Information processing method, program, and information processing system

Abstract

To efficiently generate a data set using a large-scale language model.SOLUTION: There is provided an information processing method for generating a layered data set, the information processing method including a receiving step, an outputting step, and an acquiring step, wherein in the receiving step, a design prompt is received, the design prompt includes an instruction to generate the data set and is composed of a plurality of prompts that are layered into a plurality of layers, the prompts in the adjacent layers are associated with each other, in the outputting step, the design prompt is output to a large scale language model, and in the acquiring step, the data set is acquired from the large scale language model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

[0002] Patent Document 1 discloses a prompt generation technique in the field of large-scale language models. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7313757 Summary of the Invention [Problem to be solved by the invention]

[0004] By providing prompts to a large-scale language model, various data can be generated. Sometimes users want a single, comprehensive dataset. If generation work is performed piecemeal using a large-scale language model for each part of the dataset, there is a problem of increased man-hours required to integrate the generated data.

[0005] An object of the present invention is to efficiently generate a dataset using a large-scale language model. [Means for solving the problem]

[0006] According to the present invention, there is provided an information processing method for generating a hierarchical dataset, the information processing method comprising a receiving step, an output step, and an acquisition step, wherein the receiving step receives a design prompt, the design prompt having an instruction to generate the dataset and consisting of a plurality of prompts hierarchically arranged into a plurality of layers, the prompts in adjacent layers being associated with each other, the output step outputs the design prompt to a large-scale language model, and the acquisition step acquires the dataset from the large-scale language model.

[0007] According to the present invention, it is possible to efficiently generate a data set using a large-scale language model. [Brief explanation of the drawings]

[0008] [Figure 1] 1 illustrates an information processing system 100 and an information processing device 1 according to an embodiment. [Figure 2] FIG. 2A is a hardware configuration diagram of the information processing device 1, and FIG. 2B is a functional block diagram of the control unit 12. As shown in FIG. [Figure 3] 1 is a flowchart illustrating an information processing method according to an embodiment. [Figure 4] Fig. 4A is an example of a tree structure in which design information is hierarchically arranged, and Fig. 4B is a diagram for explaining a sheet display step S1 and a receiving step S2 according to the embodiment. [Figure 5] FIG. 10 is a diagram for explaining an example of a prompt hierarchical structure in Markdown format, and is a diagram for explaining an output step S4 of the embodiment. [Figure 6] FIG. 10 is a diagram for explaining an acquisition step S5 in the embodiment. [Figure 7] FIG. 7A shows how design information for the seat ST is selected in reference step S6 of the embodiment, and FIG. 7B shows how the corresponding portion is highlighted. [Figure 8]FIG. 8A is a diagram showing an example of a prompt in Mermeid notation, which is an example of Markdown format, and FIG. 8B is a diagram showing program code (SQL code), which is an example of data generated by the prompt in FIG. 8A. [Figure 9] FIG. 8B is an example of an ER diagram drawn by the prompt in FIG. 8A. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. Various features shown in the following embodiments can be combined with each other. Furthermore, each feature can be an invention independently.

[0010] 1. Description of the configuration of the embodiment 1.1. Information Processing System Configuration 1, an information processing system 100 includes, as an example, an information processing device 1 and a large-scale language model server 2 (an example of a large-scale language model), which are connected to each other via a communication network 6 (for example, the Internet) so as to be able to exchange information.

[0011] The information processing device 1 executes each step of the information processing method according to the embodiment (described later in FIG. 3 and subsequent figures). The information processing method according to the embodiment causes the large-scale language model server 2 to generate an arbitrary hierarchical dataset. In the embodiment, the term "dataset" simply means "a collection of data," and there are no limitations on the purpose, target, use, or format of the data. In the following embodiment, as an example, the dataset generated by the large-scale language model server 2 is assumed to be program code for building a system desired by a user U.

[0012] In the embodiments, the term "program code" broadly means "information that specifies instructions for a computer." The program code in the embodiments includes source code written in any programming language other than a natural language, but is not limited to this and also includes any markup language, markdown language, or any style sheet language (e.g., CSS). The generated program code is executed in a suitable specification and environment, allowing the development and implementation of a system desired by user U.

[0013] 2 has a neural network trained using, for example, a large amount of training data. The large-scale language model server 2 receives output from the information processing device 1 as input data, generates a response to the input data, and outputs the response to the information processing device 1.

[0014] The input data received by the large-scale language model server 2 is generally referred to as a "prompt." In the embodiment, a "prompt" is input data for issuing an arbitrary instruction to the large-scale language model server 2. The prompt can take any form, such as a sentence or a diagram, and can be set as any instruction, command, question, or the like. When the prompt is expressed as a sentence, various languages ​​can be used, and any language readable by the large-scale language model server 2 can be used. For example, in the embodiment, the language used to write the prompt can be any natural language, any programming language, any database language, any markup language or markdown language, or any style sheet language (e.g., CSS).

[0015] The specific configuration of the large-scale language model server 2 is not particularly limited, but any large-scale language model known as generative AI can be used. As a large-scale language model, ChatGPT by OpenAI, Inc. may be used, for example. The large-scale language model server 2 may use only one type of large-scale language model (for example, only ChatGPT), but is not limited to this, and may use a combination of multiple large-scale language models. In the embodiment, the large-scale language model server 2 does not adopt a form of linking with a database such as RAG (Retrieval Augmented Generation). In the embodiment, by optimizing the prompts to be input to the large-scale language model server 2 (for example, the order of input), program code for constructing a system desired by the user U is efficiently generated.

[0016] 1.2. Hardware configuration of information processing device As shown in FIG. 2A, the information processing device 1 has, as an example, a communication unit 10, a memory unit 11, a control unit 12, an information 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.

[0017] The communication unit 10 may employ, for example, a wired communication means such as USB, IEEE1394, Thunderbolt (registered trademark), wired LAN network communication, etc. The communication unit 10 may also employ a configuration in which it is connected to the communication network 6 via a wireless communication means such as wireless LAN network communication, mobile communication such as 3G / LTE / 5G, or Bluetooth (registered trademark) communication. The communication unit 10 may also be configured to use both the wired communication means and the wireless communication means described above.

[0018] The memory unit 11 stores, for example, various programs, constants, variables, and setting values ​​of the information processing device 1 executed by the control unit 12. Any storage device can be used for the memory unit 11. For example, the memory unit 11 can be a storage device such as a solid-state drive (SSD), or a random access memory (RAM) that stores temporarily required information (arguments, arrays, etc.) related to program calculations. Furthermore, the information processing device 1 may use an external storage unit (for example, an external storage medium, cloud storage, etc.) in addition to the memory unit 11. In the embodiment, for example, the memory unit 11 is configured to be able to store a sheet ST and design prompts 20a and 20b (see FIGS. 4B and 5) described below, and program code (see FIG. 6).

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

[0020] The information output unit 13 is, for example, a display unit of the information processing device 1. The information output unit 13 may be included in the housing of the information processing device 1 or may be externally attached. The information output unit 13 displays a graphical user interface (GUI) screen that can be operated by the user U. The information output unit 13 may be, for example, a display device such as a CRT display, a liquid crystal display, an organic EL display, a plasma display, or an electronic paper display, as well as a display device such as an illuminable light or a projector. It is optional whether or not the information processing device 1 includes the information output unit 13. For example, the output of the information processing device 1 may be displayed on a display unit located at a location independent of the location where the information processing device 1 is installed. The information output unit 13 may also have a device that outputs audio.

[0021] As an example, the input unit 14 is configured to accept an operation input to the information processing device 1 made by the user U. The input unit 14 may be included in the housing of the information processing device 1 or may be externally attached. The input unit 14 may employ, for example, at least one of a touch panel, a switch button, a mouse, a keyboard, etc. It is optional whether or not the information processing device 1 includes the input unit 14. For example, the operation input of the user U may be accepted via an input unit of another information processing terminal provided in a remote location independent of the information processing device 1, and the information processing device 1 may accept this operation input via a network.

[0022] 1.3. Functional block configuration of information processing device The functional configuration of the information processing device 1 according to this embodiment will be described with reference to Fig. 2B. Information processing by software stored in the storage unit 11 is specifically realized by the control unit 12, which is an example of hardware, and each functional unit included in the control unit 12 is executed.

[0023] The control unit 12 of the information processing device 1 includes, for example, a sheet display unit 121, a reception unit 122, a prompt generation unit 123, an output unit 124, an acquisition unit 125, and a reference unit 126. Some of these functional units exchange information with the large-scale language model server 2.

[0024] The above-described functional units (sheet display unit 121 to reference unit 126) may be implemented by software or hardware. When implemented by software, various functions can be realized by a CPU executing a computer program. The program may be stored on a non-transitory computer-readable recording medium, provided as a downloadable file from an external server, or implemented by cloud computing, which reads a program stored in an external storage unit and realizes the functions. When implemented by hardware, various circuits such as an ASIC, FPGA, or DRP can be used. In the embodiments, various information and concepts encompassing such information are handled. These are represented by high and low signal values ​​or quantum bits as a binary bit set consisting of 0 or 1, and communication and calculations can be performed by the above-described software or hardware configuration. The software may be a general-purpose OS or a dedicated OS.

[0025] 1.4. Processing Executed by Information Processing Device (Information Processing Method) Each functional unit included in the control unit 12 in Fig. 2B is configured to be able to execute each step of the flowchart in Fig. 3. The program according to the embodiment executes each step (information processing method) in Fig. 3, thereby causing a computer (information processing device 1) to execute the information processing method according to the embodiment.

[0026] The sheet display unit 121 is configured to be able to execute a sheet display step S1 (see FIG. 4B). The reception unit 122 is configured to be able to execute a reception step S2 (see FIGS. 4B and 5). The prompt generation unit 123 is configured to be able to execute a prompt generation step S3 (see FIG. 4B). The output unit 124 is configured to be able to execute an output step S4 (see FIG. 5). The acquisition unit 125 is configured to be able to execute an acquisition step S5 (see FIG. 6). The reference unit 126 is configured to be able to execute a reference step S6 (see FIG. 7). The specific contents of each step will be described later.

[0027] 2. Description of the operation of the embodiment 2.1. Overview of Information Processing Method of the Embodiment Embodiments include several of the following features. One of its features is that it manages design information for building a desired system in a hierarchical manner. Another feature is that prompts issued to the large-scale language model server 2 are managed hierarchically. Yet another feature is that the results generated by the large-scale language model server 2 are managed hierarchically. In the hierarchical management of the embodiment, the components to be managed (each design information, each prompt, each generated result, etc.) are appropriately arranged in each layer and each node of the tree structure. Within this tree structure, the components (nodes) of adjacent layers are managed in association with each other. Each of these features can be an independent invention, but as an example, an information processing method, a program, and an information processing system that include all of these features are provided in the embodiments.

[0028] Specifically, in the embodiment, as an example, design information for obtaining a system desired by a user U is managed as a systematically hierarchical "design information tree" (see FIG. 4A). The design information tree has a plurality of design information nodes. As an example, the design information includes a requirements definition, a basic design, a workflow, etc., which are assigned to each design information node.

[0029] In addition, in the embodiment, as an example, prompts to be output to the large-scale language model server 2 are hierarchically managed in correspondence with each design information node (see FIGS. 4B and 5). In the embodiment, prompts are issued to the large-scale language model server 2 in an appropriate order according to the systematic design information tree (see FIG. 5). This enables the large-scale language model server 2 to generate any data (for example, program code).

[0030] In addition, in the embodiment, as an example, the generated data generated by the large-scale language model server 2 can be systematically managed as a "hierarchical set of data." In the embodiment, as a specific example, the generated program code is managed as a "document tree structure (i.e., program code tree structure)" (see FIG. 6). The document tree structure has document nodes, and each document node is linked to a corresponding design information node and / or prompt node. This linking allows the original design information node or prompt node to be associated with and managed as a document node generated therefrom. Corresponding locations can be easily referenced (see FIGS. 7A and 7B), which facilitates the correction, modification, deletion, or addition of prompts or generated data (program code). Any prompt can be easily added to a required node in the design information tree structure. When a prompt is corrected, deleted, or added, the corresponding generated data (corresponding program code) is corrected, deleted, or added, and the hierarchical data set (program tree structure) is automatically updated.

[0031] 2.2. Design Information The design information of the embodiment will be described. Fig. 4A shows an example of design information for configuring a system desired by a user U. This design information includes, as an example, a system overview D0, an actor D 1-1 , D 1-2 ..., Use Case D 2-1 , D 2-2 ..., screen D 3-1 , D 3-2 , D 3-3..., screen element D 4-1 , D 4-2 , D 4-3 ..., Function D 5-1 , D 5-2 , D 5-3 ..., table column D 6-1 , D 6-2 , D 6-3 In the embodiment, the system overview D0 includes design information related to the entire system, such as the system environment, specifications, etc.

[0032] In the embodiment, a hierarchical tree structure is constructed as shown in Fig. 4A. This tree structure defines the relationships between multiple pieces of design information while hierarchically organizing the design information into multiple layers L0 to L6, etc. In the embodiment, this tree structure hierarchically manages what functions are implemented for which screen elements on which screens, and what information (tables and columns) in the database are referenced or written to for that purpose.

[0033] 2.3. Information processing method and specific processing 3, the information processing method of the embodiment includes, for example, a sheet display step S1, a reception step S2, a prompt generation step S3, an output step S4, an acquisition step S5, and a reference step S6. Each step will be described in detail below.

[0034] (Sheet display step S1) Please refer to FIG. 4B. In the sheet display step S1, the sheet display unit 121 displays the sheet ST of FIG. 4B on the information output unit 13. The user U can enter design information into the sheet ST via the input unit 14. The sheet ST is divided into input areas according to layers L0 to L6, etc. In each area, the design information of each node in FIG. 4A, such as system overview, actor, use case, screen, etc., is entered as a character string. In FIG. 4B, for the sake of simplicity, only the system overview D0, use case D1, etc. are displayed. 2-1 , screen D 3-1 , D 3-2 For convenience, the manner in which the above is entered is illustrated.

[0035] (Reception step S2) The following description will be made with reference to FIGS. 4B and 5. First, the "design prompt" will be described. In the embodiment, the "design prompt" is a program code (for example, the program code PC 3-1 , PC 3-2 , PC 3-3 , . . . ) to the large-scale language model server 2. This instruction may be, for example, a directive in any language or a directive in a natural language.

[0036] In the embodiment, for example, two types of design prompts 20a and 20b are provided, which are referred to as a first design prompt 20a (see FIG. 4B) or a second design prompt 20b (see FIG. 5) for the sake of distinction.

[0037] The sheet ST in Fig. 4B is for receiving each prompt constituting the first design prompt 20a. In the receiving step S2 of the embodiment, as an example, the receiving unit 122 receives an input of the first design prompt 20a from the user U via the sheet ST in Fig. 4B.

[0038] In the embodiment, the user U enters design information (any actor, any use case, any screen, etc.) in natural language on the sheet ST. Since the large-scale language model server 2 can interpret natural language, the design information in natural language can be treated as a natural language prompt. In the embodiment, the user U issues instructions to the large-scale language model server 2 via the sheet ST based on the information in the sheet ST, so the user U essentially enters a prompt on the sheet ST. In this sense, in the embodiment, the design information entered in natural language on the sheet ST is also referred to as a first design prompt 20a. The first design prompt 20a is made up of multiple natural language prompts.

[0039] It should be noted that, in a practical example, the information processing device 1 preferably performs a process of adding specific instructions to the character string stored in each record of each node and a process of converting the character string into Markdown format in order to ultimately generate a prompt. Although there may be cases where a completed character string is initially registered as a prompt, this is because, in general, the character string stored in the sheet ST alone often does not meet the requirements as a prompt for the large-scale language model server 2. In this case, it is preferable for the information processing device 1 to generate a prompt by concatenating specific instructions for each node with the character string registered in the sheet ST. In the case of Markdown notation, a prompt can be generated by combining multiple records to create Markdown and adding commands that specify what to do with the Markdown.

[0040] In this regard, in the embodiment, as an example, a predetermined append conversion process is performed on the character strings entered in each record of each node in the sheet ST. The second design prompt 20b is obtained by performing append conversion process on each piece of design information in the first design prompt 20a to convert it into Markdown notation. In the embodiment, as an example, the sheet display unit 121 executes a preset append conversion process. This append conversion process supplements any deficiencies in character strings so as to satisfy the prompt requirements for the large-scale language model server 2, and converts the natural language prompts in each input area (each cell) of the sheet ST into Mermaid notation prompts PRM0, PRM1, PRM2, PRM3, PRM4, PRM5, PRM6, PRM7, PRM8, PRM9, PRM10, PRM11, PRM12, PRM13, PRM14, PRM15, PRM16, PRM17, PRM18, PRM19, PRM20, PRM21, PRM22, PRM23, PRM24, PRM25, PRM26, PRM27, PRM28, PRM29, PRM30, PRM31, PRM32, PRM33, PRM34, PRM35, PRM36, PRM37, PRM38, PRM39, PRM40, PRM41, PRM42, PRM43, PRM44, PRM45, PRM46, PRM47, PRM48, PRM49, PRM50, PRM51, PRM52, PRM53, PRM54, PRM55, PRM56, PRM57, PRM58, PRM59, PRM60, PRM61, PRM62, PRM63, PRM64, PRM65, PRM66, PRM67, PRM68, PRM69, PRM70, PRM71, P 1―1 The second design prompt 20b can be rewritten as a plurality of Markdown language prompts (i.e., Mermaid notation prompts PRMO, PRMO). 1―1 It is composed of...

[0041] As described above, the first and second design prompts 20a and 20b are both prompts based on the design information tree shown in FIG. 4A, and represent the same content using different notations.

[0042] In the embodiment, as an example, in each of the first and second design prompts 20a and 20b, different design information such as a system overview, actors, use cases, etc. are defined according to each layer L0 to L6, etc. As shown in FIG. 5, in the embodiment, the first and second design prompts 20a and 20b define a plurality of prompts PRM0, PRM1, PRM2, etc. that constitute the second design prompt 20b. 1―1 , PRM 1―2 ... is hierarchically organized into multiple layers L0 to L6...

[0043] In each of the first and second design prompts 20a and 20b, the prompts in adjacent layers are associated with each other. For example, in the first design prompt 20a of FIG. 4B, actor D is in layer L1. 1-1 The information is input and used in layer L2. 2-1 , D 2-2 are entered, and the relationship between these pieces of information is specified in the input fields of layer L1 and / or layer L2. Also, in the second design prompt 20b of FIG. 5, as an example, the prompts in adjacent layers are associated by writing the prompts in Markdown format, more specifically, by using Mermeid notation. Specific examples of associating prompts in adjacent layers will be described in "4. Specific examples of prompts and program code."

[0044] In the first design prompt 20a, the system overview D0 is 0-1 Design Instruction D 0-1 indicates that the data generated by the large-scale language model server 2 based on multiple prompts constitutes "one complete data set." In the embodiment, as an example, the large-scale language model server 2 generates program code as a data set, and this program code is used to construct one complete system. Therefore, in the embodiment, the design instruction D 0-1 indicates that the plurality of prompts (design information) that make up the first design prompt 20a construct a single integrated system. Correspondingly, the second design prompt 20b is a system overview prompt PRMO that is a design instruction prompt PRMO.0-1 Includes design prompt PRM 0-1 As an example, the second design prompt 20b is composed of a plurality of prompts PRMO, PRMO 1-1 , PRM 1-2 ...shows that they form a single cohesive system.

[0045] (Prompt generation step S3) In the embodiment, as an example, based on a node in an upper layer, a node in a layer below this upper layer node may be automatically generated by the large-scale language model server 2. For example, a list of required screens may be generated from a use case, a list of screen elements may be generated from the use case and the screen, or a list of functions may be generated from the screen elements. In this manner, the information processing device 1 may be equipped with a function for automatically creating each node of the design information tree in FIG. 4A in a chain reaction from the upper layer to the lower layer. In the embodiment, character strings corresponding to each record of the node automatically generated as described above may be automatically generated as a prompt for the next layer. As an example of this automatic generation, prompt generation step S3 will be described.

[0046] Please refer to FIG. 4B. In the embodiment, as an example, a prompt generation step S3 is provided. In FIG. 4B, the prompt generation step S3 is illustrated by an arrow. In the embodiment, as an example, the prompt generation unit 123 generates a prompt (for example, a prompt of a higher layer (for example, layer L2)) 2-1 ) based on the prompt of the lower layer (e.g. layer L3) (e.g. screen D 3-1 , D 3-2 )

[0047] As one form of such prompt generation step S3, the prompt generation step S3 is realized using the large-scale language model server 2. That is, in the prompt generation step S3, an upper-level prompt is output to the large-scale language model server 2 (see FIG. 2B), which causes the large-scale language model server 2 to generate a lower-level prompt.

[0048] For example, use case D as a top-level prompt 2-1 When adopting the design information of, for example, "This use case D 2-1 The prompt generation unit 123 may input a question asking, "What kind of screen is needed in the large-scale language model server 2?" as a prompt to the large-scale language model server 2. As a result, the large-scale language model server 2 receives the screen D 3-1 ~D 3-3 Alternatively, the large-scale language model server 2 may not be used. In this case, the prompt generation unit 123 may execute predetermined rule-based processing or refer to a predetermined table, etc., so that when an upper-level prompt is input, a lower-level prompt is output in response to the input.

[0049] The processing of the prompt generator 123 is applicable to both the first and second design prompts 20a and 20b.

[0050] (Output step S4) 5. In the output step S4, the output unit 124 outputs the second design prompt 20b to the large-scale language model server 2. In the embodiment, as an example, in the output step S4, the output unit 124 outputs the prompts PRMO, ... 1-1 , PRM 1-2 ... is output to the large-scale language model server 2. As a data processing method for realizing this, in the embodiment, batch processing may be adopted as an example.

[0051] (Acquisition step S5) 6. In the acquisition step S5, for example, the acquisition unit 125 acquires the program code PC from the large-scale language model server 2. 3-1 , PC 3-2 , PC 3-3 For convenience, each program code PC 3-1 , PC 3-2 , PC 3-3The acquisition unit 125 acquires each of the acquired program codes PC 3-1 , PC 3-2 , PC 3-3 ... are stored in the storage unit 11. Each program code PC 3-1 , PC 3-2 , PC 3-3 ...and each prompt PRM 3-1 , PRM 3-2 , PRM 3-3 The ... are linked to each other and stored in the storage unit 11. This link links corresponding nodes in both tree structures.

[0052] Here, actors and use cases are requirements definitions, and in response to this, each piece of design information (screens, screen elements, functions, data objects, etc.) is sequentially determined at lower levels. In one embodiment, program code is actually generated and stored from level L3 onward, which corresponds to "screen." Taking this into consideration, program code corresponding to layers L0 to L2 is not generated in Figure 6. It should be noted that there may be cases where there are no screens at level L3, which corresponds to "screen." In such cases, functions may be linked below the level using batch files or APIs. Furthermore, the top layer L0 has various uses, such as describing a system overview, defining the environment, and specifying languages ​​and frameworks. In this way, layer L0 primarily allows for setting prerequisites, and the large-scale language model server 2 can select a language and build an environment based on these.

[0053] The acquisition unit 125 of the embodiment uses the sheet ST of the sheet display unit 121 as a storage format, for example. 3-1 , PC 3-2 , PC 3-3 The user U can easily understand the correspondence between the "···" and the first and second design prompts 20a and 20b.

[0054] (See step S6) Please refer to Figures 7A and 7B. In the reference step S6, for example, the reference unit 126 displays reference information 40 on the information output unit 13. The reference information 40 includes information about the components of the program code set 30 generated in the acquisition step S5 (i.e., each program code PC 3-1 , PC 3-2 , PC 3-3 , 20b) and the corresponding parts of the prompts constituting the first or second design prompt 20a, 20b are presented to the user U. As an example, the reference information 40 of the embodiment is generated when the user U selects the design information D with the mouse cursor MC on the sheet ST. 3-3 When you select the corresponding part, the program code PC 3-3 may be highlighted (see hatching in FIG. 7B).

[0055] It should be noted that FIG. 7B shows only one example, and the present invention is not limited to this. In the embodiment, the information processing device 1 not only stores each program code but also stores the location of the program code, and preferably the location is always automatically updated. "Location" includes, for example, file name, line number, ID, class, etc. The program code and its location may be stored in the same table (table ST) as the design tree, but is not limited to this and may be stored in another table as long as the relationship can be maintained.

[0056] When a program is rewritten at a previous prompt in the order of a higher node or the same node, it is preferable that the information of the location located after the program is rewritten each time, and the location is always automatically updated.

[0057] In an embodiment, various prompt generation patterns across various hierarchies or records can be adopted. As an example, it is preferable to perform ID management for each generated prompt and automatically update the association between the generated program code and location.

[0058] 3. Description of the functions and effects of the embodiment As described above, according to the embodiment, the multiple prompts in the design prompt are appropriately hierarchically organized into multiple layers L0 to L6, etc. This allows the large-scale language model server 2 to generate a systematic dataset (as an example, systematic program code) based on the design prompts. As a result, the effort and man-hours required for integrating fragmented datasets can be reduced, allowing efficient data generation using the large-scale language model server 2. Another advantage is that the dataset generation work of the user U can be efficiently supported.

[0059] In particular, in the embodiment, systematic program code can be generated for the development of any system desired by a user U. The task of integrating fragmented program code can lead to an increase in the number of steps. The embodiment can effectively suppress this problem.

[0060] Furthermore, according to the embodiment, by presenting the reference information 40 to the user U in the reference step S6, there is an advantage that the user U can easily check or edit the data set (program code).

[0061] Furthermore, according to the embodiment, there is an advantage that the user U can easily organize the prompts for each layer L0 to L6, . . . in the sheet ST in the sheet display step S1.

[0062] Furthermore, according to the embodiment, in the prompt generation step S3, a lower-level prompt is generated based on an upper-level prompt, which has the advantage of enabling prompts to be generated systematically while reflecting the relationship between the design information between the upper and lower levels.

[0063] Furthermore, according to the embodiment, there is an advantage that the large-scale language model server 2 can be used to automatically generate lower-level prompts in the prompt generation step S3.

[0064] Furthermore, according to the embodiment, there is an advantage that prompts can be systematically given to the large-scale language model server 2 in an appropriate order from the top layer to the bottom layer in the output step S4.

[0065] Furthermore, according to the embodiment, by adopting the Markdown format for the second design prompt 20b, there is an advantage that the convenience of hierarchically managing a plurality of prompts is dramatically improved.

[0066] Also, according to the embodiment, the first design prompt 20a is a design instruction D 0-1 and the corresponding design instruction prompt PRM 0-1 can be provided to the large-scale language model server 2. This has the advantage that, after providing background information regarding a single, consolidated data set desired by the user U, the large-scale language model server 2 can process a large number of prompts in a systematic and unified manner.

[0067] 4. Examples of prompts and program codes See Figures 8A, 8B, and 9. Figure 8A is an example of a prompt in Mermaid notation, and Figure 8B is an example of program code (SQL code) generated by the large-scale language model server 2 using the content of Figure 8A as a prompt. Note, however, that some content has been omitted using "...". Also, because the example described here illustrates the association between prompts, for the sake of simplicity, we will not mention the specification of layer numbers in the prompt hierarchy.

[0068] Figure 8A prompt example PRM ex1 is an example of how to associate multiple prompts. ex1 Here is a prompt example: ex2 Contents and prompt example PRM ex3 , PRM ex4 It makes connections between each of the contents.

[0069] Fig. 9 is an ER diagram based on the contents of Fig. 8A. In the embodiment, by adopting Mermaid notation, there is an advantage that program code generation and various diagram creation (any diagram such as an ER diagram) can be realized simultaneously.

[0070] 5. Modifications, etc. The sheet ST in FIG. 4B can be in any format that can organize the design information into a list. As an example, the sheet ST can be a spreadsheet, or a data input area in any spreadsheet software, etc. The sheet ST is just an example, and any input form other than a sheet can be used. While it is preferable to have a function that organizes the design information into a list, this is not a limitation, and any data input method and data format can be used.

[0071] In the embodiment, the first design prompt 20a is converted into the second design prompt 20b and output to the large-scale language model server 2. However, this is not limiting, and as a modified example, the first design prompt 20a (the content of the sheet ST in FIG. 4B) may be output to the large-scale language model server 2 in natural language as is, in which case the second design prompt 20b may be omitted.

[0072] However, as another modification, the user U can of course manually perform any operation (such as correction, modification, addition, or deletion) on the prompt or program code described in the embodiment. For example, the user U can perform any additional input to the data of the second design prompt 20b stored in the storage unit 11. Alternatively, the user U can enter a prompt in any Markdown format or any programming language on a sheet ST or any other format, and this may be output to the large-scale language model server 2 as the first design prompt 20a. In this modification, the process of converting a natural language prompt is not required, and therefore the second design prompt 20b does not exist.

[0073] As another variation, the description method of the second design prompt 20b is not limited to Mermaid notation, and any Markdown format description method may be adopted, or any format that allows for hierarchical expression, not limited to Markdown format, may be adopted.

[0074] In the reference step S6, highlighting is just one example, and any reference function can be used. Any information that can identify the corresponding part may be displayed, such as the line number of the corresponding part.

[0075] As another modification, prompts may be provided separately from each node in an appropriate order to the large-scale language model server 2. Such a modification may be set as the default state. For example, this modification can be used in applications where it is necessary to define each element in detail, such as from a screen element to a function, or from a function to a type of data or data type referenced.

[0076] As another variation, the information processing device 1 may convert a multi-record character string at a specific node into a Markdown format such as Mermaid and output this as a prompt to the large-scale language model server 2. In particular, it is preferable that it be used in generating sequence diagrams, ER diagrams, class diagrams, etc. The format is not limited to Mermaid notation, and for example, when a relevance is selected for an arbitrary character string using a pull-down menu or the like, the information processing device 1 may automatically convert it into the specified Markdown format. This has the advantage that the user U can avoid the trouble of marking up. For example, in the case of Mermaid notation, the information processing device 1 may be configured to automatically define the following definition without the user having to enter it: ` ` ` mermaid erDiagram ||--o{ : ||--|{ :

[0077] As another variation, the information processing device 1 may generate and issue a prompt by concatenating character strings for each branch of a tree spanning multiple nodes. This can be used, for example, when defining a function in detail at multiple levels.

[0078] As a modified example of the prompt generation step S3 in the embodiment, the information processing device 1 may generate a prompt with N child node records (N is any natural number) for one parent node record. For example, if the parent node (upper layer node) has a screen and the child nodes (lower layer nodes) have multiple screen elements used on that screen, the information processing device 1 may generate a CSS according to a conversion process stored in advance using these as prompts, or may have the large-scale language model server 2 generate a CSS. For example, the CSS may be generated from an ID form, a password form, a login button, and a forgotten password button on a login screen.

[0079] In the information processing method according to the embodiment, the "hierarchical dataset" generated by the large-scale language model server 2 is not limited to program code (the program code group 30 in FIG. 6), and any dataset can be used. As a variation, the "hierarchical dataset" may be, for example, a scenario branching tree structure of an interactive story. In this case, each branching scenario may be treated as a document node, similar to the program code in FIG. 6, etc. As another example, the "hierarchical dataset" may be a tree structure of game scenario patterns. As an example, the actors and use cases in the design information tree in FIG. 4A may be replaced with, for example, Scenario 1 and Scenario 2, etc.

[0080] As another example of the information processing method according to the embodiment, the "hierarchical data set" may be any hierarchical data set in the industrial design field or the architectural design field. For example, in the industrial field, parts may be assembled by linking each layer to a BOM (Bill of Materials). Here, BOM refers to a list of parts, a bill of materials, or a bill of materials configuration in the manufacturing industry, etc., required to manufacture a product.

[0081] In the information processing method according to the embodiment, the data format included in the "hierarchical data set" is not limited to text data. The large-scale language model server 2 can employ any large-scale language model capable of generating non-text data (e.g., images, audio, music, videos, or 3D data (3D models, 3D scenes)). In this case, the design prompt may include a prompt for generating any non-text data, and any non-text data may be included in the "hierarchical data set." The above-mentioned 3D data may include, for example, three-dimensional CG and CAD (Computer Aided Design).

[0082] Furthermore, although an example of the information processing method according to the embodiment has been described in which a program code for system design is generated, the present invention is not limited to this. The information processing method according to the embodiment can also be used to generate images (objects) such as still images, videos, CG (computer graphics), CAD, etc. For example, in the case of CG, elements (e.g., objects, textures, light sources, cameras, etc.) of a scene graph (which shows the relationships between elements in an image) can be considered as nodes, and these elements are given relationships. By considering these "elements" as "layers" as described in the embodiment and creating each prompt in the manner described in the embodiment, an image (object) can be generated. By linking various libraries, such as textures, animations, etc., to the configuration of this modified example, a highly convenient image (object) creation application can be constructed.

[0083] The information processing method, program, and information processing system of the embodiment can be applied to the variations of the "hierarchical data set" listed above. Specifically, the sheet display step S1 to the reference step S6 in FIG. 3 can be applied to each of these variations. In each of these variations, the design instruction D 0-1 However, it may indicate that multiple prompts constitute "one cohesive data set."

[0084] In the embodiments, various hierarchical patterns of prompt passing are assumed, and various patterns can be adopted for passing prompts. As a first example, a simple list of natural language character strings may be passed. As a second example, a combination of Markdown notation and a command statement that specifies what to do may be passed. As a third example, a command that makes use of the hierarchical structure of the JSON format may be passed as is. As a fourth example, an image, vector data, audio, 3D objects, motion data, or interaction may be passed. In this fourth example, the storage location may be described in a list. As a fifth example, two or more of the first to fourth examples above may be combined.

[0085] Various embodiments are exemplified below, and the embodiments shown below can be combined with each other. [Appendix 1] 1. An information processing method for generating a hierarchical data set, comprising: The method includes a receiving step, an output step, and an acquisition step, In the receiving step, a design prompt is received; the design prompts have instructions for generating the data set and are comprised of a plurality of prompts arranged in a plurality of layers; the prompts of adjacent layers are related to one another; The output step outputs the design prompt to a large-scale language model; An information processing method, wherein the acquiring step acquires the dataset from the large-scale language model.

[0086] [Appendix 2] 10. The information processing method according to claim 1, An information processing method, wherein the dataset includes program code for constructing a system.

[0087] [Appendix 3] 10. The information processing method according to claim 1 or 2, further comprising a referencing step; The reference step includes displaying reference information; An information processing method, wherein the reference information indicates corresponding portions between components of the data set generated in the obtaining step and the prompts that constitute the design prompt.

[0088] [Appendix 4] An information processing method according to any one of Supplementary Note 1 to Supplementary Note 3, The method further includes a sheet display step, In the sheet display step, a sheet for receiving input of a character string is displayed; The sheet is divided into input areas according to the layers, In the receiving step, the prompt is received based on the content of the sheet.

[0089] [Appendix 5] An information processing method according to any one of Supplementary Note 1 to Supplementary Note 4, further comprising a prompt generating step; The information processing method, wherein the prompt generating step generates the prompt at a lower level based on the prompt at an upper level.

[0090] [Appendix 6] 6. The information processing method according to claim 5, In the prompt generating step, the prompt in the lower layer is generated by outputting the prompt in the upper layer to the large-scale language model.

[0091] [Appendix 7] An information processing method according to any one of Supplementary Note 1 to Supplementary Note 6, In the output step, the prompts are output to the large-scale language model in order from the top layer.

[0092] [Appendix 8] An information processing method according to any one of Supplementary Note 1 to Supplementary Note 7, The design prompt includes instructions expressed in Markdown or Markup format.

[0093] [Appendix 9] An information processing method according to any one of Supplementary Note 1 to Supplementary Note 8, the design prompt includes design instructions; The information processing method, wherein the design instructions indicate that data generated by the large-scale language model based on the plurality of prompts that make up the design prompt constitute a single coherent data set.

[0094] [Appendix 10] A program for causing a computer to execute the information processing method according to any one of Supplementary Note 1 to Supplementary Note 9.

[0095] [Appendix 11] An information processing system comprising at least one information processing device that executes each step of the information processing method according to any one of Supplementary Notes 1 to 9.

[0096] Although the embodiments have been described above, they are presented as examples and are not intended to limit the scope of the invention. The novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made. The embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0097] 1: Information processing equipment 2: Large-scale language model server 6: Communication network 10: Communications Department 11: Storage section 12: Control section 13: Information output section 14: Input section 15: Communication bus 20a: Design prompt (first design prompt) 20b: Design prompt (second design prompt) 30: Program code group 40: Reference information (highlighted) 100: Information Processing Systems 121: Sheet display area 122: Reception 123: Prompt generation section 124: Output section 125: Acquisition section 126:Reference section L1, L2, L3, L4, L5, L6: Layers D0: Design information (system overview) D 0-1 :Design information (design instructions) D 1-1 , D 1-2 : Design information (actor) D 2-1 , D 2-2 : Design information (use case) D 3-1 ~D 3-3 :Design information (screen) D 4-1 ~D 4-3 :Design information (screen elements) D 5-1 ~D 5-3 :Design information (function) D 6-1 ~D 6-3 : Design information (tables and columns) PRM0: Prompt (System Overview Prompt) PRM 0-1 : Prompt (design instruction prompt) PRM 1-1 , PRM 1-2 : Prompt (Actor) PRM 2-1 , PRM 2-2 :Prompt (use case) PRM 3-1 ~PRM 3-3 : Prompt (screen) PRM 4-1 ~PRM 4-3 :Prompt (screen element) PRM 5-1~PRM 5-3 :prompt(function) PRM 6-1 ~PRM 6-3 :Prompt(table column) PRM ex1 ~PRM ex4 :Prompt example PC 3-1 ~PC 3-3 :Program code (screen) PC 4-1 ~PC 4-3 : Program code (screen elements) PC 5-1 ~PC 5-3 :Program code (function) PC 6-1 ~PC 6-3 :Program code (tables and columns) ST: Seat MC: Mouse cursor U: User

Claims

1. 1. An information processing method for generating a hierarchical data set, comprising: The method includes a receiving step, an output step, and an acquisition step, In the receiving step, a design prompt is received; the design prompts have instructions for generating the data set and are comprised of a plurality of prompts arranged in a plurality of layers; the prompts of adjacent layers are related to one another; The output step outputs the design prompt to a large-scale language model; An information processing method, wherein the acquiring step acquires the dataset from the large-scale language model.

2. 2. The information processing method according to claim 1, An information processing method, wherein the dataset includes program code for constructing a system.

3. 3. The information processing method according to claim 1 or 2, further comprising a referencing step; The reference step includes displaying reference information; An information processing method, wherein the reference information indicates corresponding portions between components of the data set generated in the obtaining step and the prompts that constitute the design prompt.

4. 3. The information processing method according to claim 1 or 2, The method further includes a sheet display step, In the sheet display step, a sheet for receiving input of a character string is displayed; The sheet is divided into input areas according to the layers, In the receiving step, the prompt is received based on the content of the sheet.

5. 3. The information processing method according to claim 1 or 2, further comprising a prompt generating step; The information processing method, wherein the prompt generating step generates the prompt at a lower level based on the prompt at an upper level.

6. 6. The information processing method according to claim 5, In the prompt generating step, the prompt in the lower layer is generated by outputting the prompt in the upper layer to the large-scale language model.

7. 3. The information processing method according to claim 1 or 2, In the output step, the prompts are output to the large-scale language model in order from the top layer.

8. 3. The information processing method according to claim 1 or 2, The information processing method, wherein the design prompts include instructions expressed in a Markdown format or a Markup format.

9. 3. The information processing method according to claim 1 or 2, the design prompt includes design instructions; The method of claim 1, wherein the design instructions indicate that data generated by the large-scale language model based on the plurality of prompts that make up the design prompt constitutes a single coherent data set.

10. A program for causing a computer to execute the information processing method according to claim 1 or 2.

11. 3. An information processing system comprising at least one information processing device that executes each step of the information processing method according to claim 1.

Images