Problem-solving support device, method, and system

Abstract

A problem-solving support device is provided. [Solution] A computer inputs into a large-scale language model task text indicating the task to be solved and task decomposition / abstraction instruction prompts containing instructions to decompose the task text into task components and generalize and abstract specific information, and executes a process of obtaining abstracted task structure text as output from the large-scale language model.

Description

[Technical Field]

[0001] The present invention relates to a problem-solving support device, method, and system that support the creation of solutions to technical problems. [Background technology]

[0002] Various technologies have been proposed to support the human idea generation process. For example, Patent Document 1 discloses a system that supports analogical idea generation by vectorizing the idea content (text) entered by the user and searching for and presenting similar cases from a pre-stored database of thought cases. This system promotes analogical recall from different fields by removing field components from the vector data while taking field information into account. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2023-92090 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the idea generation support system disclosed in Patent Document 1 presents analogy information based on the similarity in vector space between the input idea text and thought cases in a database. However, it does not have the functionality to structurally decompose and analyze the technical problem to be solved or to systematically identify structural patterns that are effective in solving the problem. As a result, it can be difficult to search for and apply more accurate knowledge from different fields based on the essential structure of the problem. Furthermore, when a user loses track of the direction of their idea or when they have not considered it from multiple perspectives, there are insufficient specific support measures to complement this and encourage new insights.

[0005] The present invention has been made in consideration of these conventional problems, and aims to provide a new problem-solving support device, method, and system that utilizes large-scale language models (LLMs) to effectively and efficiently support problem-solving. [Means for solving the problem]

[0006] The present invention relates to a method for generating a large-scale language model using a problem text indicating a problem to be solved and a method for generating a large-scale language model using the problem text. The problem includes at least one of a purpose function indicating the purpose of the problem, a realization condition indicating the condition for realizing the purpose function, and a constraint condition that must be observed to satisfy the realization condition. Break down the problem into its components, Included in the assignment text Generalizing specific information Issues This is a problem-solving support device that inputs a problem decomposition / abstraction instruction prompt including instructions for abstraction, and executes a process of obtaining an output of abstracted problem structure text from a large-scale language model.

[0007] The present invention relates to a method for generating a large-scale language model using a problem text indicating a problem to be solved and a method for generating a large-scale language model using the problem text. The problem includes at least one of a purpose function indicating the purpose of the problem, a realization condition indicating the condition for realizing the purpose function, and a constraint condition that must be observed to satisfy the realization condition. Break down the problem into its components, Included in the assignment text Generalizing specific information Issues This is a problem-solving support method that inputs a problem decomposition / abstraction instruction prompt that instructs abstraction, and executes a process of obtaining an output of abstracted problem structure text from a large-scale language model.

[0008] The present invention is a problem-solving support system in which a problem-solving support device that functions as a SaaS management server, a user terminal, and an LLM processing device that executes processing using a large-scale language model and manages an intermediate product DB are connected via a network, and the problem-solving support device receives problem text indicating a problem to be solved from the user terminal, and The problem includes at least one of a purpose function indicating the purpose of the problem, a realization condition indicating the condition for realizing the purpose function, and a constraint condition that must be observed to satisfy the realization condition. Break down the problem into its components, Included in the assignment text Generalizing specific information Issues This is a problem-solving support system that inputs problem decomposition and abstraction instruction prompts, including instructions for abstraction, into a large-scale language model of an LLM processing device, obtains abstracted problem structure text output from the large-scale language model of the LLM processing device, and transmits it to the user's terminal. [Effects of the Invention]

[0009] The problem-solving support device, method, and system of the present invention utilizes the language processing capabilities and knowledge base of a large-scale language model (LLM) to decompose and abstract problems, thereby contributing to problem solving. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a schematic diagram of an overall configuration of a problem solving support system according to an embodiment; [Figure 2] 1 is a block diagram showing a hardware configuration of a problem solving support apparatus according to an embodiment; [Figure 3] 10 is a flowchart of a problem solving support process according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, a problem solving support apparatus, method, and system according to an embodiment will be described in detail with reference to the drawings.

[0012] (System Configuration) Fig. 1 is a schematic diagram of the overall configuration of a system including a problem solving support device 1, an LLM processing device 2, and a user terminal 3 according to an embodiment. Fig. 1 shows an example of a three-tier architecture in which a user terminal 3 (client device), a problem solving support device 1 (which serves as a SaaS management server), and an LLM processing device 2 (which serves as a processing device that executes LLM and manages an intermediate product DB) are connected via a network.

[0013] The problem solving support device 1 is an information processing terminal such as a personal computer, tablet terminal, or smartphone, and is communicably connected to an LLM processing device 2 having a large-scale language model (LLM) via a network (such as the Internet or a LAN).

[0014] The problem solving support device 1 has the function of accepting input from a user terminal 3, sending a processing request (prompt) to an LLM processing device 2 (server), receiving a response from the LLM processing device 2 and presenting it on the user terminal 3. The LLM processing device 2 is a computer that executes LLM, and performs natural language processing (decomposition and abstraction of problems, identification of structural patterns, generation of analogy information, creation of solutions, context conversion, etc.) in response to a request from the problem solving support device 1, and returns the results to the problem solving support device 1.

[0015] Note that, although this embodiment illustrates a configuration in which the main processing by the LLM is executed by an external LLM processing device 2, the present invention is not limited to this. For example, the problem solving support device 1 itself may be equipped with an LLM (local LLM, such as Gemma) and all processing may be completed within the device. In this case, the LLM processing device 2 is not required. Alternatively, a configuration may be adopted in which some processing is performed by the problem solving support device 1 and other processing is shared by the LLM processing device 2. Furthermore, the user may input and output data to the user via the input / output unit 107 of the problem solving support device 1, thereby eliminating the need for a user terminal 3.

[0016] The user terminal 3 is responsible for allowing the user to input problem text and prompts and send them to the problem-solving support device 1, which then accepts requests from the user terminal 3, requests processing from the LLM processing device 2 as necessary, receives the output results generated by the LLM processing device 2, and sends them to the user terminal 3, among other processes.

[0017] The problem-solving support device 1 is a server that performs functions such as sending and receiving data to and from the user terminal 3, managing user sessions (including issuing and managing session IDs), providing a user interface or linking with the UI on the user terminal 3, and saving and managing processing logs.

[0018] The LLM processing device 2 is a server that executes a large-scale language model (LLM) and performs major processing steps such as problem decomposition and abstraction processing, structural pattern identification processing, and cross-disciplinary solution search processing. The LLM processing device 2 may also have a database function (intermediate product DB) that stores and manages intermediate products (abstracted problem structure text, identified structural pattern text, cross-disciplinary analogy information text, created solutions, etc.) generated during these processing steps, in association with session IDs.

[0019] In this way, the main problem-solving process and management of intermediate products are performed by the LLM processing device 2, and the problem-solving support device 1 is responsible for the interface with the user and session management, etc., making it applicable to a so-called SaaS-type provision format.

[0020] The number of user terminals 3 is not limited to one, and may be multiple. The user terminal 3 has a network interface that enables wireless communication, and enables each function by loading a program stored in a storage medium into memory and running the program on a CPU. The user terminal 3 is, for example, a smartphone, tablet, or laptop computer. The configuration of the control system of the user terminal 3 is publicly known, so a description thereof will be omitted.

[0021] In the configuration shown in FIG. 1, the LLM processing device 2 may have, in addition to a problem processing function using a large-scale language model, a database function (DB function) that stores a session ID for each user of the user terminal 3 in association with intermediate and final products, such as abstracted problem structure text, identified structural pattern text, analogy information text, and generated solutions, generated in that session. This DB function allows users to later refer to or reuse processing results from past sessions via the problem-solving support device 1. The LLM processing device 2 can also read out past abstracted problem structure text, etc., associated with this session ID from the database unit (e.g., a storage area equivalent to HDD 105), and use it as re-learning data for the LLM or as part of the input for a new problem-solving process.

[0022] The problem-solving support device 1 may be connected to an LLM processing device 2 as shown in Fig. 1, and the LLM processing device 2 may perform natural language processing using a large-scale language model and problem-solving support using an intermediate product DB. Note that it is still possible for the problem-solving support device 1 to incorporate an LLM and manage the intermediate product DB (an advanced version of the configuration in Fig. 1, or a configuration in which all functions are integrated into the problem-solving support device 1 itself).

[0023] Examples of LLMs provided by the LLM processing device 2 include the GPT series (ChatGPT, GPT-4o, etc.) from OpenAI, the Gemini series from Google, and the Claude series from Anthropic, but are not limited to these, and any LLM capable of executing each processing step described in the embodiment can be used.

[0024] (Hardware configuration) 2 is a block diagram showing the hardware configuration of the problem solving support device 1 according to an embodiment. The problem solving support device 1 is configured as a general computer system including a control unit 104, a storage unit (ROM 102, RAM 103, HDD 105, etc.), a communication interface (external I / F) 106, an input / output unit 107, etc., all of which are connected via a system bus 108. The user terminal 3 and the LLM processing device 2 can also be configured as similar general computer systems.

[0025] The control unit 104 is made up of a CPU 101 and the like, and controls the operation of the entire device by loading a problem-solving support program stored in a ROM 102 or HDD 105 into a RAM 103 and executing it, thereby executing the problem-solving support process described below. The ROM 102 stores basic control programs and the like.

[0026] The RAM 103 is used as a work area for the CPU 101. The HDD 105 (or another non-volatile storage device such as an SSD) stores programs and data generated during processing (for example, session management information and log information in the problem-solving support device 1). A database (DB) unit for storing intermediate products (input problem text, abstracted problem structure text, structural pattern information (for example, a predefined structural pattern list or internal knowledge of a large-scale language model), identified structural pattern text, analogy information text, and created solutions) in the configuration of FIG. 2 is primarily constructed in the storage unit of the LLM processing device 2. In particular, the storage unit of the LLM processing device 2 is also used to associate and store the session ID for each user and various generated texts.

[0027] The communication interface 106 is an interface for communicating with external devices such as the user terminal 3 and the LLM processing device 2 via a network. The input / output unit 107 includes input devices such as a keyboard, mouse, touch panel, and microphone that accept input of problem text and instructions from the user, and output devices such as an LCD display and speaker that display and output processing results and guidance to the user. In the configuration of Figure 1, the user terminal 3 is primarily responsible for input / output functions, and the problem-solving support device 1 exchanges data with the user terminal 3 and the LLM processing device 2 via the communication interface 106.

[0028] (Problem solving support process flow) Next, the overall flow of the problem solving support process of this embodiment will be explained using the flowchart in Figure 3. In this process, in response to a request from the user terminal 3, the problem solving support device 1 interfaces with the user and manages the session, and the LLM processing device 2 executes the main processing steps using the LLM, with the two working together to advance the process.

[0029] [STEP 1: Enter assignment text] First, the user inputs problem text, which describes the technical problem they want to solve in natural language, into the problem solving support device 1 via the user terminal 3. The problem solving support device 1 accepts the input problem text, sends it to the LLM processing device 2 as needed, and stores it in the memory unit of the LLM processing device 2 (an area equivalent to the HDD 105, etc.).

[0030] An example of the assignment text is shown below. "With regard to the new delivery drone we are developing, the SwiftDeliverer Mark II, we would like to extend the battery flight time by 30% compared to the current model, while maintaining the maximum weight of the cargo that can be carried, and reduce the overall manufacturing cost of the drone by 10%."

[0031] At this stage, the problem solving support device 1 can also instruct the LLM in the LLM processing device 2 to generate a general answer to the input problem text and present the result to the user via the user terminal 3.

[0032] [STEP 2: Problem decomposition and abstraction] Next, the control unit of the LLM processing device 2 executes problem decomposition and abstraction processing based on the problem text received in STEP 1 via the problem solving support device 1. Specifically, the problem solving support device 1 inputs the problem text received in STEP 1 from the user terminal 3 and a problem decomposition and abstraction instruction prompt (which may be input from the user terminal 3 or may be prepared in advance and stored in the device; the same applies below) as input data to the LLM in the LLM processing device 2 via the communication interface 106. The LLM in the LLM processing device 2 executes processing based on the received information, generates abstracted problem structure text, and outputs it to the problem solving support device 1. The problem solving support device 1 transmits the received abstracted problem structure text to the user terminal 3 for display. The LLM processing device 2 stores the generated abstracted problem structure text in its own DB.

[0033] The problem decomposition and abstraction prompt instructs the user to explicitly decompose the problem text into its constituent parts, such as "purpose functions," "realization conditions," and "constraints," and then generalize and discard specific information (proper nouns, specific numerical targets, highly confidential technical details, etc.), converting (abstracting) the text into a "general problem structure text" to which analogies can be applied.

[0034] An example of the problem decomposition and abstraction prompt is shown below. "Please break this problem down into 'purpose function,' 'realization conditions,' and 'constraints,' and describe it as a general problem that is not dependent on a specific product name (e.g., SwiftDelivererMark II) or specific numerical targets (e.g., 30%, 10%)."

[0035] The LLM in the LLM processing device 2 performs processing based on the received problem text and problem decomposition / abstraction instruction prompt, generates a general problem structure text showing the decomposed / abstracted general problem structure text, and outputs it to the problem-solving support device 1.

[0036] An example of a typical assignment structure text output is shown below. Yes, I understand. I will break down the problem you presented and generalize it. [General assignment structure] Purpose function: - Extending the operational life of a vehicle (e.g., an air vehicle or autonomous vehicle) from its energy source (e.g., battery). At the same time, the vehicle maintains or improves the payload (maximum weight of cargo, etc.) that it can transport or carry. -Furthermore, to reduce the costs involved in manufacturing the vehicle. (Summary: Simultaneous achievement of extending operational time, maintaining / improving payload, and reducing costs) Eligibility: -Increasing the energy density of energy sources or improving energy consumption efficiency. -Lightweighting of aircraft structures and components (subject to payload maintenance constraints). · Improved cost efficiency in manufacturing processes and component sourcing. A means to evaluate and measure operational uptime, payload capability, and manufacturing costs. Design methods and techniques to manage and optimize the trade-offs between these factors (operational time, payload, cost). Constraints: Compliance with safety and reliability standards for vehicles. · Adaptability to the operational environment (e.g. weather conditions, flight area). · Compatibility with existing infrastructure and operational systems. · The level of technology, timeframe and resources available for development and manufacturing. · The characteristics and requirements of the transported object (payload) (e.g. size, shape, temperature control). -Maintaining the aircraft's basic performance (e.g. speed, range, stability).

[0037] The processing unit, which corresponds to the control unit 104 of the LLM processing device 2, stores the received general problem structure text in its own memory unit (DB, which corresponds to the HDD 105, etc.) (linking it to a session ID as necessary), and the problem solving support device 1 displays the text to the user via the user terminal 3. This abstraction makes it possible to proceed with subsequent processing without disclosing confidential information, contributing to open problem solving without the need for an NDA.

[0038] [STEP 3: Structural pattern identification process] Next, the control unit of the LLM processing device 2 executes a structure pattern identification process based on the general problem structure text obtained in STEP 2. In this process, the control unit of the LLM processing device 2 first acquires structure pattern information. This structure pattern information may be acquired, for example, by reading out a structure pattern list stored in advance in a storage unit (DB equivalent to HDD 105, etc.) of the LLM processing device 2, or may be dynamically generated or referenced by the large-scale language model (LLM) based on its internal knowledge.

[0039] A specific example of a structure pattern list as an example of structure pattern information is shown below. 1. Segmentation / Extraction: Divide the problem or object into parts and remove unnecessary parts 2. Merging / Consolidation: Combining multiple functions or systems into one 3. Dynamization: Making systems and elements change and adapt according to the situation 4. Self-service / Self-organization: The system adjusts and optimizes itself autonomously. 5. Intermediary: Solving problems by introducing intermediate elements or processes 6. Parameter Change Changes): Changes parameters such as physical state, density, temperature, flexibility, etc. 7. Feedback: Monitor outputs and results and adjust inputs and processes 8.Spatial separation Separation): Satisfying conflicting demands through spatial separation 9. Temporal Separation Separation: Satisfying conflicting demands by separating them in time 10. Local Quality: Different, optimal conditions for each part 11. Asymmetry: Making a symmetrical structure or process asymmetrical 12. Transformation: Changing an effect or energy into another useful form

[0040] Next, the control unit of the LLM processing device 2 sends the general problem structure text obtained in STEP 2, the obtained structural pattern information (e.g., the read structural pattern list), and the structural pattern identification prompt as input data to its own LLM. This structural pattern identification prompt instructs the LLM to select, based on the obtained structural pattern information, the structural pattern that is thought to best represent the general problem structure text obtained in STEP 2 or that is thought to suggest a direction for solving the problem.

[0041] An example of a structural pattern identification prompt is shown below. Please select up to three structural patterns from the [Structural Pattern List] below that you think best represent the above [General Problem Structure] (simultaneous achievement of extending operation time, maintaining / improving payload, and reducing costs = using the output content of STEP 2 above as arguments) or that suggest a direction for solving the problem, and for each, briefly explain why you think it is relevant.

[0042] The LLM in the LLM processing device 2 executes processing based on the received general problem structure text, structural pattern information, and structural pattern identification instruction prompt, identifies highly relevant structural patterns, and outputs the results (identified patterns and the reasons for their identification) as identification result text to the problem solving support device 1. The problem solving support device 1 sends the received identification result text to the user terminal 3 for display. The LLM processing device 2 stores the generated identification result text in its own DB.

[0043] An example of the identification result text output is shown below. Yes, I understand. Please select three structural patterns that you believe are most relevant to the given [general problem structure] and explain why. 1. Parameter Change Reason for Changes: This problem deals with the trade-off between multiple important performance parameters (operational time, payload, manufacturing cost) and attempts to improve them simultaneously. In particular, "improving the energy density of the energy source," "reducing weight," and "improving cost efficiency" aim to directly change physical characteristics and economic parameters, and this pattern is considered to be one of the core approaches to solving the problem. 2. Local Quality / Segmentation Reason: A mobile object (drone) is composed of many parts and subsystems (e.g., batteries, motors, structural materials, control systems). In order to simultaneously satisfy multiple requirements such as "extending operation time," "maintaining payload," and "reducing costs," it is thought that it is necessary to apply optimal design and material selection (locally optimized qualities) to each part and subsystem according to its respective role and characteristics. This is also related to the approach of separating the problem into elements and optimizing them individually. 3. Intermediary: Multiple objective functions (long flight time, payload maintenance, low cost) may be in conflict with each other (in a trade-off relationship). In order to achieve these conflicting requirements, it is expected that intermediate processes to balance the requirements, such as "design methods and technologies to manage and optimize trade-offs," or the introduction of intermediary elemental technologies such as energy management systems will be necessary. Therefore, this pattern is also considered highly relevant.

[0044] The processing unit, which corresponds to the control unit 104 of the LLM processing device 2, stores the received identification result text in its own memory unit (DB, which corresponds to the HDD 105, etc.), and the problem solving support device 1 displays the text (linked to a session ID as necessary) to the user via the user terminal 3. This step clarifies the essential structure of the problem, narrowing down the direction of the solution search.

[0045] [STEP 4: Cross-disciplinary solution search process (analogy information generation)] Next, the control unit of the LLM processing device 2 executes a process to search for and generate solutions (analogy information text) with a similar structure in a technical field different from that to which the original problem belongs, based on the identification result text obtained in STEP 3. The control unit of the LLM processing device 2 inputs the identification result text obtained in STEP 3, the general problem structure text obtained in STEP 2, and the cross-field solution search prompt as input data into its own LLM.

[0046] This prompt to search for solutions across different fields instructs the user to present multiple examples (analogy information texts) of solutions to similar problem structures from different fields that are considered to be technologically distant, along with their basic concepts (principles and approaches), based on the structural patterns contained in the identified result text and the contents of the general problem structure text.

[0047] An example of the input for the cross-disciplinary solution search prompt is shown below. In STEP 3, "parameter change" and "local properties / separation" were identified as particularly important structural patterns related to problem solving. For an example of using these patterns to solve the problem structure of "simultaneous achievement of extending operation time (improving energy efficiency), maintaining / improving payload (reducing weight), and reducing costs," please list three candidate fields from which structurally similar solutions can be expected, among different fields that are considered to be relatively distant in terms of technology, and briefly explain the similar solution in each field (analogy information text) and its basic concept (principle or approach).

[0048] The LLM in the LLM processing device 2 searches for and extracts or generates highly relevant solution examples from different fields from its knowledge base in accordance with the received identification result text obtained in STEP 3, the general problem structure text obtained in STEP 2, and the cross-field solution search prompt, and outputs this as analogy information text indicating analogy information to the problem solving support device 1. The problem solving support device 1 sends the received analogy information text to the user terminal 3 for display. The LLM processing device 2 stores the generated analogy information text in its own DB.

[0049] An example of analogy information text output is shown below. Using the patterns of "parameter change" and "local properties / separation," I will list three candidates from different, technologically distant fields that offer similar solutions to the problem structure of "simultaneous achievement of extending operation time (improving energy efficiency), maintaining / improving payload (reducing weight), and reducing costs," and explain the approach in each field. Candidate field 1: Biology (especially the energy strategies of migratory birds and deep-sea organisms) Basic concept: Living organisms have perfected the optimization of energy efficiency and resources (body weight, volume) in the struggle for survival. In particular, strategies for long-distance migration and survival in extreme environments are full of hints for solving engineering problems. Similar solutions / approaches: (Parameter changes): Optimization of flight / swimming methods (flapping vs. gliding, utilizing currents), adjustment of metabolic rate, accumulation of high-energy-density fat and its efficient use. (Local properties / separation): Achieving both lightweight and high performance through the hollow structure of bones and optimization of the functions of wings / fins by region, and concentrating resources on necessary organs. Implications for drone challenges: Applying the energy management strategies and lightweight, high-strength structural principles of living organisms could lead to breakthroughs in energy efficiency and aircraft design. Candidate Field 2: Sports Science (especially endurance athletes and equipment development) Basic Concept: Athletes must maximize their performance (time, distance, weight) within their limited physical capabilities (energy, strength) while simultaneously reducing the risk of injury (cost). Equipment development faces a similar challenge. Similar solutions / approaches: (Parameter changes): Improvement of physiological parameters (maximum oxygen uptake, lactate threshold, etc.) through training, optimization of energy replenishment through nutritional strategies, pacing strategies, and improvement of the aerodynamic characteristics and restitution coefficient of equipment. (Localized properties / isolation): Focused training of specific muscle groups, optimization of body parts to suit sport characteristics, changing materials in different parts of equipment (e.g. sole structure of running shoes, regional stiffness of bicycle frames). Implications for drone issues: Athletes' energy management (output adjustment), ability improvement through training, and approaches to balancing weight reduction, high performance, and cost in equipment development could potentially be applied to drone operation strategies and design. Candidate field 3: Social systems and logistics (e.g., supply chain optimization, urban transportation systems) Basic concept: The goal is to efficiently move and allocate resources (things, people, information, energy) in complex networks, maximizing overall throughput (efficiency) while minimizing costs (time, money, environmental impact). Similar solutions / approaches: (Parameter change): Optimization of transport routes and schedules, inventory management based on demand forecasts (parameter adjustment), optimal combination of transport modes (truck, rail, ship, plane). (Local characteristics / separation): Efficiency through hub-and-spoke networks (concentration of functions in specific locations), optimal placement of distribution centers by region, and changes to traffic control based on time of day. (Intermediation): Efficient information sharing platform, use of standardized containers and pallets. Implications for drone challenges: This could be useful not only for improving the performance of individual drones, but also for optimizing energy efficiency, payload efficiency, and operational costs for entire operational systems in which multiple drones work together (e.g., optimal charging station placement, dynamic planning of delivery routes, cooperative flight).

[0050] The processing unit, which corresponds to the control unit 104 of the LLM processing device 2, stores the received analogy information text in its own storage unit (DB, which corresponds to the HDD 105, etc.) (linking it to a session ID as necessary), and the problem solving support device 1 displays the text to the user via the user terminal 3. At this stage, the problem solving support device 1 can also present the analogy information text to the user and wait for instructions on whether to dig deeper (proceed to the next step). This may provide clues to solutions that would not have been possible with conventional thinking.

[0051] [STEP 5: Solution Creation Process] Next, the control unit of the LLM processing device 2 executes a process to generate solution ideas for the problem text received in STEP 1, based on the general problem structure text obtained in STEP 2 and the analogy information text obtained in STEP 4. The analogy information text referred to here may be one selected by the user from the user terminal 3 via the problem-solving support device 1 from among the multiple analogy information texts returned from the LLM in the LLM processing device 2 in STEP 4, or one that, if no selection is made by the user, is determined to be important and selected by the LLM in the LLM processing device 2 based on predetermined criteria (e.g., score of relevance to the original problem, novelty of the analogy, effectiveness in past similar cases, etc.).

[0052] The control unit of the LLM processing device 2 sends the general problem structure text, the content of the identified analogy information text, the problem text received in STEP 1, and a solution creation instruction prompt as input data to its own LLM. This solution creation instruction prompt instructs the LLM to combine the content of the general problem structure text, the content of the identified analogy information text, and the problem text received in STEP 1 and propose multiple specific solution ideas.

[0053] An example of a solution creation prompt is shown below. To solve the following [General Problem Structure], please propose three specific solution ideas for the original specific problem of "extending drone operation time, maintaining / improving payload, and reducing costs simultaneously," with particular reference to the analogy information presented in STEP 4, [Energy strategies and lightweight structures of living organisms (migratory birds and insects)]. Please provide an explanation for each idea so that it is clear which living organism's strategy or structure it is referencing. [General Problem Structure] (Use the output content of STEP 2 above as an argument)

[0054] The LLM in the LLM processing device 2 integrates the given information in accordance with the received general problem structure text, the content of the identified analogy information text, the problem text received in STEP 1, and the solution creation instruction prompt, generates an created solution text that shows a specific solution idea for the problem text received in STEP 1, and returns it to the problem solving support device 1. The problem solving support device 1 sends the received created solution text to the user terminal 3 for display. The LLM processing device 2 stores the generated created solution text in its own DB.

[0055] An example of the generated solution text output is shown below. Yes, I understand. Using analogies from living creatures (migratory birds and insects), I would like to propose three ideas for solutions to the drone challenge of "extending operation time, maintaining / improving payload, and reducing costs simultaneously." Idea 1: "Biomimicry: High-Density Energy System" Solution Overview: Adopt next-generation high-energy density batteries (e.g., lithium-sulfur batteries, solid-state batteries, etc., aiming for a significant improvement in energy density compared to current batteries) and introduce an energy management system that uses onboard sensors to monitor flight conditions (wind speed, wind direction, aircraft attitude, changes in payload weight, etc.) in real time, dynamically optimizing flight speed, altitude, and motor output to minimize energy consumption. Biological analogies used: High-energy-dense fat: This corresponds to the strategy of migratory birds storing high-calorie fat for long-distance flights. We aim to improve the performance of the energy source itself (parameter change). Optimizing metabolic efficiency and flight strategy: Just as birds use flapping or gliding to take advantage of air currents depending on the situation, drones also adaptively select optimal operating parameters (speed, altitude, etc.) to minimize energy consumption during flight (parameter changes, dynamic characteristics). Idea 2: "Adaptable and lightweight bionic frame structure" Solution Overview: Based on stress analysis using the finite element method (FEM), a lightweight, highly rigid frame structure is designed with an internal hollow structure like bird bones and a reinforcing structure like plant vascular bundles. This is then manufactured using metal 3D printing and additive manufacturing technology for composite materials (CFRP, etc.). Furthermore, a multi-material, local optimization design is implemented, which changes the type of material and density of the internal structure between areas that are subject to load (motor mounting points, arm bases, etc.) and areas that are not. Biological analogies used: The hollow structure of bird bones: mimics the structural principles of living organisms, which minimize weight while maintaining strength (local properties, separation). Wing function by area: This is the concept of using the most suitable materials and internal structure for each area of ​​the frame depending on the required strength, rigidity, and aerodynamic characteristics (local properties). Taking manufacturing costs into consideration, the use of expensive materials is kept to a minimum (separation). Idea 3: "Environmental Energy Harvesting & Cooperative Flight System" Solution Overview: Thin-film solar cells are installed on the surface of the drone to obtain supplemental power (parameter changes) during flight. Furthermore, when multiple drones fly in formation, a cooperative flight control algorithm is introduced (mediation) that utilizes each other's wake turbulence to reduce air resistance, like birds flying in a V formation. A navigation system is also incorporated that utilizes weather information to select routes and time periods where tailwinds and updrafts are likely to occur. Biological analogies used: Formation flying and airflow utilization in migratory birds: We will look at collective cooperative behavior to reduce individual energy consumption and strategies to skillfully utilize environmental energy (wind) (mediation, parameter change). (This includes analogies with photosynthesis in plants): Utilizing environmental energy, such as sunlight, is similar to the basic energy acquisition strategy of living organisms (conversion). Although the efficiency of solar cells may be limited due to constraints on payload and cost, they can be used as auxiliary power to extend operating time even slightly. These ideas are inspired by the energy-efficient and lightweight strategies that living organisms have evolved over time, and could offer a new approach to improving drone performance while reducing costs.

[0056] The processing unit, which corresponds to the control unit 104 of the LLM processing device 2, stores the received text of the created solution in its own storage unit (DB, which corresponds to the HDD 105, etc.) (linking it with a session ID as necessary), and the problem solving support device 1 displays the text to the user via the user terminal 3. This allows the abstract analogy to be elevated to a concrete technical idea.

[0057] [STEP 6: Context conversion processing] If necessary, the control unit of the LLM processing device 2 executes a context conversion process to redefine the problem to fit the context of a different field of expertise. For example, if the content of the analogy information text obtained in STEP 4 is related to a specific field of expertise (e.g., biology), this is effective in presenting the problem to experts in that field to obtain further knowledge and cooperation.

[0058] The control unit of the LLM processing device 2 sends the general task structure text, the content of the analogy information text (or the target field of expertise) obtained in STEP 4, and a context conversion instruction prompt as input data to its own LLM. This context conversion instruction prompt specifies the content of the general task structure text and the content of the analogy information text (or the target field of expertise) obtained in STEP 4, and instructs the LLM to redefine (context convert) the task using words and concepts that are easy for experts in that field to understand.

[0059] An example of a context conversion prompt is shown below. For the following [General Problem Structure], please redefine (contextualize) the problem from a biological perspective so that experts (e.g., biologists, ecologists, biomimetics researchers) in [Candidate Field 1: Energy Strategies of Biology (especially Migratory Birds and Deep-Sea Organisms)], which were selected as analogy candidates in STEP 4, can easily perceive it as their own research topic or area of ​​interest. In particular, please express the three goals of "extending operational time (energy efficiency)," "maintaining / improving payload (weight reduction)," and "reducing costs (contribution to survival and reproduction)" in the context of biological survival strategies and evolution, and describe how the overall research topic can be presented. [General Problem Structure] (Use the output content of STEP 2 above as arguments)

[0060] The LLM in the LLM processing device 2 performs context conversion of the task in accordance with the received general task structure text, the content of the analogy information text (or the target field of expertise) obtained in STEP 4, and the context conversion instruction prompt, and outputs a context conversion task definition text showing the result (context converted task definition) to the problem-solving support device 1. The problem-solving support device 1 sends the received context conversion task definition text to the user terminal 3 for display. The LLM processing device 2 stores the generated context conversion task definition text in its own DB.

[0061] An example of the output of the context transformation task definition text is shown below. Yes, I understand. I will recontextualize the proposed [general problem structure] for experts in biology (especially the energy strategies of flying and moving organisms) and redefine it as a research problem. [Problem definition in the context of biology (evolution and ecology)] Research topic title (example): Optimization of resource acquisition and allocation strategies in mobile organisms: Trade-offs between energy efficiency, structural constraints, and survival and reproductive success ·Problem summary: Under natural selection, organisms that travel long distances (e.g., migratory birds, fish, and flying insects) and those that operate in resource-limited environments (e.g., deep-sea organisms) face severe optimization problems related to energy budget and body structure to maximize their survival and reproductive success. This research project aims to elucidate how these organisms have integrated the following three conflicting requirements and acquired evolutionarily stable strategies: Maximizing activity time (energy efficiency): How can we maximize the total time and distance we can operate by efficiently acquiring and storing the energy needed for activities such as foraging, movement, and reproduction (e.g., high-density fat deposits) while minimizing energy expenditure during operation (e.g., metabolic rate, movement costs)? (This corresponds to the engineering challenge of "extending operational time.") Minimizing structural and physiological costs (payload and weight reduction): How do we minimize structural and physiological costs (resource requirements) such as body weight and body size while maintaining or improving mobility (e.g., flight, swimming) and necessary functions (e.g., sensory organs, predation and prey avoidance abilities, reproductive organs = payload)? What are the optimization strategies for skeletal structure, muscle placement, and organ size to reduce weight? (This corresponds to the engineering challenges of "maintaining / improving payload" and "reducing weight.") Maximizing investment efficiency in survival and reproduction (from the perspective of cost reduction): Through the energy efficiency and structural optimization mentioned above, how do they optimally allocate energy and resources throughout their lives to ultimately increase the survival rate of individuals and maximize reproductive success (the number of offspring they can leave behind)? What trade-off solutions have they evolved to address external factors (constraints) such as predation risk and environmental fluctuations? (This corresponds to the engineering challenges of "reducing manufacturing / operating costs" and "overall success.") From a biological perspective, the general problem structure presented can be reinterpreted as a fundamental evolutionary and ecological problem of "optimizing resource allocation and resolving trade-offs." Analyzing the diverse solutions of living organisms (e.g., specific metabolic pathways, morphological adaptations, behavioral strategies) from this perspective may provide new insights into engineering problem solving.

[0062] The processing unit, which corresponds to the control unit 104 of the LLM processing device 2, stores the received context conversion task definition text in its own storage unit (DB, which corresponds to the HDD 105, etc.) (linking it to a session ID as necessary), and the problem-solving support device 1 displays the text to the user via the user terminal 3. This is expected to facilitate smooth communication and collaborative research with experts in different fields.

[0063] By sequentially carrying out the above steps from STEP 1 to STEP 6, users can systematically proceed through a series of processes, including structural analysis, utilizing analogies from other fields, and creating concrete solutions to their technical challenges, with the support of LLM.

[0064] As described above, according to the embodiment, by utilizing the advanced language processing capabilities and knowledge of LLM, solutions to technical problems can be derived effectively and efficiently by executing a step-by-step process including problem decomposition and abstraction (STEP 2), structural pattern identification (STEP 3), cross-disciplinary solution search (STEP 4), solution creation (STEP 5), and context transformation (STEP 6).

[0065] In particular, by structurally grasping the problem (STEP 2) and identifying its essential patterns (STEP 3), it becomes possible to utilize knowledge from different fields at a deeper level, unlike conventional analogy searches based on simple keyword searches or superficial similarities (e.g., Patent Document 1).In identifying these structural patterns, it is possible not only to refer to a predefined list, but also to dynamically recall and generate related structural patterns from the extensive internal knowledge of a large-scale language model, which is expected to enable more flexible and comprehensive utilization of patterns.

[0066] Furthermore, because the LLM actively generates specific solutions (STEP 5), users not only get hints but also specific ideas that they can further develop. Furthermore, there are many practical advantages, such as protecting confidential information through problem abstraction (STEP 2) and promoting cross-disciplinary collaboration through context transformation (STEP 6). Furthermore, by storing the processing results for each user in the database of the LLM processing device 2, the LLM processing device 2 can support continuous problem-solving activities, such as reading previously abstracted problem structure text associated with a session ID from the database and using it for re-learning or processing.

[0067] Although an embodiment of the present invention has been described above, this embodiment is presented as an example and is not intended to limit the scope of the invention. This novel embodiment can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. This embodiment and its modifications are within the scope and spirit of the invention, and are also included in the invention and its equivalents as set forth in the claims. For example, it is possible to partially change the order of the steps, omit certain steps, or add new steps. Furthermore, a user interaction interface may be provided to accept user feedback during each step and incorporate that feedback into the processing. [Explanation of symbols]

[0068] 1 problem solving support device, 2 LLM processing device, 3 user terminal, 101 CPU, 102 ROM, 103 RAM, 104 control unit, 105 HDD (storage unit, DB unit), 106 external I / F (communication interface), 107 input / output unit, 108 system bus

Claims

1. The computer A problem-solving support device that executes a process of inputting into a large-scale language model problem text indicating a problem to be solved and problem components including at least one of a target function indicating the purpose of the problem, a realization condition indicating the conditions for realizing the target function, and a constraint condition that must be observed in order to satisfy the realization condition, and inputting into the large-scale language model problem decomposition / abstraction instruction prompts including instructions to generalize specific information included in the problem text to abstract the problem, and obtaining an abstracted problem structure text output from the large-scale language model.

2. 2. The problem-solving support device according to claim 1, further comprising the steps of: inputting the abstracted problem structure text, structural pattern information, and a structural pattern identification instruction prompt including an instruction to identify a structural pattern related to solving the abstracted problem structure text based on the structural pattern information to the large-scale language model; and obtaining an output of the identified structural pattern text from the large-scale language model.

3. 3. The problem-solving support device according to claim 2, further comprising the steps of: inputting the identified structure pattern text, the abstracted problem structure text, and a cross-field solution search instruction prompt including an instruction to search for cases that solve similar problem structures in technical fields different from the technical field to which the problem text belongs to into the large-scale language model; and obtaining an output of analogy information text in the cross-field from the large-scale language model.

4. 4. The problem-solving support device according to claim 3, further comprising the steps of: inputting the analogy information text, the abstracted problem structure text, the problem text, and a solution creation instruction prompt including an instruction to propose a specific solution idea for the problem text into the large-scale language model; and obtaining an output of a solution idea text from the large-scale language model.

5. 4. The problem-solving support device according to claim 3, further comprising: inputting, into the large-scale language model, the field of expertise to which the analogy information text belongs, the abstracted problem structure text, and a context conversion instruction prompt including an instruction to convert the abstracted problem structure text into a context of the field of expertise; and obtaining an output of a context-converted problem definition text from the large-scale language model.

6. The computer A problem-solving support method that executes a process of inputting problem text indicating a problem to be solved and problem components including at least one of a target function indicating the purpose of the problem, a realization condition indicating the conditions for realizing the target function, and a constraint condition that must be observed in satisfying the realization condition into a large-scale language model, and a problem decomposition / abstraction instruction prompt that instructs the model to generalize specific information contained in the problem text and abstract the problem, and obtaining abstracted problem structure text as output from the large-scale language model.

7. A problem-solving support system in which a problem-solving support device that functions as a SaaS management server, a user terminal, and an LLM processing device that executes processing using a large-scale language model and manages an intermediate product DB are connected via a network, The problem solving support device, A problem-solving support system that executes the process of inputting problem text indicating the problem to be solved received from the user terminal, and decomposing the problem text into problem components including at least one of a target function indicating the purpose of the problem, a realization condition indicating the conditions for realizing the target function, and a constraint condition that must be observed in satisfying the realization condition, into a large-scale language model of the LLM processing device, and a problem decomposition / abstraction instruction prompt including an instruction to generalize specific information included in the problem text to abstract the problem, obtaining abstracted problem structure text output from the large-scale language model of the LLM processing device and sending it to the user terminal.

8. The problem solving support device, a process of inputting the abstracted problem structure text, structural pattern information, and a structural pattern identification instruction prompt including an instruction to identify a structural pattern related to a solution of the abstracted problem structure text based on the structural pattern information into a large-scale language model of the LLM processing device, and obtaining an output of the identified structural pattern text from the large-scale language model of the LLM processing device; a process of inputting the identified structure pattern text, the abstracted problem structure text, and a cross-field solution search instruction prompt including an instruction to search for cases that solve similar problem structures in technical fields different from the technical field to which the problem text belongs to into a large-scale language model of the LLM processing device, and obtaining analogy information text in the cross-field from the large-scale language model of the LLM processing device; The LLM processing device The problem-solving support system according to claim 7, further executing a process of storing a session ID for each user of the user terminal, the problem structure text, the structure pattern text, and the analogy information text in association with each other.

Images