal from among the plurality of viewpoint information; generates a prompt to generate decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint, based on the selected viewpoint information; inputs the generated prompt to a large-scale language model to acquire a plurality of decomposition result information output from the large-scale language model; acquires literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputs the acquired literature information.

[0006] A second aspect of this disclosure is an information provision method in which a computer performs processing, which involves acquiring problem information, which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmitting to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints that represent a standard viewpoint for decomposing the problem used to examine the material development problem represented by the problem information; acquiring the viewpoint information selected by the user operating the user terminal from among the plurality of viewpoint information; generating a prompt to generate decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint, based on the selected viewpoint information; inputting the generated prompt into a large-scale language model to acquire a plurality of decomposition result information output from the large-scale language model; acquiring literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputting the acquired literature information.

[0007] A third aspect of this disclosure is an information provision program that causes a computer to perform the following steps: acquire problem information, which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmit to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints that represent a standard viewpoint for decomposing the problem used to examine the material development problem represented by the problem information; acquire the viewpoint information selected by the user operating the user terminal from among the plurality of viewpoint information; generate a prompt that generates decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint, based on the selected viewpoint information; input the generated prompt to a large-scale language model to acquire a plurality of decomposition result information output from the large-scale language model; acquire literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and output the acquired literature information.

[0008] According to the disclosed technology, when users attempt to develop materials using large-scale language models, it is possible to provide users with more useful information.

[0009] This figure shows an example of the schematic configuration of the information provision system 10 of the first embodiment. This is a schematic block diagram of the computers that function as each device of the information provision system 10. This figure shows an example of a sequence executed by the information provision system 10 of the embodiment. This figure shows an example of a sequence executed by the information provision system 10 of the embodiment. This figure shows an example of a sequence executed by the information provision system 10 of the embodiment. This figure shows an example of a sequence executed by the information provision system 10 of the embodiment. This figure shows an example of a sequence executed by the information provision system 10 of the embodiment. This figure shows an example of a screen displayed on the display unit of the user terminal. This figure shows an example of a screen displayed on the display unit of the user terminal. This figure shows an example of a screen displayed on the display unit of the user terminal. This figure illustrates the second embodiment. This figure illustrates the second embodiment. This figure illustrates the second embodiment.

[0010] Embodiments of the disclosed technology will be described in detail below with reference to the drawings.

[0011] [First Embodiment] <Information Provision System 10 According to the First Embodiment> Figure 1 shows the information provision system 10 according to the first embodiment. The information provision system 10 according to this embodiment includes a plurality of user terminals 12A, 12B, 12C, a server 14 which is an example of an information provision device, a large-scale language model server 16, and a database server 18. Hereinafter, one user terminal will also be simply referred to as user terminal 12. The user terminals 12, the server 14, the large-scale language model server 16, and the database server 18 are connected to each other via a communication means 19 such as the Internet, as shown in Figure 1.

[0012] The information provision system 10 according to the first embodiment provides necessary information for performing chemical calculations related to materials development using a known large-scale language model. Therefore, the information provision system 10 according to the first embodiment also serves as a navigation tool for chemical calculations related to materials development.

[0013] (User terminal 12) User terminal 12 is a terminal operated by the user. Specifically, the user exchanges information with the server 14 by operating user terminal 12.

[0014] (Server 14) Server 14 acquires information transmitted from an external source and performs processing according to that information. As shown in Figure 1, Server 14 functionally comprises a server storage unit 140 and a first control unit 142. The server storage unit 140 stores the information necessary for performing each of the processes described later. The first control unit 142 controls Server 14.

[0015] (Large-scale language model server 16) The large-scale language model server 16 acquires information transmitted from an external source and performs processing according to that information. As shown in Figure 1, the large-scale language model server 16 comprises a large-scale language model storage unit 160 and a second control unit 162.

[0016] The Large Language Model Memory Unit 160 stores known Large Language Models (LLMs). Large Language Models are so-called generative AI (Artificial Intelligence). Examples of Large Language Models include generative AI such as ChatGPT (Internet search <URL: https: / / openai.com / blog / chatgpt>) or Gemini (Internet search <URL: https: / / gemini.google.com / ?hl=ja>). Large Language Models are obtained by performing deep learning on a neural network. Prompts containing instructions are input to the Large Language Model, and inference data such as text data representing text and image data representing images are also input. The Large Language Model infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference includes, for example, analysis, classification, prediction, and / or summarization. In this embodiment, the Large Language Model is a concept that includes so-called AI agents (or agentic AI). When using an AI agent, for example, if the AI ​​agent receives input from the user, it will operate autonomously by performing processes such as creating a prompt itself.

[0017] The second control unit 162 controls the large-scale language model server 16.

[0018] (Database Server 18) The database server 18 acquires information transmitted from an external source and performs processing according to that information. As shown in Figure 1, the database server 18 functionally comprises a database 180 and a third control unit 182.

[0019] Database 180 stores multiple pieces of bibliographic information as information necessary for executing each of the processes described later. Bibliographic information includes, for example, academic literature, patent documents, or literature related to research or business conducted by the organization to which the user belongs. Generally, "literature" often refers to materials, books, publications, papers, and other documents that have a predetermined format. However, the "bibliographic information" used in this embodiment is not limited to this and is a broad concept that also includes textual information that does not have a fixed format. Specifically, internal company information used within the company to which the user belongs is also included in bibliographic information. For example, textual information used within a company that contains at least one sentence falls under bibliographic information in this embodiment. In addition, information on the company's own technology, which is information on technology owned by the company to which the user belongs, also falls under bibliographic information in this embodiment.

[0020] The third control unit 182 controls the database server 18.

[0021] The user terminal 12, the large-scale language model server 16, and the database server 18 can be implemented, for example, by the computer 50 shown in Figure 2. The computer 50 includes a central processing unit (CPU) 51, a memory 52 as a temporary storage area, and a non-volatile storage unit 53. The computer 50 also includes an input / output interface (I / F) 54 to which external devices and output devices are connected, and a read / write (R / W) unit 55 that controls the reading and writing of data to the recording medium. The computer 50 also includes a network I / F 56 that connects to a network such as the Internet. The CPU 51, memory 52, storage unit 53, input / output I / F 54, R / W unit 55, and network I / F 56 are connected to each other via a bus 57.

[0022] [Operation of the Information Provision System 10 in the First Embodiment] Next, the specific operation of the information provision system 10 in the first embodiment will be described. In each device of the information provision system 10, the processes shown in Figures 3, 4, 5, 6, and 7 are executed.

[0023] First, the user accesses the server 14 by operating their user terminal 12. The server 14 transmits predetermined input screen information to the user terminal 12. The user terminal 12 displays the input screen information transmitted from the server 14 on its display unit (not shown).

[0024] Next, in step S100 of Figure 3, the user terminal 12 receives the problem information entered by the user. The problem information is information relating to a problem in material development and is expressed in natural language.

[0025] Figure 8 shows an example of a screen displayed on the display unit (not shown) of the user terminal 12. As shown in Figure 8, for example, the user operates their user terminal 12 and enters an example of problem information, "I want to improve the performance of Li-ion conductivity at the all-solid-state battery interface," into the text box W1 on the screen G1 displayed on the display unit (not shown) of the user terminal 12, and presses the OK button. This sends the problem information from the user terminal 12 to the server 14. In the example shown in Figure 8, the problem information is referred to as a "sub-theme." In Figure 8, the problem information is shown as a sub-theme because, although the theme of improving the performance of all-solid-state batteries is a material development challenge, inputting such a broad theme into the large-scale language model described later may not necessarily produce appropriate output. Therefore, in the example shown in Figure 8, the explanation for the problem information input field is "Enter the sub-theme you want to solve" to encourage the user to narrow the scope of the theme. The server 14 may also be configured to support the creation of problem information. For example, if the user enters "all-solid-state battery" into the text box W1, the server 14 may suggest related words (for example, "I want to improve the performance of Li ion conductivity"). The generation of related words may be performed by a large-scale language model, as described later, or by storing a dictionary of words in the server 14's memory and performing the process based on that. Furthermore, the user's input of task information is not limited to text. For example, the user may input data in a format other than text, such as tabular data, image data, or chemical structure data of compounds. Moreover, text and data in a format other than text may be combined. For example, the user may be able to input chemical structure data of a compound along with the text, "I want to calculate the physical properties of this compound."

[0026] Next, in step S102, the first control unit 142 of the server 14 receives task information transmitted from the user terminal 12.

[0027] In step S104, the first control unit 142 of the server 14 acquires multiple viewpoint information based on the problem information received in step S102, for example. The viewpoint information is information that represents a reference viewpoint for breaking down a problem, used to examine the problem of material development, and is pre-prepared information. For example, the first control unit 142 of the server 14 acquires multiple viewpoint information by reading multiple viewpoint information that is pre-stored in the server storage unit 140.

[0028] Figure 8 shows W2 as an example of multiple perspective information. The multiple perspective information W2 shown in Figure 8 is a set of perspectives that serve as criteria for breaking down the problem, "We want to improve the performance of Li-ion conductivity at the solid-state battery interface." In Figure 8, "No specific perspective," "Break down by component," "Break down by manufacturing process," and "Consider development / research" are shown as examples. Users can also customize the perspective information, which is the perspective, by pressing, for example, the "Customize perspective" button. In Figure 8, the perspective information is referred to as a perspective.

[0029] The perspective information shown in Figure 8 is just one example; other perspectives such as "decompose by internal company division (of the company to which the user belongs)" or "decompose by physical properties" are also possible. For example, if the perspective information "decompose by internal company division" is selected, it is necessary to input information about that internal company division. It is possible to decompose the project information using information about the company's research themes or research organizations. For example, it is possible to decompose it using internal company divisions such as "charge carrier research themes" or "mass production process research themes."

[0030] In step S106, the first control unit 142 of the server 14 transmits the multiple viewpoint information acquired in step S104 to the user terminal 12. However, the first control unit 142 of the server 14 does not transmit the multiple viewpoint information as is, but rather transmits information to the user terminal 12 for generating a display screen containing the multiple viewpoint information. As a result, for example, the display unit (not shown) of the user terminal 12 displays a screen like the one shown in Figure 8.

[0031] In step S107, the user terminal 12 receives multiple viewpoint information transmitted from the server 14 in step S106 and displays it on its own display unit (not shown). For example, multiple viewpoint information W2 is displayed on screen G1 as shown in Figure 8.

[0032] The user selects the desired viewpoint information from among multiple viewpoint information displayed on the display unit (not shown) of their user terminal 12. For example, the user selects one viewpoint information from among multiple viewpoint information by operating their user terminal 12.

[0033] In step S108, the user terminal 12 receives the selection result of the viewpoint information selected by the user.

[0034] In step S110, the user terminal 12 sends the selection result received in step S108 to the server 14.

[0035] In step S112, the first control unit 142 of the server 14 receives the selection result transmitted from the user terminal 12 in step S110.

[0036] In step S114, the first control unit 142 of the server 14 generates a first prompt for generating decomposition result information based on the viewpoint information selected by the user in step S108. The decomposition result information is information that represents the result of decomposing the material development problem represented by the problem information entered by the user in step S100 into elements from the selected viewpoint.

[0037] Figure 9 shows an example of a screen displayed on the display unit (not shown) of the user terminal 12. W3 shown in Figure 9 is an example of decomposition result information when the viewpoint information is "decompose by constituent elements". In the example in Figure 9, "Let's check the ionic conductivity", "Let's check the interfacial contact resistance", "Let's check the structure and composition of the interfacial layer", "Let's check the temperature", "Let's check the stress and strain", and "Let's check the temperature and mobility of the charge carriers" are shown as examples of decomposition result information W3. By inputting the first prompt generated in step S114 to the large-scale language model, decomposition result information W3 as shown in Figure 9 is output from the large-scale language model. In the example shown in Figure 9, the message "Consider countermeasures for all elements" is also displayed.

[0038] Furthermore, if, for example, the user selects viewpoint information, the first control unit 142 of the server 14 generates a first prompt according to the following prompt template. The [Issue Information] section of the prompt template contains text information representing the issue information, and the [Viewpoint Information] section contains text information representing the viewpoint information.

[0039] (Example of the first prompt) Decompose the [Problem Information] into its elements according to the following rules and generate themes that should be addressed to improve performance. (Rule 1) Decompose the [Problem Information] according to the [Perspective Information (e.g., chemical or physical components)] as themes that should be researched in order to realize the [Problem Information], and list and display the decomposition results. (Rule 2) If there are 10 or more components, the scope of the problem in the [Problem Information] is too broad, so generate a message informing the user to narrow down the scope of the problem.

[0040] In step S116, the first control unit 142 of the server 14 sends the first prompt generated in step S114 to the large-scale language model server 16.

[0041] In step S118, the second control unit 162 of the large-scale language model server receives the first prompt sent from the server 14 in step S116.

[0042] In step S120, the second control unit 162 of the large language model server 16 inputs the first prompt received in step S118 to the large language model stored in the large language model storage unit 160.

[0043] In step S122, the second control unit 162 of the large language model server 16 acquires a plurality of decomposition result information output from the large language model.

[0044] In step S124, the second control unit 162 of the large language model server 16 transmits the plurality of decomposition result information acquired in step S122 to the server 14.

[0045] In step S126, the first control unit 142 of the server 14 receives the plurality of decomposition result information transmitted from the large language model server 16 in step S124.

[0046] In step S128, the first control unit 142 of the server 14 transmits the plurality of decomposition result information received in step S126 to the user terminal 12. Note that the first control unit 142 of the server 14 does not transmit the plurality of decomposition result information as it is, but transmits, for example, information for generating a display screen including the plurality of decomposition result information to the user terminal 12. Thereby, for example, a screen as shown in FIG. 9 is displayed on the display unit (not shown) of the user terminal 12.

[0047] In step S130, the user terminal 12 receives the plurality of decomposition result information transmitted from the server 14 in step S128 and displays it on its own display unit (not shown). For example, a plurality of decomposition result information W3 is displayed within the screen G2 as shown in FIG. 9.

[0048] The user selects the decomposition result information that the user desires from among the plurality of decomposition result information displayed on the display unit (not shown) of the user's own user terminal 12. For example, the user selects one decomposition result information from among the plurality of decomposition result information by operating the user's own user terminal 12. For example, the user selects the decomposition result information "Let's check regarding ionic conductivity" shown in FIG. 9 by operating the user's own user terminal 12.

[0049] In step S132, the user terminal 12 receives the selection result of the decomposition result information selected by the user.

[0050] In step S134, the user terminal 12 transmits the selection result received in step S132 to the server 14.

[0051] In step S136, the first control unit 142 of the server 14 receives the selection result transmitted from the user terminal 12 in step S134.

[0052] Next, in steps S138 to S144, the first control unit 142 of the server 14 obtains the first document information related to the decomposition result information from among a plurality of document information based on the selection result of the decomposition result information obtained in step S136.

[0053] Specifically, the first control unit 142 of the server 14 uses a known Retrieval Augmented Generation (RAG) to obtain the first document information related to the decomposition result information from among a plurality of document information, and includes information related to the first document information in a prompt to obtain an appropriate output from a large language model. RAG is a method of obtaining a document suitable for a question from a database using so-called vector search or the like, and causing a large language model to answer based on the information in the document. For example, RAG performs preprocessing such as vectorization or conversion to a QA format on the content of a document, prepares a database for each document in advance, and uses at least one of the task information and the decomposition result information selected by the user to extract a document highly relevant to the input information. At this time, by including documents in the database in advance, for example, as documents related to material development or in-house documents, the probability of extracting highly relevant documents can be increased.

[0054] For example, the first control unit 142 of server 14 adds two pieces of text information—the task information and the decomposition result information selected by the user—and vectorizes them. Using vector search, it extracts documents with high geometric similarity as the first document information. Based on the first document information, the first control unit 142 of server 14 causes a large-scale language model to generate text that includes a detailed explanation of the decomposition result information.

[0055] Specifically, in step S138, the first control unit 142 of the server 14 sends a signal to the database server 18 requesting bibliographic information related to the disassembly result information represented by the selection result, based on the selection result of the disassembly result information received in step S136.

[0056] In step S140, the third control unit 182 of the database server 18 receives the request signal sent from the server 14 in step S138.

[0057] In step S142, the third control unit 182 of the database server 18 retrieves first document information related to the decomposition result information from among multiple document information by searching its own database 180. Then, the third control unit 182 of the database server 18 transmits the first document information to the server 14.

[0058] In step S144, the first control unit 142 of the server 14 receives the first document information transmitted from the database server 18 in step S142.

[0059] In step S146, the first control unit 142 of the server 14 generates a second prompt for generating explanatory information about the disassembly result information selected by the user, based on the first document information received in step S144.

[0060] For example, the first control unit 142 of the server 14 generates a second prompt using the following prompt template. The [Issue Information] section of the prompt template contains text information representing the issue information, the [(Selected) Decomposition Result Information] section contains text information representing the decomposition result information selected by the user, and the [First Document Information] section contains text information related to the extracted first document information. The [First Document Information] section may contain one or more documents. If all of the decomposition result information generated by the large-scale language model were to be input back into the large-scale language model, the data would become too large or the number of times the large-scale language model is executed would increase too much. Therefore, in this embodiment, it is assumed that only the decomposition result information selected by the user is input into the large-scale language model.

[0061] (Example of the second prompt) For the [Selected Decomposition Result Information] needed to solve or achieve the [Problem Information], refer to the following paper and generate text to present a detailed explanation and references to the user. References: [First Document Information]

[0062] Note that if you enter the entire paper represented by the first reference information into the [First Reference Information] section above, the prompt may become too long (for example, if the first reference information is a review article, it is expected that the number of pages will exceed 50). Therefore, you may choose to run the above RAG to extract the chapters that are found in the search and input only the text information of those chapters into the large-scale language model. Alternatively, when storing the literature in the literature database included in the RAG system, you may store the literature chapter by chapter.

[0063] Alternatively, as shown in the example below, you can pre-define the style for displaying each piece of bibliographic information and include it in the second prompt.

[0064] Example: Author(s). Article Title. Journal Name, Publication Year, Volume Number, Issue Number, Pages (Start Page - End Page). *The format "Volume Number, Issue Number," can also be abbreviated as "Volume Number (Issue Number),".

[0065] In step S148, the first control unit 142 of the server 14 sends the second prompt generated in step S146 to the large-scale language model server 16.

[0066] In step S150, the second control unit 162 of the large-scale language model server 16 receives the second prompt sent from the server 14 in step S148.

[0067] In step S152, the second control unit 162 of the large-scale language model server 16 inputs the second prompt received in step S150 to the large-scale language model stored in the large-scale language model storage unit 160.

[0068] In step S154, the second control unit 162 of the large-scale language model server 16 acquires first explanatory information representing an explanation of the decomposition result information output from the large-scale language model.

[0069] In step S156, the second control unit 162 of the large-scale language model server 16 transmits the first explanatory information acquired in step S154 to the server 14.

[0070] In step S158, the first control unit 142 of the server 14 receives the first explanatory information transmitted from the large-scale language model server 16 in step S156.

[0071] In step S160, the first control unit 142 of the server 14 transmits the first document information acquired in step S144 and the first explanatory information acquired in step S158 to the user terminal 12.

[0072] In step S162, the user terminal 12 receives the first bibliographic information and the first commentary information transmitted from the server 14 in step S160.

[0073] The user confirms the first bibliographic information and the first explanatory information displayed on the display unit (not shown) of their user terminal 12.

[0074] As shown in Figure 9, for example, if the decomposition result information selected in step S132 is "Let's check the ionic conductivity," then the first explanatory information W4 and the first reference information W5, which represent an explanation of the decomposition result information, are displayed on the display unit (not shown) of the user terminal 12.

[0075] If the user wishes to obtain further considerations, they input an operation to the user terminal 12 to instruct the generation of considerations (expressed as "actions" in the example in Figure 5). For example, if the user presses the "Consider next action" button by operating their user terminal 12, the generation of consideration information related to the first explanatory information W4 and the first bibliographic information W5 will begin. If the "Consider actions for all elements" button shown in Figure 9 is pressed, the generation of consideration information for all decomposition result information will be performed.

[0076] In step S164, when the user terminal 12 receives an operation from the user indicating an instruction to generate a draft proposal, it sends a signal to the server 14 requesting the generation of the draft proposal.

[0077] In step S166, the first control unit 142 of the server 14 receives the request signal transmitted from the user terminal 12 in step S164.

[0078] In step S168, the first control unit 142 of the server 14 generates a third prompt for generating proposal information for considering experiments or chemical calculations using the method represented by the decomposition result information based on the first literature information. The proposal information is information for considering experimental methods or chemical calculations related to the decomposition result information, and is, for example, information regarding experiments or chemical calculations necessary to consider the contents of the decomposition result information based on the first literature information.

[0079] For example, the first control unit 142 of the server 14 generates a third prompt using the following prompt template. The [Issue Information] section of the prompt template contains text information representing the issue information, the [(Selected) Decomposition Result Information] section contains text information representing the decomposition result information selected by the user, and the [First Document Information] section contains text information related to the extracted first document information. The [First Document Information] section may contain one or more documents.

[0080] (Example of the third prompt) We want to consider [(selected) decomposition result information] in order to solve or achieve [problem information]. Refer to the following paper and list several proposals for what points should be considered. References: [First reference information]

[0081] Figure 10 shows an example of a screen displayed on the display unit (not shown) of the user terminal 12. The "Actions" shown in Figure 10 correspond to multiple proposed actions when the selected decomposition result information is "Let's check the ionic conductivity." By inputting the third prompt generated in step S168 to the large-scale language model, multiple proposed actions as shown in Figure 10 are output from the large-scale language model.

[0082] In step S170, the first control unit 142 of the server 14 sends the third prompt generated in step S168 to the large-scale language model server 16.

[0083] In step S172, the second control unit 162 of the large-scale language model server 16 receives the third prompt sent from the server 14 in step S170.

[0084] In step S174, the second control unit 162 of the large-scale language model server 16 inputs the third prompt received in step S172 to the large-scale language model stored in the large-scale language model storage unit 160.

[0085] In step S176, the second control unit 162 of the large-scale language model server 16 acquires multiple draft information output from the large-scale language model.

[0086] In step S178, the second control unit 162 of the large-scale language model server 16 transmits the multiple proposal information acquired in step S176 to the server 14.

[0087] In step S180, the first control unit 142 of the server 14 receives multiple draft information transmitted from the large-scale language model server 16 in step S178.

[0088] Next, in steps S182 to S188, the first control unit 142 of the server 14 acquires second document information related to each of the multiple draft information from among the multiple document information, based on the multiple draft information acquired in step S180.

[0089] Specifically, the first control unit 142 of the server 14 uses a known RAG (Retrieval Augmented Generation) to obtain second-hand document information related to the proposed plan information from among multiple pieces of document information, and includes information related to that second-hand document information in the prompt, thereby obtaining appropriate output from the large-scale language model. For example, the first control unit 142 of the server 14 vectorizes each of the generated proposed plan information and uses RAG to extract the second-hand document information corresponding to each of the proposed plan information.

[0090] First, in step S182, the first control unit 142 of the server 14 sends a signal to the database server 18 requesting second document information related to the proposed information, based on the proposed information received in step S180.

[0091] In step S184, the third control unit 182 of the database server 18 receives the request signal sent from the server 14 in step S182.

[0092] In step S186, the third control unit 182 of the database server 18 retrieves second document information related to the proposed information from among multiple document information by searching its own database 180. Then, in step S186, the third control unit 182 of the database server 18 transmits the retrieved second document information to the server 14.

[0093] In step S188, the first control unit 142 of the server 14 receives the second document information transmitted from the database server 18 in step S186.

[0094] In step S190, the first control unit 142 of the server 14 generates a fourth prompt for generating second explanatory information that represents an explanation of the proposed information based on the second document information.

[0095] For example, the first control unit 142 of the server 14 generates a fourth prompt using the following prompt template. The [Issue Information] section of the prompt template contains text information representing the issue information, the [(Selected) Decomposition Result Information] section contains text information representing the decomposition result information selected by the user, and the [Second Document Information] section contains text information relating to the extracted second document information. The [Second Document Information] section may contain one or more documents.

[0096] (Example of the fourth prompt) To solve or achieve the [Problem Information], we would like to consider the [Proposed Plan Information] as part of the [(Selected) Decomposition Results Information]. Please refer to the following paper and generate a detailed explanation of the [Proposed Plan Information]. References: [Second Document Information]

[0097] In step S192, the first control unit 142 of the server 14 sends the fourth prompt generated in step S190 to the large-scale language model server 16.

[0098] In step S194, the second control unit 162 of the large-scale language model server 16 receives the fourth prompt sent from the server 14 in step S192.

[0099] In step S196, the second control unit 162 of the large-scale language model server 16 inputs the fourth prompt received in step S194 to the large-scale language model stored in the large-scale language model storage unit 160.

[0100] In step S198, the second control unit 162 of the large-scale language model server 16 acquires the second explanatory information output from the large-scale language model. The second explanatory information may include the results of linking with the organizational chart of the organization to which the user belongs.

[0101] In step S200, the second control unit 162 of the large-scale language model server 16 transmits the second explanatory information acquired in step S198 to the server 14.

[0102] In step S202, the first control unit 142 of the server 14 receives the second explanatory information transmitted from the large-scale language model server 16 in step S200.

[0103] In step S204, the first control unit 142 of the server 14 transmits the draft information received in step S180, the second literature information received in step S188, and the second explanatory information received in step S202 to the user terminal 12. However, the first control unit 142 of the server 14 does not transmit the above information as is, but rather transmits information to the user terminal 12 for generating a display screen that includes the above information. As a result, for example, a screen like the one shown in Figure 10 is displayed on the display unit (not shown) of the user terminal 12.

[0104] In step S206, the user terminal 12 receives the proposed solutions information, second literature information, and second explanatory information transmitted by the server 14 in step S204 and displays them on its own display unit (not shown). For example, on screen G3 as shown in Figure 10, multiple proposed solutions information (expressed in Figure 10 as "Action 1: Model the crystal structure of the characteristics of the ion conduction path," "Action 2: Model the atomic arrangement of the interface structure and the presence of impurities," and "Action 3: Model the thickness of the interface layer of the interface structure"), second literature information (the "Paper Information" and "Internal Information" parts of Figure 10), and second explanatory information (the "Purpose" part of Figure 10) are displayed.

[0105] In the example shown in Figure 10, the objective of the proposed plan is generated as the second explanatory information. However, it is also possible to generate the second explanatory information to include further considerations (for example, what experiments or calculations should be performed), and what kind of modeling and computational resources should be prepared. In this case, for example, by inputting available computational resources into the prompts for input to the large-scale language model, it becomes possible to generate the second explanatory information described above.

[0106] As described above, the server 14 of the information provision system 10 in the first embodiment acquires problem information, which is information about a material development problem transmitted from a user terminal and expressed in natural language. The server 14 transmits to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints that represent a standard viewpoint for decomposing the problem used to examine the material development problem represented by the problem information. The server 14 acquires the viewpoint information selected by the user operating the user terminal from among the plurality of viewpoint information. Based on the viewpoint information selected by the user, the server 14 generates a prompt to generate decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint. The server 14 inputs the generated prompt into a large-scale language model and acquires a plurality of decomposition result information output from the large-scale language model. Then, based on the acquired decomposition result information, the server 14 acquires literature information related to the decomposition result information from a plurality of literature information and outputs the acquired literature information. This makes it possible to provide the user with more useful information when the user tries to develop materials using the large-scale language model.

[0107] [Second Embodiment] Next, a second embodiment will be described. In the second embodiment, based on information about the material development challenges faced by the user and information about the company's own technologies existing within the company to which the user belongs (hereinafter also simply referred to as "company technology information"), the company identifies its own technologies that are useful or not useful in solving the material development challenges faced by the user. In other words, the second embodiment analyzes the strengths and weaknesses of the company's own technologies in solving the user's challenges. Useful company technologies are those that directly solve the user's challenges (for example, company technologies that solve the challenges on their own) or those that do not directly solve the user's challenges but can solve them in combination with other technologies. On the other hand, not useful company technologies are those that cannot solve the user's challenges.

[0108] Figures 11 and 12 are diagrams illustrating the second embodiment.

[0109] The information provision system 10 of the second embodiment identifies proprietary technologies that are useful or not useful in solving the user's material development challenges, based on the user's material development challenges and proprietary technology information, and determines the priority of approaches to solving the user's challenges. For example, it presents a priority order of proprietary technologies that are likely to solve the user's challenges and approaches using those proprietary technologies. For example, as shown in Figure 11, the perspective information V of the first embodiment described above is identified from the material development challenge-related information P of the company to which the user belongs or from the challenge information of the first embodiment described above. Note that the challenge-related information is, for example, information related to material development challenges extracted from the company's papers or patent information. Then, multiple decomposition result information D1 to D5 is generated from the perspective information V. Next, the information provision system 10 of the second embodiment obtains the degree of relevance between the decomposition result information and the proprietary technology information for each combination of the multiple decomposition result information D1 to D5 and the multiple proprietary technology information T by referring to multiple proprietary technology information T. Then, it presents approaches using the proprietary technology as a priority based on that degree of relevance. For example, as shown in Figure 11, a structural understanding and prioritization of the user's problem is performed. Specifically, consider the case shown in Figure 11 where the correlation between the decomposition result information D1 and the company's technology T1 is high, and the correlation between the decomposition result information D2 and the company's technology T2 is also high. In this case, the approach using the company's technologies T1 and T2, which have high correlations, will have a higher priority, and the priority of other approaches will be lower. The information provision system 10 of the second embodiment outputs the company's technology information obtained in this way and the priority of the approach using that company technology information as results. This makes it possible to perform a structural understanding and prioritization of the user's problem. Alternatively, if some kind of company technology (for example, hollow polymer technology) exists, it becomes possible to identify a problem that is suitable for it.

[0110] Alternatively, for example, in the second embodiment, useful and unuseful proprietary technologies are identified, and based on the degree of relevance between these identified results and external technologies different from the proprietary technologies, external technologies that should be combined with the proprietary technologies to solve the user's problem are selected. Then, in the second embodiment, information (e.g., explanatory text) is output showing how the user's problem can be solved by combining the selected external technologies with the proprietary technologies. For example, if a proprietary technology that does not directly solve the user's problem but can solve it when combined with other technologies is identified as a useful proprietary technology, then external technologies that can solve the problem when combined with that proprietary technology are identified, and that information is output. Alternatively, for example, if only unuseful proprietary technologies are identified, then external technologies that directly solve the problem are identified, and that information is output.

[0111] Alternatively, for example, in the second embodiment, useful and unuseful proprietary technologies are identified, and based on these identification results, a combination of multiple proprietary technologies is selected to solve the user's problem. Then, in the second embodiment, information (e.g., explanatory text) is output showing how the user's problem can be solved by the selected combination of proprietary technologies. Even if a problem cannot be solved by any one of the multiple proprietary technologies, it may be possible to solve the problem by combining one proprietary technology with another.

[0112] Alternatively, for example, in the second embodiment, useful and unuseful proprietary technologies are identified, and based on these identification results, a combination of multiple proprietary technologies is selected to solve the user's problem. Then, in the second embodiment, information (e.g., explanatory text) is output showing how the user's problem can be solved by the selected combination of proprietary technologies.

[0113] For example, as shown in Figure 12, an understanding of technical approaches to compensate for shortcomings in the company's own technology is carried out. Specifically, as shown in Figure 12, first, multiple decomposition result information D1 to D5 are generated. Now, consider a case where the company has set a challenge, such as the performance achievement rate being 60% when using the company's own technology represented by the company's own technology information T1, and the goal is to increase this performance achievement rate to 100%. In this case, the information provision system 10 of the second embodiment, by referring to the company's own technology information T1, obtains the degree of relevance between the decomposition result information and the company's own technology information for each combination of the multiple decomposition result information D1 to D5 and the company's own technology information T1. Then, suppose that decomposition result information D1 and D4 with a high degree of relevance to the company's own technology information T1 are identified. The information provision system 10 of the second embodiment, by referring to the company's internal database or existing technology information, identifies other technology information TA and TB that should be combined with the company's own technology information T1. For example, other technology information TA is a different company technology from the company's own technology information T1. Furthermore, for example, other technical information (TA) may be external technologies existing outside the company to which the user belongs. This can result in, for example, a 100% performance achievement rate. Therefore, when performance cannot be achieved with current in-house technology alone for a specific challenge, it becomes possible to identify technical approaches that can compensate for the shortcomings. The method for identifying the technical information to combine based on the acquired relevance can be set by the user or the AI ​​(for example, the AI ​​agent mentioned above). Specifically, for example, the user can specify to combine information that has a low relevance to the company's technology and a high relevance to the technology of another company. This makes it easier to extract other companies' technical assets that the company does not possess. Moreover, this setting can be set not only by the user, but also by the AI, or the AI ​​can make recommendations from all output options according to predetermined criteria.

[0114] Thus, in the second embodiment, based on the proprietary technologies currently possessed by the company to which the user belongs, it becomes possible to present information such as "what is currently lacking and what should be done now" in order for the user to solve the material development challenges. This will be explained in detail below.

[0115] Figure 13 is a diagram illustrating the processing flow performed by the information provision system 10 of the second embodiment. The first control unit 142 of the server 14 of the information provision system 10 of the second embodiment performs each of the processes shown in Figure 13. Since the configuration of the information provision system 10 of the second embodiment is the same as that of the first embodiment, a detailed explanation is omitted. The literature database 180 of the second embodiment stores multiple pieces of proprietary technical information, which are examples of literature information. Therefore, each of the multiple pieces of literature information in the second embodiment corresponds to proprietary technical information representing technical information owned by the company to which the user belongs.

[0116] (Improved Resolution) First, in Phase Ph1, the problem is structured. For example, by performing the same processing as in the first embodiment described above, multiple perspective information V1 to V5 are obtained from the problem information or problem-related information P. Note that the problem-related information P can be identified from the publication information or patent information of the company to which the user belongs.

[0117] (Factor Decomposition) Next, in Phase Ph2, the approach is structured. Specifically, by performing the same process as in the first embodiment described above, the viewpoint information V selected by the user is decomposed into factors, and multiple decomposition result information D1 to D5 are identified. In this process, existing technical information TS may be referenced. In the first embodiment described above, as shown in Figure 9, examples of decomposition result information included "Let's check the ionic conductivity," "Let's check the interfacial contact resistance," "Let's check the structure and composition of the interfacial layer," "Let's check the temperature," "Let's check the stress and strain," and "Let's check the temperature and mobility of the charge carriers." However, the decomposition result information is not limited to these. For example, information such as "Improve ionic conductivity," "Reduce interfacial contact resistance," "Change the structure and composition of the interfacial layer," "Change the temperature," "Change the stress and strain," and "Change the temperature and mobility of the charge carriers" may also be used as decomposition result information. The degree of relevance between such decomposition result information and the company's own technical information is calculated in the "Relevance Determination" phase described later.

[0118] (Relevance Determination) In Phase Ph3, the positioning of the company's own technology is understood. The first control unit 142 of the server 14 in the second embodiment obtains the degree of relevance between the decomposition result information and the company's own technology information for each combination between each of the multiple decomposition result information and each of the multiple company's own technology information T, based on the multiple decomposition result information D1 to D5. For example, the first control unit 142 of the server 14 obtains the degree of relevance using the RAG described above. Specifically, the first control unit 142 of the server 14 vectorizes each of the multiple decomposition result information and each of the multiple company's own technology information using RAG, and determines the target population according to the similarity between the vectors. For example, the first control unit 142 of the server 14 sets each of the company's own technology information whose similarity between the vector of the decomposition result information and the vector of the company's own technology information is greater than or equal to a predetermined value as the target population. Then, the first control unit 142 of the server 14 has a large-scale language model calculate the degree of relevance between each of the multiple company's own technology information included in the population and each of the multiple decomposition result information. For example, the first control unit 142 of the server 14 inputs each of the multiple proprietary technical information items and each of the multiple decomposition result information items included in the population into the large-scale language model, and also inputs a prompt into the large-scale language model instructing it to calculate the degree of association between the decomposition result information and the proprietary technical information for each combination of each of the multiple proprietary technical information items and each of the multiple decomposition result information items. As a result, the degree of association is calculated for each combination of each of the multiple proprietary technical information items and each of the multiple decomposition result information items. For example, as shown in Figure 13, different degrees of association are calculated for each of the multiple decomposition result information items D1 to D5 (the differences in patterns indicate different degrees of association).

[0119] The first control unit 142 of the server 14 may output the company's own technical information and its relevance at this point. This allows user U to recognize that the disassembly result information and the company's own technical information are related, and to confirm the company's own technical information that is useful for solving the problem. For example, consider the case where the relevance between the disassembly result information D1 and the company's own technical information T1, as shown in Figure 13, is high, and the relevance between the disassembly result information D2 and the company's own technical information T2 is also high. In such a case, information can be obtained such as that the company's own technology T1 can solve the problem related to the disassembly result information D1 on its own, or that the company's own technology T2 can solve the problem related to the disassembly result information D4 on its own. Also, for example, if the relevance between the disassembly result information D2 and the company's own technical information T3, as shown in Figure 13, is low, the relevance between the disassembly result information D3 and the company's own technical information T5 is zero (extremely low), and the relevance between the disassembly result information D5 and the company's own technical information T4 is moderate, then in the "idea proposal" phase described later, technologies to be combined with the company's own technology will be proposed.

[0120] (Idea Proposal) In Phase 4, the idea is refined into a concrete idea. The first control unit 142 of the server 14 identifies target technology information that is necessary to solve the problem and is to be combined with the company's own technology, based on multiple levels of relevance. For example, the first control unit 142 of the server 14 acquires target technology information by inputting each of the multiple decomposition result information, each of the multiple levels of relevance information, and predetermined instruction information into a large-scale language model. The instruction information is information that indicates the type of information to be output from the large-scale language model. For example, information that instructs the system to search for external technology necessary to deal with a certain decomposition result information, information that instructs the system to search for external technology TX that will be effective when combined with the company's own technology T1 to deal with a certain decomposition result information, or information that instructs the system to search for another company technology T2 that will be effective when combined with the company's own technology T1 to deal with a certain decomposition result information is input into the large-scale language model as instruction information. The large-scale language model then acquires the target technology information by, for example, referring to existing technology information TS via the internet. Therefore, the target technical information is technical information different from the company's own technical information, and is either external technical information existing outside the company, or internal technical information different from the internal technical information to be combined. Furthermore, the large-scale language model outputs textual information according to the instruction information. For example, textual information such as "We propose approaching this problem by combining our own technology A and external technology B. This is because..." is output from the large-scale language model.

[0121] Furthermore, the first control unit 142 of the server 14 may be configured to only perform the process of obtaining proprietary technical information related to the disassembly result information from among multiple pieces of proprietary technical information based on the acquired disassembly result information, and outputting the acquired proprietary technical information.

[0122] As described above, the literature database 180 of the server 14 of the information provision system 10 in the second embodiment stores proprietary technical information representing the technical information owned by the company to which the user belongs, as each of the multiple pieces of literature information. Based on the multiple pieces of decomposition result information acquired, the server 14 of the information provision system 10 in the second embodiment acquires the degree of relevance between the decomposition result information and each of the multiple pieces of proprietary technical information for each combination between each of the multiple pieces of decomposition result information and each of the multiple pieces of proprietary technical information, and outputs the acquired proprietary technical information and the degree of relevance. Specifically, based on the multiple pieces of decomposition result information, the server 14 of the information provision system 10 in the second embodiment acquires the degree of relevance between the decomposition result information and each of the multiple pieces of proprietary technical information for each combination between each of the multiple pieces of decomposition result information and each of the multiple pieces of proprietary technical information, and identifies target technical information that is necessary to solve the problem and represents the technology to be combined with the proprietary technology. The server 14 of the information provision system 10 in the second embodiment then outputs the acquired proprietary technical information and the target technical information. Note that the target technical information is technical information different from the proprietary technical information and is external technical information existing outside the company, or proprietary technical information different from the proprietary technical information to be combined. This allows users to gain insights into what is currently lacking or what needs to be done at this time to solve their problems, based on internal company information and their own technical data.

[0123] Furthermore, the second embodiment allows for proposals that take into account the accumulation of the company's past technologies. For example, since the company's technical information related to factor X (decomposition result information X) and factor Y (decomposition result information Y) exists, it becomes possible to obtain suggestions such as increasing the priority of factor Z (decomposition result information Z). Alternatively, for example, since the company's technical information related to factor X (decomposition result information X) and factor Y (decomposition result information Y) exists, and it is possible to identify factor Z (decomposition result information Z) by combining these, it becomes possible to obtain suggestions such as it would be good to use the company's technology. As mentioned above, the instruction information input to the large-scale language model is selected in advance by the user, so information that meets the user's requirements is output from the large-scale language model.

[0124] Furthermore, the technology disclosed herein is not limited to the embodiments described above, and various modifications and applications are possible without departing from the gist of this disclosure.

[0125] [Modification 1] In the above embodiment, the server 14 was described as ultimately outputting draft information, second literature information, and second explanatory information, but it is not limited to this. For example, the server 14 may acquire literature information related to the decomposition result information from among multiple literature information based on the decomposition result information output from the large-scale language model, and output the acquired literature information as the final output. As mentioned above, the literature information may be academic literature or internal company documents, etc.

[0126] [Modification 2] In the above embodiment, the example of the server 14 using various prompt templates when generating prompts was described, but it is not limited to this. For example, when the server 14 generates prompts, it may use an aggregated template as shown below. The aggregated template below is a template that includes selected viewpoint information and issue information, and prompts are generated based on pre-set rules. In the following example, prompts are generated according to rules corresponding to the viewpoint information.

[0127] (Aggregated template) Based on the following [Perspective Information], decompose the [Issue Information] into its elements. Apply Rule X when the Perspective Information is A, and Rule Y when the Perspective Information is B.

[0128] Furthermore, for example, the second, third, or fourth prompt may include bibliographic information of the literature.

[0129] [Modification 3] The viewpoint information of the above embodiment may include at least one of the material structure and the material manufacturing process.

[0130] [Modification 4] When the server 14 generates the proposal information, it may also calculate the confidence level of the proposal information. For example, the server 14 calculates the confidence level of the proposal represented by the proposal information based on rules set in advance according to the attributes of the second literature information related to the proposal information, and outputs the confidence level of the proposal. Alternatively, for example, the server 14 obtains the confidence level of the proposal represented by the proposal information by inputting a confidence prompt, which is a prompt for generating the confidence level of the proposal represented by the proposal information, into a large-scale language model, and outputs the confidence level of the proposal. In this case, the server 14 may, for example, sort and output multiple proposals according to the confidence level of the proposal information. The server 14 may also determine the confidence level of each proposal (for example, how effective the proposal seems, how much it will help improve performance, etc.) based on whether a corresponding paper or in-house research result can be found for each proposal in RAG. Furthermore, the confidence level may be determined from references selected from the database, based on rules such as: internal research results > papers with an impact factor (IF) higher than a certain value > papers with an impact factor (IF) below a certain value. In this case, the IF can be determined by journal or by individual document. If multiple documents are extracted, the value of the most similar document is intended to be used, but the mean or median may also be used. Alternatively, the confidence level may be generated by, for example, sorting the papers for each proposed idea, in order of importance for solving the [problem information], as output from the large-scale language model. In addition, accuracy can be improved by having the large-scale language model determine whether "this has already been researched or not" based on the reference papers (for example, by prompting it to determine that if a paper corresponding to one proposed idea does not match that proposed idea (does not contain details), it has not been researched). Also, if the large-scale language model determines that a topic has not been researched, it may be asked to generate an estimate of why it hasn't been researched.

[0131] [Modification 5] Server 14 may also obtain a computational model or research procedure to solve the material development problem represented by the problem information. In this case, Server 14 generates a research prompt, which is a prompt to obtain a computational model or research procedure to solve the material development problem represented by the problem information, based on the proposed information and the problem information. Server 14 then inputs the generated research prompt into the large-scale language model, obtains the computational model or research procedure output from the large-scale language model, and outputs the computational model or research procedure. In this case, for example, Server 14 may have the large-scale language model refer to papers, present research results for those that have already been researched, and for those that have not yet been researched (for example, if the crystal structure has not been modeled), indicate that there are no research results and include in the research prompt a sentence such as "Show how it can be modeled." In this case, Server 14 may have the large-scale language model determine whether or not there are research results, and if there are no research results, it may have the large-scale language model create an SQL statement that causes RAG to search again for "literature to realize the proposed information (for example, papers for modeling crystal structures)," and then present how to model using the papers extracted using that SQL statement. In addition, at this time, the server 14 may also search for literature that shows why modeling is impossible, and may be instructed to indicate that it is impossible to model the large-scale language model. In this case, the prompt generated by the server 14 may be, for example, the following:

[0132] Prompt example: [Proposed Information] is a proposal being considered as part of [Decomposition Results Information] to solve or realize [Problem Information]. Propose the extraction of literature to be referenced and the method of execution for implementing these proposals according to the following rules. (Rule 1) From the extracted literature below, determine whether [Proposed Information] has already been studied. If it has already been studied, create and present a computational model for implementing [Proposed Information] based on the literature describing the research results. (Rule 2) If [Proposed Information] has not yet been studied, extract papers that would be useful for researching [Proposed Information]. Furthermore, present the research procedure for [Proposed Information] using those reference papers. (Rule 3) If there is evidence in the literature that [Proposed Information] is unrealistic, state the reasons why it is unrealistic. Rule 3 takes precedence over Rule 2.

[0133] [Modification 6] In the above embodiment, the case in which the database server 18 has one database 180 was described as an example, but it is not limited to this. For example, there may be multiple databases, and different types of bibliographic information may be stored in each of these multiple databases. In this case, the server 14 may store multiple bibliographic information in each of the multiple databases, and when retrieving specific bibliographic information from the multiple bibliographic information, it may retrieve the specific bibliographic information from the multiple bibliographic information based on title information or perspective information of the multiple bibliographic information that has been prepared in advance for each database. Specifically, the server 14 prepares an overview of what kind of bibliographic information is stored in the database and a collection of titles of the bibliographic information stored in the database for each referenced database, and causes the vector search or large-scale language model to determine which database to refer to. In this case, for example, the server 14 may determine the referenced database for each decomposition result information or proposal information. Also, for example, the server 14 may determine the referenced database based on decomposition criterion information (for example, because when the user selects "structural decomposition by element", it is not necessary to refer to bibliographic information on "manufacturing process").

[0134] [Variation 7] Also, if there are multiple databases as in Variation 6, access permissions may be set for each of the databases. Since a user using this system may not necessarily have access permissions to the database being referenced, the server 14 checks the user's access permissions. For example, chemical companies always have confidential parameters that have been calculated or obtained through experiments but have not been announced to the public, and access to these is restricted to only a limited number of people. Therefore, when a user logs into this system, only databases that the user has access to are designated as reference databases. Specifically in this case, when the server 14 retrieves specific literature information from among multiple literature sources, it may determine whether the user has access permissions to the target database, and if the user has access permissions to the target database, it may retrieve the specific literature information from that database. In addition, when determining the reference database, the server 14 may also check the user's access permissions at the same time, and if it is determined that it would be better to refer to the reference database but the user does not have access permissions, it may present a message saying, "The information is available within the company, but you do not have access permissions." Furthermore, in this case, to reduce the possibility of access restrictions being bypassed by prompt hacking by a third party, it is possible to separate the large-scale language model that determines the referenced database from the large-scale language model that determines whether the referenced database is accessible.

[0135] [Modification 8] When the server 14 retrieves specific bibliographic information from among multiple bibliographic records, it may determine which bibliographic information to retrieve based on the number of times the bibliographic information retrieved for reference has been referenced. In this case, the server 14 may record the number of times each bibliographic record has been referenced and use the number of references to display a ranking or to calculate the reliability. It should be noted that there may be business models in which fees are paid to the bibliographic providers (especially academic journals) based on the number of references, so the server 14 should keep track of the number of references.

[0136] [Modification 9] The server 14 may identify the department information related to the proposal represented by the proposal information based on the organizational chart information of the organization to which the user belongs and the list of research themes for each department that makes up the organization, and output the department information. In this case, the server storage unit 140 of the server 14 stores the organizational chart information and the list of research themes. Specifically, the server 14 determines the organization related to each proposal based on the organizational chart and the list of research themes for each organization that have been acquired in advance, and presents it to the user.

[0137] [Modification 10] In addition to information on the computational model relating to materials development, the server 14 may also output computational model execution information necessary to execute the computational model. Computational model execution information includes, for example, information on the program for executing the computational model and the computational resources necessary to execute the computational model. Computational resource information includes, for example, whether or not computation by a quantum computer is required, the amount of computation and computation time required by the quantum computer or classical computer for the computation, and the cost required for the computation. In this case, based on the computational model information generated by the large-scale language model, the server 14 generates a program for calculating the computational model or calculates the computational resources required to calculate the computational model. Furthermore, for example, if a user attempts to perform an actual calculation based on the above information, the server 14 may obtain the cost according to the actual calculation process. Note that the generation of computational model execution information may be performed by the large-scale language model instead of the server 14. For example, the large-scale language model may be made to generate a program for executing the computational model by inputting the computational model and the rules for creating an execution program for performing chemical calculations. Alternatively, the large-scale language model may be made to generate information on the computational resources necessary for the calculation by inputting the computational model and information on the performance and computation costs of a predetermined quantum computer or classical computer.

[0138] Furthermore, in each of the above embodiments, the server 14 was described as acquiring viewpoint information selected by the user operating the user terminal from among multiple viewpoint information, but it is not limited to this. Also, in the example, the server 14 was described as acquiring decomposition result information selected by the user from among multiple decomposition result information, but it is not limited to this. Each of the above selections may be made (or automatically set) by AI (for example, an AI agent) rather than by the user. Furthermore, the selection of viewpoint information may also be a selection of purpose, and the configuration may also allow the user to select viewpoint items that direct the output, such as whether it is new development, exploration of new areas of the company's technology, or competitive analysis. For this reason, purpose selection can also be equivalent to the selection of viewpoint information. In this case, the server 14 may perform processing such as outputting items with high company relevance and high competitor relevance if it is a competitive analysis.

[0139] Furthermore, in the above embodiments, the term "processor" refers to a broad type of processor, including general-purpose processors (e.g., CPUs) and dedicated processors (e.g., GPUs: Graphics Processing Units, ASICs: Application Specific Integrated Circuits, FPGAs: Field Programmable Gate Arrays, programmable logic devices, etc.).

[0140] Furthermore, the operation of the processor in the above embodiments may not be performed by a single processor, but may be performed by multiple processors located in physically separate locations working together. Also, the order of the processor's operations is not limited to the order described in each of the above embodiments, but may be changed as appropriate.

[0141] Furthermore, although the "system" in this embodiment is described as being composed of multiple devices as an example, it may also be composed of a single device that has some of the functions of multiple devices.

[0142] Furthermore, the processing performed by each device according to the above embodiment may be software-based processing, hardware-based processing, or a combination of both. Additionally, the processing performed by each device may be stored as a program on a storage medium and distributed.

[0143] The program of this application can be provided as a program product. A program product includes all forms of products for providing a program. For example, a program product includes a program provided via a network such as the Internet, and non-temporary computer-readable recording media such as CD-ROMs and DVDs on which the program is stored.

[0144] All documents, patent applications, and technical standards described herein are incorporated by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted to be incorporated by reference.

[0145] (Note) The following is a note regarding the nature of this disclosure.

[0146] (Note 1) An information providing device that acquires problem information, which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmits to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints that represent a standard viewpoint for decomposing a problem used to examine the material development problem represented by the problem information; acquires the viewpoint information selected by the user operating the user terminal from among the plurality of viewpoint information; generates a prompt to generate decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint, based on the selected viewpoint information; inputs the generated prompt into a large-scale language model to acquire a plurality of decomposition result information output from the large-scale language model; acquires literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputs the acquired literature information. (Note 2) The information providing device described in Note 1, wherein the prompt is a first prompt, a second prompt is generated to generate explanatory information representing an explanation of the decomposition result information based on the literature information, the generated second prompt is input to a large-scale language model to obtain the explanatory information output from the large-scale language model, and the obtained literature information and the explanatory information are output.(Note 3) The information providing device described in Note 2, wherein the aforementioned document information is first document information, the aforementioned explanatory information is first explanatory information, and when acquiring the first document information related to the decomposition result information, a plurality of decomposition result information is transmitted to the user terminal, the decomposition result information selected by the user from among the plurality of decomposition result information is acquired, a third prompt is generated for generating proposal information for considering an experiment or chemical calculation using the method represented by the decomposition result information based on the first document information related to the selected decomposition result information, the generated third prompt is input to the large-scale language model to acquire the proposal information output from the large-scale language model, a second document information (second reference) related to the proposal information is acquired from a plurality of documents (by RAG) based on the acquired proposal information, a fourth prompt is generated for generating second explanatory information representing an explanation of the proposal information based on the second document information, the generated fourth prompt is input to the large-scale language model to acquire the second explanatory information of the proposal information output from the large-scale language model, and the acquired proposal information and second explanatory information are output. (Note 4) The information providing device according to Note 1 or Note 2, wherein the second prompt includes bibliographic information of the literature information. (Note 5) The information providing device according to any one of Notes 1 to 4, wherein the perspective information includes at least one of the structure of the material and the manufacturing process of the material. (Note 6) The information providing device according to any one of Notes 1 to 5, wherein the literature information is academic literature or internal company documents. (Note 7) The information providing device according to Note 3, which calculates the reliability of the proposal represented by the proposal information based on rules set in advance according to the attributes of the second literature information related to the proposal information, and outputs the reliability of the proposal. (Note 8) The information providing device according to Note 3, which obtains the reliability of the proposal represented by the proposal information by inputting a reliability prompt, which is a prompt for generating the reliability of the proposal represented by the proposal information, to the large-scale language model, and outputs the reliability of the proposal.(Note 9) An information providing device according to any one of Notes 1 to 8, which generates a research prompt, which is a prompt for obtaining a computational model or research procedure for solving the material development problem represented by the problem information, based on the problem information, and inputs the generated research prompt into the large-scale language model to obtain the computational model or research procedure output from the large-scale language model. (Note 10) An information providing device according to any one of Notes 1 to 9, which has multiple databases containing multiple bibliographic information, and when obtaining specific bibliographic information from the multiple bibliographic information, it obtains specific bibliographic information from the multiple bibliographic information based on title information or perspective information of the multiple bibliographic information prepared in advance for each database. (Note 11) An information providing device according to any one of Notes 1 to 10, which has access rights set for multiple databases, and when obtaining specific bibliographic information from the multiple bibliographic information, it determines whether the user has access rights to the target database, and if the user has access rights to the target database, it obtains specific bibliographic information from the database. (Note 12) When obtaining specific bibliographic information from the multiple bibliographic information, the information to be obtained is determined based on the number of times the bibliographic information obtained for reference has been referenced, as described in any one of Notes 1 to 11. (Note 13) Based on the organizational chart information of the organization to which the user belongs and the list of research themes for each department that constitutes the organization, the information providing device identifies the department information related to the proposal represented by the proposal information and outputs the department information, as described in Note 3. (Note 14) The information providing device according to Note 9 outputs the computation model from the large-scale language model and executes the generation of a program for calculating the computation model or the calculation of the computational resources of a quantum computer for calculating the computation model. (Note 15) The information providing device according to Note 1 generates prompts to be input to the large-scale language model based on pre-set rules.(Note 16) The information providing device described in Note 1, wherein each of the multiple pieces of literature information is proprietary technical information representing technical information owned by the company to which the user belongs, and based on the acquired disassembly result information, the device acquires proprietary technical information related to the disassembly result information from among the multiple pieces of proprietary technical information and outputs the acquired proprietary technical information. (Note 17) The information providing device described in Note 1, wherein each of the multiple pieces of literature information is proprietary technical information representing technical information owned by the company to which the user belongs, and based on the acquired disassembly result information, the device acquires the degree of relevance between the disassembly result information and the proprietary technical information for each combination between each of the multiple pieces of disassembly result information and each of the multiple pieces of proprietary technical information, and outputs the acquired proprietary technical information and the degree of relevance. (Note 18) The information providing device described in Note 1, wherein each of the multiple pieces of literature information is proprietary technical information representing technical information owned by the company to which the user belongs, and based on the multiple pieces of decomposition result information obtained, the degree of relevance between the decomposition result information and the proprietary technical information is obtained for each combination between each of the multiple pieces of decomposition result information and each of the multiple pieces of proprietary technical information, and based on the multiple degrees of relevance, target technical information representing the technology necessary to solve the problem and to be combined with the proprietary technology is identified, and the obtained proprietary technical information and the target technical information are output. (Note 19) The information providing device described in Note 18, wherein the target technical information is technical information different from the proprietary technical information and is external technical information existing outside the company. (Note 20) The information providing device described in Note 18, wherein the target technical information is proprietary technical information different from the proprietary technical information to be combined.(Note 21) An information provision method in which a computer performs the following processes: acquires problem information which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmits to the user terminal a plurality of viewpoint information which are pre-prepared viewpoints that represent a standard viewpoint for decomposing a problem used to examine the material development problem represented by the problem information; acquires the viewpoint information selected by the user operating the user terminal from among the plurality of viewpoint information; generates a prompt to generate decomposition result information which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint based on the selected viewpoint; inputs the generated prompt into a large-scale language model to acquire a plurality of decomposition result information output from the large-scale language model; acquires literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputs the acquired literature information. (Note 21) An information provision program for causing a computer to execute a process that: acquires problem information which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmits to the user terminal a plurality of viewpoint information which are pre-prepared viewpoints that represent a standard viewpoint for decomposing the problem used to examine the material development problem represented by the problem information; acquires the viewpoint information selected by the user operating the user terminal from among the plurality of viewpoint information; generates a prompt to generate decomposition result information which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint based on the selected viewpoint; inputs the generated prompt into a large-scale language model to acquire a plurality of decomposition result information output from the large-scale language model; acquires literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputs the acquired literature information.

[0147] The disclosure of Japanese Patent Application No. 2024-171155, filed on 30 September 2024, is incorporated herein by reference in its entirety. All documents, patent applications, and technical standards described herein are incorporated herein by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted to be incorporated by reference.

Claims

1. An information providing device that acquires problem information, which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmits to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints that serve as criteria for decomposing the problem used to examine the material development problem represented by the problem information; acquires selected viewpoint information from the plurality of viewpoint information; inputs a prompt generated based on the selected viewpoint information to a large-scale language model, which is a prompt for generating decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint, thereby acquiring a plurality of decomposition result information output from the large-scale language model; acquires literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputs the acquired literature information.

2. The information providing device according to claim 1, wherein the prompt is a first prompt, a second prompt is generated for generating explanatory information representing an explanation of the decomposition result information based on the literature information, the generated second prompt is input to a large-scale language model to obtain the explanatory information output from the large-scale language model, and the obtained literature information and the explanatory information are output.

3. The information providing device according to claim 2, wherein the literature information is first literature information, the explanatory information is first explanatory information, and when acquiring the first literature information related to the decomposition result information, a plurality of decomposition result information is transmitted to the user terminal, the selected decomposition result information is acquired from the plurality of decomposition result information, a third prompt is generated for generating proposal information for considering an experiment or chemical calculation using the method represented by the decomposition result information based on the first literature information related to the selected decomposition result information, the generated third prompt is input to the large-scale language model to acquire the proposal information output from the large-scale language model, second literature information related to the proposal information is acquired from a plurality of literature based on the acquired proposal information, a fourth prompt is generated for generating second explanatory information representing an explanation of the proposal information based on the second literature information, the generated fourth prompt is input to the large-scale language model to acquire second explanatory information of the proposal information output from the large-scale language model, and the acquired proposal information and second explanatory information are output.

4. The information providing device according to claim 2, wherein the second prompt includes bibliographic information of the document information.

5. The information providing device according to claim 1 or 2, wherein the viewpoint information includes at least one of the structure of the material and the manufacturing process of the material.

6. The information providing device according to claim 1 or 2, wherein the aforementioned literature information includes at least one of academic literature and internal company documents.

7. An information providing device according to claim 3, which calculates the reliability of the proposal represented by the proposal information and outputs the reliability of the proposal based on rules set in advance according to the attributes of the second document information related to the proposal information.

8. The information providing device according to claim 3, which inputs a confidence prompt, which is a prompt for generating the confidence level of the proposal represented by the proposal information, to the large-scale language model, thereby obtaining the confidence level of the proposal represented by the proposal information and outputting the confidence level of the proposal.

9. An information providing device according to claim 1 or 2, which generates a research prompt, which is a prompt for obtaining a computational model or research procedure for solving a material development problem represented by the problem information, based on the problem information, and inputs the generated research prompt into the large-scale language model to obtain the computational model or research procedure output from the large-scale language model.

10. An information providing device according to claim 1 or 2, wherein multiple databases store multiple bibliographic information, and when retrieving specific bibliographic information from among the multiple bibliographic information, the device retrieves the specific bibliographic information from among the multiple bibliographic information based on title information or perspective information of the multiple bibliographic information that is prepared in advance for each database.

11. An information providing device according to claim 1 or 2, wherein access permissions are set for multiple databases, and when obtaining specific bibliographic information from the multiple bibliographic information, the device determines whether the user has access permissions to the target database, and if the user has access permissions to the target database, it obtains the specific bibliographic information from the database.

12. When obtaining specific bibliographic information from multiple bibliographic sources, the information to be obtained is determined based on the number of times the bibliographic information obtained for reference has been referenced, as described in claim 1 or claim 2.

13. An information providing device according to claim 3, which identifies department information related to the proposal represented by the proposal information based on organizational chart information of the organization to which the user belongs and a list of research themes for each department that constitutes the organization, and outputs the department information.

14. The information providing device according to claim 9, wherein the large-scale language model outputs the computation model, and the device generates a program for computing the computation model or calculates the computational resources of a quantum computer for computing the computation model.

15. The information providing device according to claim 1, wherein the prompts to be input to the large-scale language model are generated based on pre-set rules.

16. The information providing device according to claim 1, wherein each of the plurality of document information is proprietary technical information representing technical information owned by the company to which the user belongs, and based on the acquired disassembly result information, the device acquires proprietary technical information related to the disassembly result information from among the plurality of proprietary technical information and outputs the acquired proprietary technical information.

17. The information providing device according to claim 1, wherein each of the plurality of document information is proprietary technical information representing technical information owned by the company to which the user belongs, and based on the plurality of decomposition result information obtained, the degree of relevance between the decomposition result information and the proprietary technical information is obtained for each combination between each of the plurality of decomposition result information and each of the plurality of proprietary technical information, and the obtained proprietary technical information and the degree of relevance are output.

18. The information providing device according to claim 1, wherein each of the plurality of literature information is proprietary technical information representing technical information owned by the company to which the user belongs, and based on the obtained plurality of decomposition result information, the degree of relevance between the decomposition result information and the proprietary technical information is obtained for each combination between each of the plurality of decomposition result information and each of the plurality of proprietary technical information, and based on the plurality of degrees of relevance, target technical information representing the technology necessary to solve the problem and to be combined with the proprietary technology is identified, and the obtained proprietary technical information and the target technical information are output.

19. The information providing device according to claim 18, wherein the target technical information is technical information different from the company's own technical information and is external technical information existing outside the company.

20. The information providing device according to claim 18, wherein the target technical information is different from the company's own technical information that is to be combined.

21. An information provision method in which a computer performs the following processes: acquires problem information, which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmits to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints that represent a standard viewpoint for decomposing the problem used to examine the material development problem represented by the problem information; acquires selected viewpoint information from the plurality of viewpoint information; inputs a prompt generated based on the selected viewpoint information to a large-scale language model, which is a prompt that generates decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint, thereby acquiring a plurality of decomposition result information output from the large-scale language model; acquires literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputs the acquired literature information.

22. An information provision program for causing a computer to execute a process that involves: acquiring problem information, which is information relating to a material development problem transmitted from a user terminal and expressed in natural language; transmitting to the user terminal a plurality of viewpoint information, which are pre-prepared viewpoints representing a standard viewpoint for decomposing the problem used to examine the material development problem represented by the problem information; acquiring selected viewpoint information from the plurality of viewpoint information; inputting a prompt generated based on the selected viewpoint information to a large-scale language model, which is a prompt that generates decomposition result information, which is the result of decomposing the material development problem represented by the problem information into elements from the selected viewpoint; acquiring a plurality of decomposition result information output from the large-scale language model; acquiring literature information related to the decomposition result information from a plurality of literature information based on the acquired decomposition result information; and outputting the acquired literature information.

Images