Information processing method, program, and information processing device

Abstract

The purpose of the present invention is effective assistance of document creation.　Provided is an information processing method for assisting document creation, the information processing method including causing at least one computer to execute prompt creation processing, generation processing, and document creation processing. In the prompt creation processing, a prompt is created on the basis of user input information, which is based on input by a user, and a prompt template. In the generation processing, a generative AI model is caused to generate generated data based on the prompt. In the document creation processing, document data is created on the basis of the generated data.

Description

Information processing method, program, and information processing apparatus

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

[0002] Patent Document 1 discloses a technology that assists in document creation.

[0003] Japanese Patent Publication No. 2009-080645

[0004] The inventors of this application have discovered a novel support technology that streamlines document creation by incorporating the results of recent technological innovations in the field of artificial intelligence.

[0005] This invention aims to provide effective support for document creation.

[0006] According to the present invention, an information processing method for supporting document creation is provided, wherein a prompt creation process, a generation process, and a document creation process are performed on at least one computer, the prompt creation process creates a prompt based on user input information based on user input and a prompt template, the generation process causes a generation AI model to generate generation data based on the prompt, and the document creation process creates document data based on the generation data.

[0007] According to the present invention, it is possible to effectively support document creation by utilizing a generative AI model.

[0008] Figure 1 illustrates an information processing device 1 and a related server according to the embodiment. Figure 2A illustrates a hardware configuration diagram of the information processing device 1, and Figure 2B illustrates a functional block diagram of the control unit 12. Figure 3 is a flowchart for explaining an example of an information processing method according to the embodiment. Figure 4 is an example of an input UI screen 130 according to the embodiment. Figure 5A is an example of a preview UI screen 140 (display language: Japanese), and Figure 5B is an example of record M006 (heading string Hd Ｍ００６This figure illustrates the content to be described according to the Results. Figure 6A is an example of the editing UI screen 150, and Figure 6B shows an example after editing work has been performed. Figure 7 shows an example of the popup UI screen 160 presented by the evaluation process. Figure 8 is a flowchart to explain a modified example of the embodiment.

[0009] One or more embodiments of the present invention will be described below with reference to the drawings. The various features shown in the embodiments below can be combined with each other. Furthermore, each feature constitutes an independent invention.

[0010] For example, one of the simple embodiments provided by this disclosure is a "document creation support technology using a generative AI model." More specifically, it is a "document creation support technology that prompts a generative AI model to generate data and automatically creates document data by combining the generated data." Any one of the various features of the embodiment may be combined with this simple document creation support technology, or any combination of multiple various features may be combined. This technology can be provided in the form of an information processing method, a computer program, an information processing device, or an information processing system. In one embodiment, as an example, it supports the creation of papers (especially academic papers).

[0011] In the following explanation, the terms "and / or" have the following meanings. For example, using elements α and β, "α and / or β" encompasses "α only from α and β", "β only from α and β", and "α and β". It should be noted that, in principle, the addition of other elements (e.g., γ) is not excluded. Unless explicitly prohibited or impossible to combine, "α and / or β" can be arbitrarily combined with other elements such as γ via "and" or "or".

[0012] In the following explanation, the term "primarily" can be used to express either a quality or a proportion. When "primarily" expresses a quality, for example, "main part" means the main part, the important part, or the part that influences the overall trend. On the other hand, when "primarily" expresses a proportion, "primarily" means "majority." The majority may be a majority of the whole (more than 50%), or it may be the group with the largest proportion within the whole (e.g., the group of 40% of 40%, 20%, 20%, and 20%). In embodiments, the proportion that "primarily" means may be defined, for example, within the following numerical ranges. Specifically, for example, the proportion that "primarily" means may be 50% or more, 60% or more, or 70% or more. The percentage represented by "mainly" may be, for example, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95% or more, or it may be 100%, or it may be within any two ranges of the values ​​listed here (e.g., 40-90%). The above properties or percentages of "mainly" may be used, for example, to represent the "percentage of specific characters (e.g., English letters)" in a string, or the "percentage of specific words (e.g., English words)" in text or content.

[0013] 1. Description of the Configuration of the Embodiment 1-1. System Overall Configuration Diagram 1 illustrates an information processing system 100 of the embodiment. The information processing system 100 includes, as an example, an information processing device 1 of the embodiment and related servers. In the embodiment, as an example, the related servers include at least one generation AI model server 2 and at least one document publishing system server 3. The related servers may also include at least one of the following: an arbitrary translation server, an arbitrary analysis application server, and an arbitrary RAG (Retrievable-Augmented Generation) system server. In the embodiment, for the sake of simplicity, an example is given of a user U operating the information processing device 1. However, this is only an example, and for example, the information processing device 1 may be provided as a server, or user U may access the information processing device 1 via a communication network 6 using their own mobile terminal, or the information processing method of the embodiment may be provided as a cloud web service.

[0014] In Figure 1, as an example, multiple generative AI model servers 2 are connected to the information processing device 1 via a communication network 6. Each generative AI model server 2 has, for example, a neural network trained using a large amount of training data. Each generative AI model server 2 has the function of generating a response to data (prompt) output from the information processing device 1 and outputting it to the information processing device 1. The generative AI model server 2 is not particularly limited, but any model, such as a so-called large-scale language model (LLM), can be used. In this embodiment, as an example, multiple generative AI model servers 2 can be used interchangeably or in combination.

[0015] The document publishing system server 3 shown in Figure 1 is connected to the information processing device 1 via a communication network 6. The document publishing system server 3 can employ various known document publishing systems. For example, a research paper publishing system may also be used.

[0016] 1-2. As shown in the hardware configuration diagram 2A of the information processing device, the information processing device 1 includes, as an example, a communication unit 10, a storage unit 11, a control unit 12, an output unit 13, and an input unit 14, and these components are electrically connected within the information processing device 1 via a communication bus 15. The information processing device 1 is also a "computer" that performs information processing in the embodiment.

[0017] The communication unit 10 may employ wired communication methods such as USB, IEEE 1394, Thunderbolt®, or wired LAN network communication. The communication unit 10 may also be configured to connect to the communication network 6 via wireless communication methods such as wireless LAN network communication, 3G / LTE / 5G mobile communication, or Bluetooth® communication. Furthermore, the communication unit 10 may be configured to use both wired and wireless communication methods in combination.

[0018] The storage unit 11 stores various values, such as various programs, constants, variables, and settings of the information processing device 1, which is executed by the control unit 12, as an example. The storage unit 11 can employ a storage device such as a solid-state drive (SSD) or a random access memory (RAM) that stores temporarily necessary information (arguments, arrays, etc.) related to program calculations. In addition to the storage unit 11, the information processing device 1 may also use an external storage unit (for example, an external storage medium, cloud storage, etc.).

[0019] The control unit 12 is configured to perform processing and control related to the information processing of the information processing device 1. The control unit 12 can be configured as, for example, a central processing unit (CPU), and in this embodiment, the control unit 12 is an example of a processor capable of executing programs related to each step of the flowchart described later. The control unit 12 realizes various functions related to the information processing device 1 by, for example, reading programs stored in the storage unit 11. Furthermore, the information processing of the software in the information processing device 1 is realized by, for example, the processing of various programs stored in the storage unit 11 by the control unit 12 as hardware.

[0020] The output unit 13 is, for example, the display unit of the information processing device 1. The output unit 13 may be included in the housing of the information processing device 1, or it may be externally mounted. The output unit 13 displays a graphical user interface (GUI) screen that can be operated by the user. The output unit 13 may employ display devices such as a CRT display, liquid crystal display, organic EL display, plasma display, electronic paper display, head-mounted display, or smart glasses, as well as a lit-up light or projector. It is optional whether or not the information processing device 1 includes an output unit 13. For example, as a modification, the output of the information processing device 1 may be displayed on a display unit located in a separate location independent of where the information processing device 1 is installed. The output unit 13 may also have a device that outputs sound.

[0021] The input unit 14 is configured, for example, to receive operation inputs made by a user of the information processing device 1. The input unit 14 may be included in the housing of the information processing device 1 or it may be an external component. The input unit 14 can be, for example, a touch panel, switch buttons, a mouse, a keyboard, etc. Whether or not the information processing device 1 includes an input unit 14 is optional. For example, as a modification, operation inputs to the information processing device 1 may be received by the information processing device 1 via an information processing terminal located in a separate location independent of where the information processing device 1 is installed.

[0022] 1-3. The functional configuration of the information processing device 1 according to this embodiment will be described with reference to the functional block diagram 2B of the information processing device. The control unit 12 of the information processing device 1 includes, as an example, a session management unit 12a, an information acquisition unit 12b, a result data acquisition unit 12b1, a translation unit 12c, a prompt creation unit 12d, a reference information acquisition unit 12e, a generation unit 12f, a pre-acquisition unit 12g, an operation reception unit 12h, a document creation unit 12j, an evaluation unit 12k, and a posting unit 12m. In order to provide each of these functional units, information processing by software stored in the storage unit 11 is specifically realized by the control unit 12, which is an example of hardware. When various data are sent and received between the information processing device 1 and the generation AI model server 2, an API (Application Programming Interface) is intervened. The various data sent and received include, for example, prompt P ｎ , Reference information Ref1 or generated data G ｎ Includes, etc.

[0023] Each of the above functional units can be provided by any means, for example, by software or by hardware. When implemented by software, various functions can be realized by the CPU executing a computer program. The program may be stored on a non-temporary recording medium that the computer can read, provided for download from an external server, or implemented by so-called cloud computing, which reads a program stored in an external memory unit to realize the function. When implemented by hardware, it can be implemented by various circuits such as ASICs, FPGAs, or DRPs. In the embodiment, various information and concepts encompassing it are handled, which are represented by high or low signal values ​​or qubits as a set of binary bits composed of 0s or 1s, and communication and calculations can be performed by the above-mentioned software or hardware configurations. The software may be a general-purpose OS or a dedicated OS.

[0024] Each functional unit in Figure 2B is configured to execute each process (S1 to S11) in the flowchart of Figure 3, which will be described later. The session management unit 12a is configured to execute session management process S1 (see Figure 3). The information acquisition unit 12b is configured to execute information acquisition process S2 (see Figure 3), and user input information IN ｎ It is configured to be obtainable.

[0025] The result data acquisition unit 12b1 is configured to execute the result data acquisition process S2a (see Figure 3). The result data acquisition unit 12b1 incorporates, for example, an analysis application 12c1. The analysis application 12c1 receives input data INA, performs analysis processing internally, and outputs result data R1. As an alternative, the analysis application 12c1 may be provided on an external server (such as a cloud web service), in which case it is configured to be accessible from the information processing device 1 via the communication network 6.

[0026] The translation unit 12c is configured to execute translation processing SS1 (see Figure 3). The translation unit 12c is configured to perform translation between any multiple different natural languages ​​(e.g., English and Japanese). The translation unit 12c is configured to be callable by other functional units as needed, for example. In this embodiment, for example, the translation unit 12c causes the generation AI model server 2 to perform the translation. For example, the translation unit 12c can execute the translation process by providing the generation AI model server 2 with translation instruction prompts and the text to be translated, for example.

[0027] The prompt creation unit 12d is configured to execute the prompt creation process S3 (see Figure 3), and prompt P ｎ It is possible to create this. The reference information acquisition unit 12e is configured to execute the reference information acquisition process S4 (see Figure 3) and to acquire the reference information Ref1.

[0028] The generation unit 12f is configured to execute generation process S5 (see Figure 3). The pre-import unit 12g is configured to execute pre-import process S5a (see Figure 3). The generation unit 12f and the pre-import unit 12g are configured to access the generation AI model server 2. The generation AI model server 2 receives prompt P ｎ And receiving the reference information Ref1, the generated data G is generated accordingly. ｎ Answer the question.

[0029] The document creation unit 12j is configured to execute the document creation process S8 (see Figure 3). The operation reception unit 12h is configured to execute the operation reception process S9 (see Figure 3). The evaluation unit 12k is configured to execute the evaluation process S10 (see Figure 3). The posting unit 12m is configured to execute the posting process S11 (see Figure 3) and can send the document data D1 to the document publication system server 3.

[0030] A more detailed explanation of the processing of each functional unit will be provided in the flowchart shown in Figure 3, which will be described later.

[0031] 2. Description of the Operation and Processing Content of the Embodiment 2-1. Prerequisites The operation of the embodiment (i.e., the information processing content) will be explained using Figures 3 to 7. Figure 3 is a flowchart for explaining an example of the information processing method of the embodiment. Figure 3 specifies the specific processing steps of the embodiment. The program according to the embodiment can execute the information processing method of the embodiment on the computer (information processing device 1) by executing each process (information processing method) in Figure 3.

[0032] 2-1-1. UI Screen Figure 4 is an example of the input UI screen 130 of the embodiment. The input UI screen 130 is displayed on the display unit of the output unit 13. The input UI screen 130 includes common items, keyword input items, reference information items, and generation AI model selection items. The common items include a user-specified language menu 131a, a new session button 131b, an input field increase button 133, an item transition button 134, and a generation start button 139. The keyword input items include a keyword type selection menu 132a and a keyword input area 132b. The data import items include an input area 135a for the import data reference destination, a detailed input button 135b for the import data, an analysis setting checkbox 136a, and an analysis application selection button 136b. The reference information items include a reference information reference destination input area 137a and a pre-import checkbox 137b. The AI ​​model generation selection items include the AI ​​model generation selection menu 138a and the usage condition setting button 138b. The input UI screen 130 is an example, and as another example, item-specific UI screens for each of the above items may be provided, and screen transitions may occur in any order.

[0033] User U enters various information into the UI screen 130 and then clicks the "Start Generation" button 139. This triggers the execution of each processing step (S2 onwards) related to document creation within the flowchart shown in Figure 3.

[0034] Figures 5A, 5B, 6A, 6B, and 7 are examples of UI screens in each process described later, and are displayed on the display unit of the output unit 13.

[0035] 2-1-2. Table TB In the embodiment, as an example, as shown in FIG. 2A, the table TB is stored in the storage unit 11. In the embodiment, as an example, a record ID (M001 to M011...) is assigned to each record of the table TB. In the embodiment, the record ID is associated with the heading of the document data D1. The record ID (M001 to M011...) is also referred to as the "heading ID". In the embodiment, the table TB, as an example, includes at least the following columns (Hd ｎ 、IN ｎ 、PrT ｎ 、P ｎ 、G ｎ ).

[0036] Heading string: Hd ｎ (Hd Ｍ００１ 、Hd Ｍ００２ …) User input information: IN ｎ (IN Ｍ００１ 、IN Ｍ００２ …) Prompt template: PrT ｎ (PrT Ｍ００１ 、PrT Ｍ００２ …) Prompt: P ｎ (P Ｍ００１ 、P Ｍ００２ …) Generated data: G ｎ (G Ｍ００１ 、G Ｍ００２ …)

[0037] The table TB is used in internal processes such as the prompt creation process S3 and the generation process S5 described later. In each column (Hd ｎ 、IN ｎ 、PrT ｎ 、P ｎ 、G ｎ ), the subscript n functions as an identifier, and each field data can be referred to by specifying an arbitrary record ID.

[0038] The heading string Hd ｎ stores each heading of the document data D1 (papers in the embodiment). In the embodiment, as an example, the stored data is in the base language (e.g., English). For example, in the case of an academic paper, the heading string fields Hd Ｍ００１ 、Hd Ｍ００２HD Ｍ００３ HD Ｍ００４ ...Hd Ｍ００１１ Each of the data items may be, in order, "Title", "Author Names", "Abstract", "Introduction", "Experiment", "Results", "Discussion", "Conclusion", "Acknowledgements", "References", and "Appendices". For example, heading string Hd Ｍ００４ This will be the "Introduction".

[0039] 2-1-3. Base Language of the Language Model As described above, in the embodiment, the Generative AI Model Server 2 includes a language model as an example. In the embodiment, the "base language" of the language model is utilized as an example. This base language is a natural language to which the language model has adapted through learning. In the embodiment, the base language of the Generative AI Model Server 2 is English as an example. In the embodiment, various measures are taken to utilize the above-mentioned "base language" as an example. For example, in the embodiment, the column Hd of table TB ｎ , PrT ｎ , P ｎ G ｎ The string data within is primarily written in the base language.

[0040] Further details about the base language will be provided in the section on modifications (third modification) below, so please refer to that as needed. In this embodiment, a large-scale language model is used as an example, but a small-scale language model may also be used. There may be multiple tables TB (tables TB1, TB2, etc.), in which case table TB1 may be the base language (e.g., English), and table TB2 may mainly contain the same content as table TB1 but in a non-base language (e.g., Japanese).

[0041] 2-1-4. Prompt Templates A "prompt template" is a standardized structure for giving instructions to a generating AI model. A prompt template has a "variable section." Any information can be inserted into this variable section. Specific words, phrases, or sentences can be inserted into the variable section. Prompts can be created by embedding specific words or phrases (variable section).

[0042] In this embodiment, the "prompt template" is, for example, a predetermined string of arbitrary length. Prompt template PrT ｎ It is sufficient that it can generate prompts by combining it with user input information, and there are no particular limitations. For example, prompt template PrT ｎ This may be one or more phrases, one or more sentences, one or more paragraphs, or text. In this embodiment, as an example, the prompt template PrT ｎ However, the data is pre-stored in the memory unit 11 and read out each time by rule-based processing from table TB as needed.

[0043] Each prompt template PrT of the embodiment ｎ For example, the "Base Language Prompt Template PrT" is created in advance in the base language. ｎ In a base language prompt template, the prompt template string is primarily written in the base language. For example, all or at least the main part of the string may be written in the base language, or the proportion of characters or words in the base language may be the majority. The main part is, for example, a command statement to the generative AI model, and / or a condition statement that directly affects the accuracy of the generative AI model's response, etc.

[0044] In this embodiment, as an example, each heading string Hd of document data D1 ｎ And each column IN ｎ , PrT ｎ , P ｎ G ｎEach field is associated with the structure of table TB.

[0045] 2-2. Explanation of Each Processing Step Next, an example of the specific content of each processing step in Figure 3 will be explained. When the processing in Figure 3 is started, the UI screen 130 is displayed first, and the system enters a waiting state.

[0046] In the UI screen 130 of Figure 4, user U can select any user-specified language from the languages ​​pulled down by the user-specified language menu 131a by operating the input unit 14. In this embodiment, as an example, the user-specified language is set to Japanese. This allows for, for example, the display language on each UI screen and / or user input information IN ｎ The language will be set to Japanese.

[0047] 2-2-1. Session Management Process S1 In the process shown in Figure 3, first, as an example of the embodiment, the session management process S1 is executed by the information processing device 1. In the session management process S1, the information processing device 1 executes the following processes (d1) to (d3) as an example.

[0048] (d1) The information processing device 1 performs session switching when predetermined conditions are met. Various conditions can be used as predetermined conditions, but one example is when the new session button 131b (see Figure 4) is clicked. As another condition, a user account may be provided for each user, and session switching may be performed automatically when the user account is switched (including when logging out, etc.).

[0049] (d2) In response to the session switching, the information processing device 1 displays the user input information IN for the current session. ｎ , prompt P ｎ and generated data G ｎ Delete one or more of them. In this embodiment, as an example, delete all of them. After deletion, start the next session. This has the advantage of preventing information mixing and ensuring accuracy, as well as ensuring information confidentiality, etc.

[0050] (d3) Within one session, user input information INｎ , prompt P ｎ or generated data G ｎ The data is retained (for example, stored in arbitrary memory and reused). As an example, in the prompt creation process S3, generation process S5, and document creation process S8 described later, user input information IN ｎ , prompt P ｎ or generated data G ｎ It becomes reusable.

[0051] If the new session button 131b (see Figure 4) is not clicked, the session management process S1 does not need to be executed. For example, if the same user logs in again without a session switch, the previous session and the current session may be treated as a series of sessions, or they may be considered as the same session.

[0052] 2-2-2. Information Acquisition Process S2 Next, in this embodiment, as an example, the information acquisition process S2 is executed by the information processing device 1. In the information acquisition process S2, the information processing device 1 receives "User Input Information IN ｎ The user U inputs user input information by operating the input unit 14. ｎ It is possible to input data. The input information can be stored in the storage unit 11 of the information processing device 1.

[0053] In one embodiment, as an example, user input information IN ｎ This includes "keywords" and "imported data." First, let's discuss "keywords." User U selects the keyword type to input by operating the keyword type selection menu 132a (in the example in Figure 4, "Research Objective"), and then enters the keyword in the keyword input area 132b. Here, "keyword" simply means "a string of characters to convey the main points," and there is no limit to the length of the string. The keyword entered in the keyword input area 132b can be any word, phrase, or sentence.

[0054] Next, we will discuss "imported data." In information import processing S2, the information processing device 1 takes "chart data" as an example and inputs it as user input information IN. ｎIt will be imported as such. Here, "figures and tables" can be figures, tables (including data lists), graphs, line drawings, or photographs. There are no restrictions on the data format of the figure and table data.

[0055] 2-2-2-1. Translation Process SS1 In this embodiment, as an example, the translation process SS1 is executed by the information processing device 1. In the translation process SS1 of this embodiment, as an example, the "base language prompt P ｎ The information processing device 1 uses the generating AI model server 2 to perform translation so that the "base language prompt P" is provided to the generating AI model server 2 in the generation process S5. ｎ "Then, prompt P ｎ However, it is primarily written in the base language.

[0056] As a specific example, in translation processing SS1, the information processing device 1 receives user input information IN ｎ This is converted to a base language (e.g., English). As a result, in this embodiment, the prompt creation process S3 and the generation process S5 can be executed in an "English-based" manner, for example. In this embodiment, "English-based" means "primarily written in English," "based on English," or "primarily English."

[0057] 2-2-2-2. Result Data Acquisition Process S2a Next, in this embodiment, as an example, the information processing device 1 is made to execute the result data acquisition process S2a. In this embodiment, as an example, experimental data (in Figure 4, "BBB.csv") is imported from the UI screen 130 in Figure 4 as an example of input data INA. Alternatively, clicking the detailed input button 135b may display an input form as a pop-up. Supplementary information as an example of input data INA may be entered into this input form. The supplementary information may be, for example, details of the imported experimental data (e.g., "experimental method" and "experimental conditions").

[0058] In the result data acquisition process S2a of the embodiment, as an example, the information processing device 1 performs the following (a1) and (a2). (a1) The result data R1 is acquired by having the analysis application 12c1 perform analysis based on the input data INA. For example, analysis may be performed based on the input data INA "BBB.csv" shown in Figure 4. The input data INA is user input information IN ｎ Even a portion of the information contained in it is acceptable. (a2) Result data R1 is entered into user input information IN ｎ To incorporate it as such.

[0059] The analysis application 12c1 may, for example, be an analysis application that incorporates a neural network (NN) model. The analysis application 12c1 may take in experimental data, experimental methods and experimental conditions, and perform data analysis based on the taken-in data. In this embodiment, Multi-Sigma® is used as an example of the analysis application 12c1. The result data R1 may be a figure or table (including graphs, etc.), but may also be any other arbitrary data. It may also be text, strings, or numerical data. In place of the analysis application 12c1, any survey application, any evaluation application, or any estimation application may be used.

[0060] In this embodiment, as an example, the result data R1 includes figures and tables. Figures and tables may be assigned figure numbers, etc., within the analysis application 12c1, or user input information IN ｎ When the data is imported, a figure number or similar may be assigned. As an example, let's continue the explanation by referring to "Figure Q" in Figure 5B. The result data R1 (Figure Q) is, as an example, imported into the value of the user input information variable {FIG(X1)} in record M006 of table TB (see Figure 2A). The header string Hd in record M006 Ｍ００６ This is the Results. Prompt template PrT Ｍ００６ For example, "Explain the results of {FIG(X1)}." is one such prompt template. Ｍ００６ By combining this with the result data R1 (for example, Figure Q), the prompt PＭ００６ can be generated. This prompt P Ｍ００６ can be used to cause the generation AI model server 2 to generate an explanatory text about the content of the result data R1.

[0061] 2-2-3. Prompt Creation Process S3 Next, in the embodiment, as an example, the information processing apparatus 1 is caused to execute the prompt creation process S3. In the prompt creation process S3, the control unit 12 uses the user input information IN ｎ and the prompt template PrT ｎ to create a prompt P ｎ based on these.

[0062] An example of the prompt creation process S3 will be described. For convenience of explanation, the record M004 is used as an example. The heading string Hd Ｍ００４ of the record 004 is, as an example, an introduction. The prompt template PrT Ｍ００４ has a variable {KW1}. It is assumed that a keyword corresponding to the keyword type "purpose of research" is incorporated into the value of this variable {KW1}. The prompt template PrT ｎ is provided in advance and stored in the table TB. The prompt template PrT Ｍ００４ is, as an example, "What is the social background of the need for {KW1}?" The Japanese translation is "Please tell me the social background for the need of {KW1}."

[0063] In the embodiment, as an example, "reduction of greenhouse gas emissions" is input as a keyword related to "purpose of research" (see FIG. 4). In the embodiment, as an example, the English translation of this keyword "reduction of greenhouse gas emissions" is combined with the prompt template PrT Ｍ００４ to generate a prompt P Ｍ００４ corresponding to the heading M004. In the prompt P Ｍ００４ , "{reduce greenhouse gas emissions}" is substituted into the variable {KW1} of the prompt template PrT Ｍ００４ . In the embodiment, as an example, the prompt P is created by such a mechanism.ｎ It is possible to create this. Note that prompt P ｎ This is not limited to text; diagrams, charts, and other elements may be embedded in any form.

[0064] 2-2-4. Reference Information Acquisition Process S4 Next, in this embodiment, as an example, the information processing device 1 is made to execute the reference information acquisition process S4. In the reference information acquisition process S4, the information processing device 1 is capable of accepting input of at least one reference information Ref1. The reference information Ref1 may be information such as any paper, or information such as literature in a field related to the document data D1 to be created. The reference information Ref1 may be the electronic data of the literature itself, but is not limited to this, and may also be literature identification information that can be obtained by searching for literature (for example, in the case of an academic paper, the title and author, literature number, etc.). This is because if literature identification information is obtained, the information processing device 1 can acquire the reference information data via the generation AI model server 2 or another search engine. The reference information Ref1 can be used in the pre-import process S5a described later, and / or can be used in any way within the generation process S5 (such as an attachment to the document data D1, or a citation of any part).

[0065] 2-2-5. Generation Process S5 Next, in this embodiment, as an example, the information processing device 1 is made to execute the generation process S5. Here, in the generation process S5 of this embodiment, as an example, the information processing device 1 includes the pre-import process S5a as a substep.

[0066] 2-2-5-1. Pre-import processing S5a In the generation processing S5 of the embodiment, as an example, the information processing device 1 prompts the generation AI model server 2 with a prompt P ｎ Based on the reference information Ref1, generated data G ｎ This generates [something]. As an example of a mechanism to achieve this, the embodiment provides a pre-importation process S5a.

[0067] In the pre-loading process S5a of the embodiment, as an example, the information processing apparatus 1 makes the reference information Ref1 readable by the generation AI model server 2 without the learning of the generation AI model server 2. Specifically, in the pre-loading process S5a, the generation AI model server 2 may cooperate with a RAG system server (not shown). For example, the reference information Ref1 may be imported into the database of the RAG system server. Alternatively, for example, the generation AI model server 2 may temporarily hold the reference information Ref1 to make it readable. This temporary holding may, for example, transmit the content of the reference information Ref1 to the generation AI model server 2 for on-site analysis, or make the information in the file available as context. For example, in-context learning (ICT) may be used. As a result, the generation AI model server 2 can be made capable of answering with high accuracy by referring to the reference information Ref1.

[0068] 2-2-5-2. Communication with the Generation AI Model Server 2Following the above pre-loading process S5a, in the generation process S5, the information processing apparatus 1 causes the generation AI model server 2 to generate generation data G ｎ based on the prompt P ｎ . Specifically, as an example, the information processing apparatus 1 refers to the table TB1 to provide each prompt P ｎ to the generation AI model server 2 to generate an answer, and receives the answer and stores it as the generation data G ｎ .

[0069] First, the information processing apparatus 1 transmits the prompt P ｎ to the generation AI model server 2 via the API. In the generation process S5, if necessary, together with the prompt P ｎ , attached data (for example, any user input information IN ｎ such as chart data, or the result data R1, etc.) may be transmitted to the generation AI model server 2.

[0070] The information processing apparatus 1 acquires an answer from the generation AI model server 2 via the API. The generation data G ｎThis can be text (written text), a diagram or chart, or text with embedded diagrams or charts. The acquired generated data G ｎ This is stored in table TB. For example, prompt P Ｍ００４ Generated data G based on this Ｍ００４ This may be generated and stored in table TB.

[0071] 2-2-5-3. Differentiation of Multiple Generation AI Models In one embodiment, for example, in generation process S5, the information processing device 1 uses multiple generation AI model servers 2 in a differentiated manner. More specifically, in one embodiment, "Chat GPT" and "Consensus Copilot" can be used interchangeably depending on the situation.

[0072] In this embodiment, specific examples of "method for determining a generation AI model" and "method for differentiating the use of multiple generation AI models" are described in relation to the use of multiple generation AI models.

[0073] 2-2-5-3-1. Method for Determining the Generated AI Model The method for determining the generated AI model server 2 used in the generation process S5 is not limited and may be "changeable" or "fixed". If changeable, manual change or automatic processing can be used. If "fixed", it may always be fixed or fixed in principle.

[0074] In one embodiment, "manual change" is used as an example, and the user can specify it using the generation AI model selection menu 138a in Figure 4. Another example is "fixed," meaning that multiple generation AI models are pre-set and the user cannot specify or change them. Yet another example is "automatic processing," where processing is provided to automatically determine, set, or suggest the optimal or preferred generation AI model according to the type and content of the document. If manual change is not used, the generation AI model selection menu 138a may not be necessary.

[0075] The large-scale language models usable in generation process S5 may include, for example, various generative AI models from OpenAI Inc. (GPT (Generative Pretrained Transformer)-3 (registered trademark), GPT-4 (registered trademark), ChatGPT, etc.).

[0076] As another example, a generative AI model specialized for a specific field or application may be adopted. For example, if document data D1 is an academic paper, a generative AI model specialized for academic papers may be adopted. For instance, a generative AI model based on an academic paper AI search model is configured to identify academic papers that match the search keywords, and to extract answers to questions from the identified papers. Another advantage is that the cited references presented along with the answers are accurate. An example of such a generative AI model is "Consensus Copilot" described in the embodiment.

[0077] Other models that can be used in the generated AI model server 2 will be explained later in the modified version (third modified version).

[0078] 2-2-5-3-2. Method for Differentiating the Use of Generated AI Models There are no limitations on the method for differentiating the use of the generated AI model server 2. The content of the method of differentiation may be "specifiable" or "fixed". If it is specifiable, manual specification or automatic processing can be adopted. If it is "fixed", it may always be fixed or fixed in principle. Various variations are envisioned for the specific content of the method of differentiation, but for example, the following (Y1) and (Y2) may be included. (Y1) Processing unit of document data D1 (Y2) Method for determining the differentiation conditions

[0079] First, let's describe the processing unit of the document data D1 (Y1) mentioned above. A single document data can be divided into multiple parts according to arbitrary criteria. A single document data can be arbitrarily divided according to the text position and / or semantic content. A different generation AI model can be used for each divided part. For example, for each generation process S5 of each item in the document data D1 (as an example, each prompt P Ｍ００１ ~P Ｍ０１１For example, different generating AI model servers 2 may be used. Another example is that different generating AI model servers 2 may be used depending on the differences in information such as text, figures, and numerical data. The user may also be able to specify which one to use.

[0080] Next, the method for determining the (Y2) usage conditions will be described. For example, the usage conditions may be user-specifiable. As a specific example, in this embodiment, clicking the usage condition setting button 138b (see Figure 4) next to the generated AI model selection menu 138a allows the user to set the usage conditions for each generated AI model server 2. As another example, the usage conditions for multiple generated AI model servers 2 may be fixed in principle to pre-set conditions. As yet another example, processing may be provided for automatically determining, suggesting candidates for, or automatically setting the optimal or preferred usage conditions.

[0081] Referring to Figure 4, in this embodiment, as an example, when the usage condition setting button 138b is clicked, a pop-up screen etc. is displayed, and each heading (specifically, each heading string Hd of table TB) ｎ A checkbox corresponding to the heading may be presented. By checking each checkbox, the user may be able to specify which generative AI model should generate the body text for which heading. In addition, while the generative AI model server 2 can use a large-scale language model for text generation, it is not limited to this and other generative AI models (e.g., image generation models) may be used in combination.

[0082] Furthermore, if the user has specified only one generation AI model in Figure 4, the generation process S5 may be completed using only that single generation AI model. In this case, the distinction described above does not apply.

[0083] 2-2-6. Loop Determination Process S6 In this embodiment, as an example, the information processing device 1 is configured to repeat the prompt creation process S3 and the generation process S5.

[0084] In the loop determination process S6 of the embodiment, as an example, it is determined whether the generation process S5 has been completed for all pre-defined text blocks. A "text block" refers to a single text area in the document data D1, and specific examples include paragraphs, sections, or subsections. Each heading (Hd Ｍ００１ ~HD Ｍ０１１ A text block corresponding to the above is provided, and in the embodiment, as an example, each text block contains each generated data G in table TB. Ｍ００１ ~G Ｍ０１１ The content is inserted. The processing result of processing step S6 is, for example, the required generated data G. ｎ If all the necessary conditions are met, the answer is affirmative (YES); if they are incomplete, the answer may be negative (NO).

[0085] 2-2-7. Block Change Process S7 If the loop determination process S6 is negative (NO), proceed to the block change process S7. In the block change process S7, the text block (heading) to be processed is moved to the next text block (heading). In this embodiment, as an example, the heading string Hd ｎ Depending on the relationship between them, generated data G ｎ The generation order is predetermined. Specifically, the items to be processed are set in a predetermined order between records M001 and M011, and the prompt creation process S3 is executed sequentially accordingly. The processing order is not necessarily limited to ascending or descending order of record IDs, but is predetermined to an appropriate order according to the document type, etc.

[0086] Here, in the prompt creation process S3 of the embodiment, as an example, the information processing device 1 generates data G that was already generated in the previous generation process S5. ｎ This is used to create the prompt. "Already generated data" is, as an example, generated data G obtained in other processing steps within the same document creation. ｎ Alternatively, generated data G obtained from other document creation within the same session. ｎ But that's fine.

[0087] To illustrate an example, let's explain using table TB (see Figure 2A). In this embodiment, as an example, document data D1 has a first heading (e.g., Hd Ｍ００６ ) and the second heading (for example, HD Ｍ００７ ) and the first and second headings are placed in different positions within document data D1. Here, as an example, in table TB, the user input information IN of record M007 Ｍ００７ The value {G Ｍ００６ Assume that} is set. Value {G Ｍ００６} is the generated data G for heading M006. Ｍ００６ This is an instruction to incorporate it.

[0088] In the prompt creation process S3 of the embodiment, as an example, the information processing device 1 performs the following (x1) and (x2). (x1) The information processing device 1 generates the generated data G generated in the generation process S5 for the first heading. ｎ At least a portion of it, User input information for the second heading IN ｎ It will be incorporated as follows. For example, the first heading HD Ｍ００６ Generated data G Ｍ００６ The string is, Second heading HD Ｍ００７ User input information for IN Ｍ００７ It may be incorporated as follows. (x2) The information processing device 1 receives the user input information IN acquired in (x1) above. ｎ (Example: IN Ｍ００７ ) to the second heading (e.g., HD Ｍ００７ ) and the associated prompt template PrT ｎ (Example: PrT) Ｍ００７ ) is combined with this. This results in prompt P ｎ (Example: P Ｍ００７ ) generates. Example prompt P Ｍ００７ The second heading is HD Ｍ００７ Generated data G Ｍ００７ This is a prompt to obtain [the desired result].

[0089] This will result in the generation data G that was generated in the past. ｎ (In other words, generated data G) Ｍ００６ The string (of the above) can be recursively used to create a prompt for heading M007.

[0090] Although omitted for convenience in the embodiment, similarly, prompt templates PrT are also used for other records in table TB. ｎ These are stored in advance. In some records, user input information IN ｎ In the column, generate data G in other records. ｎ Some or all of the above may be applied. This may enable the prompt creation process S3 and the generation process S5 to be executed automatically and sequentially across multiple different blocks (specifically, multiple heading areas) in the document data D1.

[0091] The processing loop of processing steps S3, S5, S6, and S7 is repeated, generating data G Ｍ００１ ~G Ｍ０１１ These are continuously generated. Eventually, the necessary generated data G ｎ All the necessary components are present. In this embodiment, for example, at this point the determination result of processing step S7 becomes positive (YES), and the process exits the processing loop and proceeds to document creation processing S8. Note that the above loop processing may also be implemented using the standard processing functions of the control unit 12, and the loop determination processing S6 and block modification processing S7 are not shown in the functional block diagram of Figure 2B.

[0092] 2-2-8. Document Creation Process S8 Next, in this embodiment, as an example, the information processing device 1 is made to execute the document creation process S8. In the document creation process S8, the information processing device 1 generates data G ｎ Document data D1 is created based on this. In the document creation process S8 of this embodiment, as an example, the information processing device 1 generates each generated data G ｎ These are placed in positions corresponding to each heading in document data D1 according to a pre-set correspondence. In this embodiment, as an example, each heading string Hd ｎ The position is predetermined according to the type or content of document data D1 (see Figure 5A). As shown in Figure 5B, the heading string Hd ｎ At the corresponding position, each generated data G ｎ It is inserted. The corresponding position is, for example, the bottom or right side in a horizontal document, and the left side in a vertical document.

[0093] The document creation process S8 presents the UI screen 140 (preview screen) as illustrated in Figure 5A. This allows user U to view the contents of document data D1. As illustrated in Figure 5A, the document data D1 has pre-set positions corresponding to each record M001 to M011. Generated data G is set at the positions corresponding to each record M001 to M011. ｎ Contents (each generated data G) Ｍ００１ ~G Ｍ０１１ Each of the strings is inserted. Figure 5B shows an example of the contents of record M006 (Result). Note that the document data D1 can be downloaded by clicking the save button 141. Note that the positions corresponding to each record M001 to M011 may be in a fixed order set in advance, or they may be changed arbitrarily by user U.

[0094] In this embodiment, as an example, an output language specification menu 145 is provided on the preview UI screen 140. By operating the output language specification menu 145, the user can specify the output language of the document data D1 (for example, the language for presentation, saving, transmission, or printing) to any language (for example, Japanese).

[0095] It should be noted that, in this embodiment, as an example, the document creation process S8 includes the translation process SS1. The translation process SS1 processes each generated data G in table TB. Ｍ００１ ~G Ｍ０１１ The string is translated into the user-specified output language (e.g., Japanese), and the translation result is inserted at each position in document data D1.

[0096] 2-2-9. Operation Reception Processing S9 Next, the information processing device 1 of the embodiment can, as an example, execute operation reception processing S9. In operation reception processing S9, the information processing device 1 receives document data D1 and generated data G ｎ or prompt P ｎ User editing operations regarding this are possible.

[0097] In the preview UI screen 140 illustrated in Figure 5A, clicking the edit button 142 activates the edit mode. In edit mode, clicking any heading area in document data D1 (for example, the M004 area) transitions to the edit UI screen 150 illustrated in Figure 6A. For example, heading M004 "Introduction" is editable. Furthermore, clicking the edit button 152 in Figure 6A makes the text in the display area 151 editable. Figure 6B shows an example after editing, with the underlined section "In Japan..." inserted. Clicking the save button 156 updates document data D1 with the edited content.

[0098] Furthermore, when the insert button 153 is clicked on the editing UI screen 150 in Figure 6A, one or more user-saved documents previously saved by user U may be presented as insertion candidates. This allows the user-saved documents to be inserted at any position within the display area 151. Also, when the prompt editing button 154 is clicked, the currently displayed generated data G may be presented. ｎ The prompt P used to generate ｎ For example, a prompt P may appear as a pop-up. ｎ The user U may be able to edit and save it. When the regenerate button 155 is clicked, the edited prompt P ｎ Then, the generation process S5 may be executed again. The generated data G obtained as a result of the re-execution. ｎ The display area 151 may then be updated.

[0099] As described above, in the document creation process S8 of the embodiment, the information processing device 1 generates data G ｎ Document data D1 can be created based on user editing operations. User editing operations generate data G ｎ This may include any editing operations on the content, and any editing operations may include adding, deleting, modifying, or otherwise generating data G ｎ This may include combining data or appending to user-saved text prepared in advance.

[0100] It should be noted that, in this embodiment, as an example, the operation reception process S9 includes the translation process SS1. The translation process SS1 generates each of the generated data G in table TB. Ｍ００１ ~G Ｍ０１１ Each string may be translated into a user-specified language (e.g., Japanese), and the translation result may be displayed in the display area 151. Furthermore, the edited content illustrated in Figure 6B may be translated into a base language (e.g., English) by the translation process SS1, and then re-stored in the table TB.

[0101] 2-2-10. Evaluation Process S10 Next, the information processing device 1 of the embodiment can, as an example, perform the evaluation process S10. When the evaluation button 143 is clicked on the UI screen 140 illustrated in Figure 5A, the pop-up UI screen 160 illustrated in Figure 7 is displayed. In the evaluation process S10 of the embodiment, as an example, the information processing device 1 presents the evaluation of the document data D1 on the pop-up UI screen 160. The translation process SS1 may perform the necessary translations according to the evaluation items (e.g., document comparison between different natural languages).

[0102] The evaluation presented in the evaluation process S10 of the embodiment may, as an example, include at least one of the following (b1) to (b3): ​​(b1) Similarity between the content of other documents and the content of document data D1; (b2) Novelty score of the content of document data D1; (b3) Importance score of the content of document data D1.

[0103] Furthermore, in the evaluation process S10 of the embodiment, the following specific examples (b11) to (b13) may be presented as more specific examples tailored to the paper: (b11) Presentation of similar parts when the evaluation includes (b1) above; (b12) Novelty evaluation of the description of the Method and the description of the Results when the evaluation includes (b2) above and the document data D1 is a paper; (b13) Social significance based on the description of the introduction and / or the description of the Results (e.g., degree of technological contribution, technological progress, technological improvement, etc.) when the evaluation includes (b3) above and the document data D1 is a paper.

[0104] "Similarity" is related to (b1) and (b11) above and can be obtained, for example, by any comparison process that compares the similarity between multiple document data. For example, document data D1 and any other document data (for example, reference information Ref1) may be compared at the sentence level, paragraph level, section level, or as a whole. Any large-scale language model, any natural language machine learning model, or any rule-based document comparison program may be used as the document comparison tool. For example, the percentage of similarity may be presented depending on the amount of sentences that are judged to be identical, substantially identical, or similar. Specifically, "presentation of similar parts" may be compared with any other document data, for example, at the sentence level or paragraph level. Similarity evaluation also has the advantage of enabling plagiarism checks.

[0105] In this embodiment, for example, when the confirmation button 161 on the pop-up UI screen 160 in Figure 7 is clicked, the user may transition to another UI screen. On the other UI screen, sentences that are judged to be similar may be highlighted, for example, or they may be displayed side-by-side in two columns, for example. In the example in Figure 7, the result "5% similarity for 1 item (similar document XXXXX)" is obtained, but this is just an example. Multiple similar documents may be presented, and the similarity score for each similar document may be presented.

[0106] The "novelty score" is related to (b2) and (b12) above. For example, the novelty score may be high if a specific part of a document's data is not described in any other document. In particular, in the example of (b12) above, the descriptions of "methods" and "results" are evaluated in detail. For example, if the same literature is found through a search for both methods and results, the novelty score may be presented as zero or sufficiently low. It may be calculated as follows, for example, in a table.

[0107]

[0108] The "importance score" is related to (b3) and (b13) above and may be presented based on, for example, words, sentences, paragraphs, or context within document data D1. For example, the "importance score" may be calculated by taking into account the content of the "introduction" and / or "results" in document data D1. For example, if the "Introduction" of document data D1 clearly states how it contributes to society, or explains that the degree of contribution is higher than before, a higher importance score may be presented. For example, the "results" may quantitatively evaluate how much it contributes. If the "results" section includes descriptions such as a very high reduction rate (tens of percent or more) in energy consumption, or a significant improvement in automobile fuel efficiency, a higher importance score may be presented. The importance score for the entire document data D1 may be calculated by taking into account the importance scores of both the "introduction" and the "results".

[0109] 2-2-11. Posting Process S11 Next, the information processing device 1 of the embodiment can execute posting process S11 as an example. When the post button 144 is clicked with one or more output languages ​​specified in the output language specification menu 145 on the preview UI screen 140 illustrated in Figure 5A, posting process S11 is executed. In posting process S11, the information processing device 1 can send one or more specified output language versions of document data D1 to one or more document publishing system servers 3. The document publishing system servers 3 may be arbitrarily specified by the user U or may be pre-configured. This makes it possible to publish or publish document data D1 in any language version. The input language specified by the user (see user-specified language menu 131a in Figure 4) may be automatically set, temporarily set, or suggested as the output language.

[0110] After that, the routine ends. User U may recreate the same document data D1 within the same session, create new document data D1 after switching sessions, or log out. According to the embodiment described above, it is possible to effectively support document creation by utilizing the generation AI model.

[0111] 3. The details of the modified embodiments are illustrative and various modifications are possible. For example, any one of the modified embodiments described below may be adopted, or a combination of multiple modified embodiments may be adopted.

[0112] 3-1. First Variation (User Input Information, etc.) User Input Information IN ｎ Various types of information can be used. For example, user input information IN ｎ If the input is textual information, there are no restrictions on the length of the string, etc. Acceptable textual information may include, for example, any character, number, symbol or term, any keyword, any phrase, any sentence, any paragraph, or any text, or one or more arbitrary documents, any document, or any book. User Input Information IN ｎ Other examples may include arbitrary chart data, analytical data, or any other arbitrary data. Other arbitrary data may include, for example, 3D models, three-dimensional point cloud data, big data, or multimedia data (e.g., video or audio). If user input information is directly included... ｎ If the data cannot be incorporated into the generating AI model, it can be integrated with a large-scale language model by utilizing preprocessing or multimodal AI technology. By combining several of the examples above, user input information can be input. ｎ It can be incorporated as such.

[0113] 3-2. Second variation (Generated data G) ｎ (Recursive use, etc.) In the information acquisition process S2, the generated data G output by the generated AI model server 2 in previous processing steps ｎ Some or all of the keywords may be automatically entered into the keyword input field, suggested, or presented as selection options. For example, this has the advantage of improving work efficiency when the same user reuses the same experimental data, etc., to write papers on the same or similar themes.

[0114] 3-3. Third Modification (Language Correspondence and Translation Processing, etc.) 3-3-1. Details of the Base Language The base language will be explained in more detail below. As mentioned above, the base language is a natural language to which the language model has adapted through learning. Various natural languages ​​differ from one another in terms of words, grammatical structure, or word order. If the natural languages ​​mainly included in the training data are different, the way the language model is trained will be different, and the natural languages ​​to which the language model adapts will also be different. "Adaptation" means that through learning (pre-training or post-training), the internal parameters of the language model are optimized or specialized for a specific language.

[0115] To give a concrete example, in a language model that incorporates a neural network (NN), each node in each layer has internal parameters such as weights and biases. These parameters are adjusted during training. Many recent generative large-scale language models (e.g., GPT) are based on a transformer architecture and incorporate multiple NNs. If the natural language primarily used in the pre-training data is different, the parameter tuning results for each NN will also differ.

[0116] Taking this into consideration, the base language may be defined by one or more of the following (Lb1) to (Lb4).

[0117] 3-3-1-1: (Lb1) When the pre-trained data in the training phase of the pre-trained data language model (generating AI model server 2) is mainly based on a specific natural language, this specific natural language may be considered the base language. For example, if "mainly" means proportion, then in training data containing multiple natural language contents, the majority (largest group) natural language is "the specific natural language on which the training data is mainly based." For example, if the training data consists of 40% English, 20% Japanese, 20% German, and 20% Chinese, then the majority is English, and the base language may be considered to be English. When defining proportions of training data etc. using the term "mainly," various definitions using specific numerical ranges can be adopted, as described at the beginning of the embodiment.

[0118] 3-3-1-2: (Lb2) When the tuning of the tuned language model (generating AI model server 2) is mainly based on a specific natural language, this specific natural language may be considered the base language. Specifically, this includes cases where the language used is biased towards a specific natural language during post-training adjustments (tuning) or fine-tuning. Additional post-training in fine-tuning may also be mainly based on a specific natural language. The definition by proportion is the same as for the pre-training data in (Lb1) above.

[0119] 3-3-1-3: (Lb3) Benchmark score The benchmark score of the language model (generating AI model server 2) is higher for the base language than for other natural languages. By comparing different languages ​​in benchmark tests, the language to which the language model is adapted can be identified retrospectively. More preferably, the base language is the natural language that has the highest overall benchmark score and / or the highest benchmark score particularly related to response accuracy or response speed.

[0120] 3-3-1-4: (Lb4) When the AI ​​model server 2 (language model) is being trained using data from the Internet of High Resource Languages, a high resource language on the Internet may be considered the base language. Examples of high resource languages ​​include English, Russian, German, Spanish, French, Japanese, Portuguese, Italian, Persian, Polish, Chinese, etc. The base language may be a single natural language selected from these high resource languages.

[0121] The base language may be one natural language selected from the group of English, Japanese, French, German, Spanish, Chinese, and Korean. The choice of base language will be influenced by the bias in the training data (language user population, economic scale, international influence, level of academic research activity, etc.) and the nationality and official language of the language model development company. Currently, the majority of internet content is in English, and academic papers are mainly in English. On the other hand, non-English candidates often include information in Japanese and other languages ​​listed above.

[0122] Preferably, the base language of a language model is the natural language to which the language model has "best" adapted through learning. The natural language that best fits each of the definitions (Lb1) to (Lb4) above may be considered the base language.

[0123] 3-3-2. Specific Examples of Related Processes Next, specific examples of each process related to the language are described. For example, the following (L1a) and / or ((L1b) are possible, and these have already been used in embodiments. (L1a) Prompt Template PrT ｎ However, it may also be one that has been prepared in advance, mainly in the base language. (L1b) In translation processing SS1, the generated data G is prepared so that the document data D1 can be output in one or more user-specified languages ​​(see menu 145 in Figure 5A). ｎ You may perform a translation for this.

[0124] In this embodiment, as an example, arbitrary internal processing is performed in English, but it is not limited to this. As another example, if the base language is any non-English language, various processing may be performed in that non-English language. For language support, for example, any number of tables TB may be prepared, and each table etc. may have a header string Hd in a different language ｎ and prompt template PrT ｎ The following may be stored:

[0125] As an example, the large-scale language models described below are thought to have English as their base language. Specific examples include OpenAI's GPT series (i.e., GPT-3®, GPT-4®, ChatGPT, etc.), LLaMA®, Claude, PaLM®, GroK, Microsoft Copilot, or Consensus Copilot. Each language model with English as its base language is optimized or specialized for English. As an example, tsuzumi or Fugaku-LLM are thought to have Japanese as their base language. As an example, ERNIE is thought to have Chinese as its base language.

[0126] There are no limitations on the specific means of implementing the translation process SS1, and various known translation applications can be used. As a variation, translation may be performed by communicating with a neural machine translation (NMT: a machine translation engine using a neural network) translation server. Further examples include the use of rule-based machine translation (RBMT) or statistical-based machine translation (SMT) alone or in conjunction with it. Additionally, a lookup-type translation engine or a dictionary-based translation system may be used alone or in conjunction with it, and these may be primarily used for translation at the word level, keyword level, or phrase level. The translation process SS1 may perform translation by transmitting the target language and the text to be translated to any translation engine server.

[0127] To enhance the "multilingual support function," for example, the translation process SS1 may support not only Japanese but also one or more non-English languages. In this embodiment, the user-specified language may be changed at any processing step in the flowchart of Figure 3. The translation process SS1 may allow the user to specify / change any user-specified language at any point in each processing step, such as the document creation process S8.

[0128] It should be noted that, in one embodiment, the prompt creation process S3 and the generation process S5 proceed sequentially in the base language (specifically, English).

[0129] In this embodiment, as an example, a table TB based on a base language and a translation process SS1 are employed. In this embodiment, for example, sentences are written according to the grammatical structure and word order of the base language, and for example, the proportion of words from the base language is high. In this embodiment, as an example, even if words from a non-base language are included, for example, prompt P ｎ The element that determines the content of the instructions is a string of the base language, and / or, for example, each generated data G ｎ The main part of it (e.g., a string of characters such as a topic sentence) is the base language.

[0130] In this embodiment, a translation process SS1 is employed, and various documents and other elements are presented in the user interface in the user-specified language. Document creation processes (especially data communication with the generation AI model server 2) can consistently be based on the base language. This approach ensures accuracy while simultaneously providing convenience and work efficiency for the user U.

[0131] 3-3-4. Omission of Translation Processing Depending on the type of document data D1, for example, the required information accuracy may differ. If the information accuracy is sufficient to meet the user's needs, translation into the base language may not be necessary. In this case, the translation process SS1 may be omitted while using an arbitrary large-scale language model.

[0132] 3-3-5. Summary The language support in the embodiment can be summarized in the table below as an example. In the translation process SS1 of the embodiment, a base language and a non-base language (e.g., a user-specified language) are appropriately set as the source language or target language of the translation in order to perform the necessary translation between the following languages.

[0133]

[0134] 3-4. Fourth Modification (Regarding Pre-import Processing) The pre-import processing S5a can employ various techniques to enable the generation AI model server 2 to access various information based on the reference information Ref1. As a modification, the generation AI model server 2 may be made to perform training using the reference information Ref1 as training data. As one modification, if a pre-trainable generation AI model is provided, the pre-import processing S5a may be configured to cause this generation AI model to perform "pre-training". Pre-training may be fine-tuning, additional training, retraining, or retraining.

[0135] 3-5. Fifth Modification (Utilization of Multiple Generating AI Models, etc.) In this embodiment, as an example, multiple generating AI model servers 2 are used interchangeably or in combination. In "using interchangeably," for example, multiple generating AI model servers 2 may be selectively used within the same session depending on different scenes or situations (see the processing of generation process S5).

[0136] On the other hand, another example that may be adopted is "combined use". In "combined use", for example, within the same session, one prompt P ｎ Each of the multiple generative AI model servers 2 may provide an answer to this. Each answer may be used as is or processed and then combined to generate combined generative data. One of the multiple models may be used primarily, with the others used as secondary.

[0137] When adopting the above-mentioned "using different models" or "using them in combination," the multiple generative AI models used may, for example, fall under any one of the following (e1) to (e6): (e1) Generative AI models of the same series but different versions. (e2) Generative AI models whose training data differs in at least a portion of it. (e3) Generative AI models of different types. For example, a text generation AI and an image generation AI can be used in combination or interchangeably. (e4) Generative AI models whose post-training tuning differs. (e5) Generative AI models from different providers. If the providers are different, the training data etc. will usually be different, so the responses of the generative AI models may differ. (e6) Multiple generative AI models with different databases in the RAG system. If the database, i.e., external resources are different, the responses of the generative AI models may differ.

[0138] A typical example of a generative AI model server 2 is a large-scale language model. In large-scale language models, even if the model architecture is the same, different training data will result in different generated data G. ｎ The output is as follows. It should be noted that the above (e1) to (e6) are not necessarily mutually exclusive and may overlap.

[0139] As other examples, (e7) or (e8) below may be adopted. (e7) In a generation AI multi-agent system, the generation AI model used by the first agent and the generation AI model used by the second agent can be used in combination or interchangeably. (e8) Role-playing agents (characters) within the same generation AI model, instructed to behave as different agents, can be used in combination or interchangeably.

[0140] Multiple AI models may be selected from the AI ​​model generation selection menu 138a. Communication settings such as APIs may be switched according to the specified AI model generation. Note that the above fifth modification is merely an example, and in other modifications, only one AI model generation may be used.

[0141] 3-6. Sixth Modification (Position of Operation Acceptance Process) The operation acceptance process S9 is not limited to the processing step position shown in Figure 3. As one modification, as shown in Figure 8, the operation acceptance process S9 may be placed between the generation process S5 and the loop determination process S6. In this case, the generated data G ｎ Each time a file is retrieved, the user U can be prompted to confirm and edit it. The generated data G is obtained after confirmation and editing. ｎ However, since this information is reused in the prompt creation process S3 for the next text block (heading), it also has the advantage of being able to reflect the user U's intentions from the initial stage.

[0142] 3-7. Seventh Modification (Types of Document Data, etc.) There are no limitations on the document data D1 that can be created in this embodiment. Suitable documents for use in this embodiment include, for example, the following: For example, it can be used for papers (specifically academic papers or research papers), published technical reports, patent documents, any technical documents, or various reports. Other examples include textbooks, books, brochures, manuals, specifications, Q&A sheets, or contracts. It is preferable that the reference information Ref1 is related to the document data D1 to be created.

[0143] In the document creation support of the embodiment, there are no limitations on the field or specialized area. Document data D1 may belong to any of the fields of natural science, medical science (medicine, dentistry, pharmacy), or humanities and social science. For example, it may cover biology, physics, engineering, agriculture, biotechnology, computer and information science, medical science, pharmacy, education, psychology, or earth science. An example of a document suitable for use in the embodiment is "a document with a specified theme and a somewhat standardized format for headings and order." It is also suitable for creating documents related to data analysis of arbitrary data (experimental data, statistical data, or simulation data, etc.), or for creating documents that include explanations of figures and tables.

[0144] Furthermore, this seventh modification can be combined with the fifth modification, but in this case, depending on the type or field of document the user wishes to create, multiple different generative AI models may be used interchangeably or in combination. When employing domain-specific generative AI models, it may be possible to use or combine generative AI models according to their areas of expertise. When creating interdisciplinary research papers, it may be possible to use or combine multiple generative AI models specialized for each field.

[0145] 3-8. Eighth Modification (Inclusion of Reference List, etc.) As a modification, the references for each output result may be listed together in a predetermined format at an arbitrary location (for example, at the end) of the document data D1. In this embodiment, record M010 corresponds to "References," so the generated data G Ｍ０１０ References or cited information may be inserted and arranged in a single format. A predetermined format may be, for example, for a paper, "Number) Author name, journal abbreviation, volume (bold), starting page or paper number -- page (year of publication)." Alternatively, one or more of the items in this example may be arbitrarily adopted. For example, in document creation process S8, the document information contained in document data D1 may be extracted. Another example is that in prompt creation process S3, the "Information source presentation instruction" is set to prompt P ｎThis may be added. The instruction to provide information sources requests the presentation of references, citations, or website URLs related to the answer. Adding the instruction to provide information sources may be uniform or optional. Generated Data G ｎ If the document contains information such as references, this reference information may be extracted in document creation process S8. This eighth modification is even more effective when combined with a high-precision generation AI model in the field of academic papers, such as "Consensus GPT".

[0146] 3-9. Ninth Modification (Prompt Template, etc.) In this embodiment, the prompt template PrT ｎ Table TB is used for handling such data, but this is just one example. It is possible to use any other data format and processing method, not just tables. Furthermore, in this embodiment, a prompt template PrT is used for each header (each record M001 to M011). ｎ The system stores and manages, but is not limited to, prompt templates, etc., in any unit different from headings.

[0147] Prompt Template PrT ｎ This is not limited to cases where the content is fixed in advance in the form of a table or the like. Another example is using an arbitrary prompt generator to create a prompt template PrT ｎ The creation (generation) or customization of prompts may be performed dynamically. For example, a "prompt for prompt template generation" may be given to the prompt generator, for example, "Generate a prompt template for writing text following a heading based on a keyword." In this case, an arbitrary keyword (i.e., "user input information") and an arbitrary heading (for example, "Introduction") become the variable parts of the prompt for prompt generation.

[0148] 3-10. In other embodiments, any one or more of the functional blocks (12a to 12m) in Figure 2B and the processes (S1 to S11 and SS1) in Figure 3, etc., can be omitted. Also, the flowchart of the embodiment (see Figure 3 or Figure 8) is just an example, and the order of processes can be arbitrarily changed except for those whose order of processes is necessarily determined.

[0149] For example, one of the simple embodiments provided by the present disclosure is a "document creation support method using a generation AI model." One form of this simple document creation support method employs a prompt creation process S3, a generation process S5, a document creation process S8, a loop determination process S6, and a block modification process S7, and can be provided by omitting other processing steps. Any one or any multiple of the various features of the embodiments can be combined with this simple document creation support method. Any one or any multiple of the processing steps S1, S2, S2a, S4, S5a, S6, S9 to S11, and SS1 in Figure 3 can be combined with this simple document creation support method. The translation process SS1 can be employed in relation to "any processing that handles language" (either inside or before / after such processing). The same modifications can be made to the functional block diagram in Figure 2B.

[0150] In the embodiments, each functional unit of the control unit 12 (see Figure 2B) is presented as an example and does not limit the present invention. The flowchart in Figure 3 is also presented as an example and does not limit the present invention. In the program, information processing method, and information processing apparatus provided based on the embodiments, the "unit configuration" in the actual product, etc., does not have to be divided as shown in Figure 2B or Figure 3. Specifically, the "unit configuration" here includes, for example, components, modules, processes, steps, circuits, or circuit elements. For example, each functional unit in Figure 2B may be provided as an individual unit configuration, but is not limited thereto, and each functional unit in Figure 2B does not have to correspond one-to-one with the unit configuration of the actual product. For example, one unit configuration may provide multiple functional units, or multiple functional units may provide one functional unit. For example, at least a part of one unit configuration and at least a part of another unit configuration may cooperate to provide one or more functional units. It is sufficient that the control unit 12 as a whole is capable of achieving the functions provided by each or a combination of the above functional units.

[0151] 4. Additional Notes, etc. Various embodiments are illustrated below. The embodiments shown below can be combined with each other.

[0152] [Note 1] An information processing method for supporting document creation, wherein a prompt creation process, a generation process, and a document creation process are performed on at least one computer, the prompt creation process creates a prompt based on user input information based on user input and a prompt template, the generation process causes a generation AI model to generate generation data based on the prompt, and the document creation process creates document data based on the generation data.

[0153] According to the configuration described in Appendix 1 above, it is possible to effectively support document creation by utilizing a generative AI model.

[0154] [Appendix 2] An information processing method as described in Appendix 1, wherein in the prompt creation process, at least a portion of the generated data is used for prompt creation.

[0155] According to the configuration described in Appendix 2 above, the generated data can be recursively used to create prompts.

[0156] [Appendix 3] An information processing method according to Appendix 1 or Appendix 2, wherein the prompt is associated with the heading of the document data, and in the document creation process, each of the generated data obtained in the generation process is placed in the position corresponding to each of the headings in the document data according to the association relationship.

[0157] According to the configuration described in Appendix 3 above, document data can be created accurately and efficiently.

[0158] [Appendix 4] An information processing method described in any one of Appendix 1 to 3, wherein the information acquisition process is further performed by a computer, and in the information acquisition process, chart data is acquired as the user input information.

[0159] According to the configuration described in Appendix 4 above, the degree of freedom in document content can be improved.

[0160] [Appendix 5] An information processing method according to any one of Appendix 1 to 4, wherein the generating AI model includes a language model, and the computer further performs a translation process, in which the translation process performs translation such that the prompt of the base language is given to the generating AI model in the generation process, and the base language is a natural language to which the language model has adapted through learning.

[0161] The configuration described in Appendix 5 above has the advantage of making it easier to effectively extract the performance of the generative AI model.

[0162] [Note 6] An information processing method described in Note 5, wherein (La1) or (La2) below: (La1) The prompt template is mainly created in advance in the base language. (La2) In the translation process, translation is performed on the generated data so that the document data can be output in one or more user-specified languages.

[0163] According to Appendix 6 above, the advantages are (La1) improved accuracy of the language model's responses, or (La2) improved user convenience through language support.

[0164] [Appendix 7] An information processing method described in Appendix 5 or Appendix 6, wherein the base language is defined by one or more of the following (Lb1) to (Lb4): (Lb1) The base language is the specific natural language when the pre-training data in the training stage of the language model is mainly based on a specific natural language. (Lb2) The base language is the specific natural language when the tuning of the language model is mainly based on a specific natural language. (Lb3) The benchmark score of the language model is higher for the base language than for other natural languages. (Lb4) The language model is trained using data on the internet, and the base language is a high-resource language on the internet.

[0165] According to the above note 7, it is possible to appropriately select the base language.

[0166] [Note 8] An information processing method described in any one of Notes 1 to 7, wherein the result data acquisition process is further performed by a computer, and the result data acquisition process performs the following (a1) and (a2): (a1) Obtain result data by having an application program other than the generated AI model perform investigation, evaluation, analysis or estimation. (a2) Take the result data as the user input information.

[0167] According to the configuration described in Appendix 8 above, it is possible to support the creation of documents that incorporate the results of other analyses, etc.

[0168] [Appendix 9] An information processing method described in any one of Appendix 1 to 8, wherein the operation reception process is further performed by a computer, and the operation reception process accepts user editing operations on the document data, the generated data, or the prompt.

[0169] According to the configuration described in Appendix 9 above, users can actively intervene in the document creation process.

[0170] [Note 10] An information processing method described in any one of Notes 1 to 9, wherein an evaluation process is further performed by a computer, the evaluation process presents an evaluation of the document data, and the evaluation includes at least one of the following (b1) to (b3): ​​(b1) Similarity between the content of other documents and the content of the document data (b2) Novelty of the content of the document data (b3) Importance of the content of the document data

[0171] According to the configuration described in Appendix 10 above, it is possible to provide users with information to help them make decisions regarding document evaluation.

[0172] [Appendix 11] An information processing device for the information processing method described in Appendix 10, wherein the evaluation process presents one of the following (b11) to (b13): (b11) Presentation of similar parts when the evaluation includes (b1); (b12) Novelty evaluation of the description of the method and the description of the results when the evaluation includes (b2) and the document data is a paper; (b13) Social significance based on the description of the introduction and / or the description of the results when the evaluation includes (b3) and the document data is a paper.

[0173] According to the configuration described in Appendix 11 above, it is possible to provide more detailed information for document evaluation.

[0174] [Appendix 12] An information processing method described in any one of Appendix 1 to 11, wherein the posting process is further performed by a computer, and in the posting process, the document data is transmitted to the document publishing system.

[0175] The configuration described in Appendix 12 above offers the advantage of providing users with seamless support for a series of tasks, from document creation to submission. When Appendix 12 is combined with Appendix 10 or Appendix 11, users have the advantage of being able to submit documents after considering the evaluation results of those documents.

[0176] [Appendix 13] An information processing method described in any one of Appendix 1 to 12, wherein a computer is further made to perform a reference information acquisition process, the reference information acquisition process accepts input of reference information, and the generation process causes the generation AI model to generate the generated data based on the prompt and the reference information.

[0177] The configuration described in Appendix 13 above has the advantage of being able to improve the accuracy of the generated data responses.

[0178] [Appendix 14] An information processing method as described in Appendix 13, wherein the generation process includes a pre-import process, and the pre-import process enables the generation AI model to access the reference information by performing (c1) or (c2) below. (c1) Enables the generation AI model to read the reference information without requiring the generation AI model to learn. (c2) Causes the generation AI model to learn based on the reference information.

[0179] According to the configuration described in Appendix 14 above, there is an advantage in that the accuracy of the generated AI model's responses can be improved in the field of reference information.

[0180] [Appendix 15] An information processing method described in any one of Appendix 1 to 14, wherein in the generation process, multiple generation AI models are used separately or in combination.

[0181] According to the configuration described in Appendix 15 above, it is possible to leverage the advantages of each individual generative AI model.

[0182] [Appendix 16] An information processing method described in any one of Appendices 1 to 15, wherein the computer further performs session management processing, and the session management processing performs the following (d1) and (d2): (d1) When predetermined conditions are met, a session is switched. (d2) In response to the session switch, one or more of the user input information, prompts and generated data in the current session are deleted, and the next session is started.

[0183] According to the configuration described in Appendix 16, it is possible to suppress the mixing of information between different sessions.

[0184] [Appendix 17] A program that causes at least one computer to execute the information processing method described in any one of Appendix 1 to 16.

[0185] The program described in Appendix 17 above can be stored on a computer-readable, non-temporary recording medium.

[0186] [Appendix 18] An information processing device that performs each of the processing steps of the information processing method described in any one of Appendix 1 to 16.

[0187] The above appendix 18 refers to "at least one information processing device," which may be a single information processing device or multiple information processing devices working together.

[0188] While embodiments have been described above, these are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications are permitted. The embodiments and their variations are included within the scope and essence of the invention, as well as within the scope of the invention and its equivalents as described in the claims.

[0189] 1: Information processing device, 2: Generation AI model server, 3: Document publishing system server, 6: Communication network, 10: Communication unit, 11: Storage unit, 12: Control unit, 12a: Session management unit, 12b: Information acquisition unit, 12b1: Result data acquisition unit, 12c: Translation unit, 12c1: Analysis application (application program), 12d: Prompt creation unit, 12e: Reference information acquisition unit, 12f: Generation unit, 12g: Pre-import unit, 12h: Operation reception unit, 12j: Document creation unit, 12k: Evaluation unit, 12m: Posting unit, 13: Output unit, 14: Input unit, 15: Communication bus, 100: Information processing system, 130: Input UI screen, 131a: User-specified language menu, 131b: New session button, 132a: Keyword type selection menu, 132b: Keyword input area, 133: Input field increase button N, 134: Item transition button, 135a: Input area, 135b: Detailed input button, 136a: Analysis settings checkbox, 136b: Analysis app selection menu, 137a: Reference information destination input area, 137b: Pre-import checkbox, 138a: Generate AI model selection menu, 138b: Usage conditions setting button, 139: Start generation button, 140: Preview UI screen, 141: Save button, 142: Edit button, 143: Evaluation button, 144: Post button, 145: Output language specification menu, 150: Edit UI screen, 151: Display area, 152: Edit button, 153: Insert button, 154: Prompt editing button, 155: Regenerate button, 156: Save button, 160: Pop-up UI screen, 161: Confirmation button, INA: Input data, IN ｎ : User input information, KW1: Variable, R1: Result data, Ref1: Reference information, TB: Table, M001-M011: Record ID, HD ｎ : Headline text, PrT ｎ : Prompt template (base language prompt template), P ｎ : Prompt (base language prompt), G ｎ : Generated data (string data), D1: Document data, U: User

Claims

1. An information processing method for supporting document creation, comprising: causing a prompt creation process, a generation process, and a document creation process to be executed on at least one computer; in the prompt creation process, a prompt is created based on user input information based on user input and a prompt template; in the generation process, a generation AI model is caused to generate generation data based on the prompt; and in the document creation process, document data is created based on the generated data.

2. An information processing method according to claim 1, wherein in the prompt creation process, at least a portion of the generated data is used for prompt creation.

3. An information processing method according to claim 1, wherein the prompt is associated with the heading of the document data, and in the document creation process, each of the generated data obtained in the generation process is placed at a position in the document data corresponding to each of the headings in the document data according to the association relationship.

4. An information processing method according to claim 1, wherein the information acquisition process is further performed by a computer, and in the information acquisition process, chart data is acquired as user input information.

5. An information processing method according to claim 1, wherein the generating AI model includes a language model, and the computer further performs a translation process, wherein the translation process performs translation such that the prompts in the base language are given to the generating AI model in the generation process, and the base language is a natural language to which the language model has adapted through learning.

6. An information processing method according to claim 5, wherein (La1) or (La2) below: (La1) The prompt template is mainly created in advance in the base language. (La2) In the translation process, translation is performed on the generated data so that the document data can be output in one or more user-specified languages.

7. An information processing method according to claim 5, wherein the base language is defined by one or more of the following (Lb1) to (Lb4): (Lb1) The base language is the specific natural language when the pre-training data in the training stage of the language model is mainly based on a specific natural language. (Lb2) The base language is the specific natural language when the tuning of the language model is mainly based on a specific natural language. (Lb3) The benchmark score of the language model is higher for the base language than for other natural languages. (Lb4) The language model is trained using data on the internet, and the base language is a high-resource language on the internet.

8. An information processing method according to claim 1, wherein the result data acquisition process is further performed by a computer, and the result data acquisition process performs the following (a1) and (a2): (a1) Obtain result data by having an application program other than the generated AI model perform investigation, evaluation, analysis or estimation. (a2) Take the result data as user input information.

9. An information processing method according to claim 1, wherein the operation reception process is further performed by a computer, and the operation reception process accepts user editing operations on the document data, the generated data, or the prompt.

10. An information processing method according to claim 1, wherein an evaluation process is further performed by a computer, the evaluation process presents an evaluation of the document data, and the evaluation includes at least one of the following (b1) to (b3): ​​(b1) Similarity between the content of another document and the content of the document data (b2) Novelty of the content of the document data (b3) Importance of the content of the document data 11. An information processing device according to claim 10, wherein the evaluation process presents one of the following (b11) to (b13): (b11) Presentation of similar parts when the evaluation includes (b1); (b12) Novelty evaluation of the description of the method and the description of the results when the evaluation includes (b2) and the document data is a paper; (b13) Social significance based on the description of the introduction and / or the description of the results when the evaluation includes (b3) and the document data is a paper.

12. An information processing method according to claim 1, wherein the posting process is further performed by a computer, and the posting process transmits the document data to a document publishing system.

13. An information processing method according to claim 1, wherein a computer further performs a reference information acquisition process, the reference information acquisition process accepts input of reference information, and the generation process causes the generation AI model to generate the generated data based on the prompt and the reference information.

14. An information processing method according to claim 13, wherein the generation process includes a pre-import process, and the pre-import process enables the generation AI model to access the reference information by performing (c1) or (c2) below: (c1) Enables the generation AI model to read the reference information without training the generation AI model. (c2) Causes the generation AI model to perform training using the reference information.

15. An information processing method according to claim 1, wherein in the generation process, a plurality of generation AI models are used separately or in combination.

16. An information processing method according to claim 1, wherein a computer further performs session management processing, and the session management processing performs the following (d1) and (d2): (d1) When predetermined conditions are met, a session is switched. (d2) In response to the session switch, one or more of the user input information, prompts and generated data in the current session are deleted, and the next session is started.

17. A program that causes at least one computer to execute the information processing method described in any one of claims 1 to 16.

18. An information processing device that performs each process of the information processing method described in any one of claims 1 to 16.

Images