Business formation system and business formation method

Abstract

Starting from the concept of a hardware product business, we identify the best trading partners for each component related to this hardware. [Solution] The business composition system 5 includes a business component decomposition unit 22 that receives data on a business plan 42 and decomposes it into elements or parts that make up the business plan 42; a company information extraction unit 24 that, if the elements decomposed by the business component decomposition unit 22 are parts, extracts company information on multiple companies that produce these parts; an LLM index value calculation unit 26a that calculates, using a large-scale language model server 31, LLM index values ​​that evaluate the company information on the multiple companies extracted by the company information extraction unit 24; and a ranking unit 27 that ranks the multiple companies based on the LLM index values.

Description

[Technical Field]

[0001] The present invention relates to a business formation system and a business formation method. [Background technology]

[0002] In recent years, technologies such as generative artificial intelligence (AI) and large language models (LLM) have brought about major changes in industrial structure. This has led to increased market entry by companies from other industries and emerging companies. This dramatic change in industrial structure is likely to transform the hardware product business into a software or application business model. In reality, automobiles are often equipped with communication functions, expanding their functionality as user devices.

[0003] The business model for smartphone application and software businesses is to quickly release prototypes to the market and update them based on market feedback. Just like the smartphone application and software businesses, hardware product businesses are also expected to quickly release prototypes to the market and update them based on market feedback.

[0004] In the product business, unlike the application or software businesses, it is difficult to quickly release prototypes to the market. This is different from the application or software businesses, which can release programs with minimal functionality in stages. In the hardware product business, physical parts are required, and evaluation and collaboration with partner companies that manufacture these parts is necessary.

[0005] Patent Document 1 describes an invention of a system that assists in the selection of companies to purchase from in order processing. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-094727 Summary of the Invention [Problem to be solved by the invention]

[0007] The invention described in Patent Document 1 is effective for selecting a company to purchase from. However, it does not take into consideration the comprehensive selection of companies that manufacture multiple parts required for prototyping in the product business, for example.

[0008] In the product business, it is not possible for one company to manufacture all the parts and turn them into products; rather, collaboration with business partners who have the parts and technologies is necessary. Finding these business partners and holding meetings with the people in charge at each company can take up a huge amount of time.

[0009] Therefore, an object of the present invention is to find the best business partner for each component related to the hardware based on the concept of the hardware product business. [Means for solving the problem]

[0010] In order to solve the above-mentioned problems, the business formation system of the present invention includes a business component decomposition unit that receives data of a business plan and decomposes the business plan into elements or parts that make up the business plan, a company information extraction unit that extracts company information on multiple companies that produce the parts if the elements decomposed by the business component decomposition unit are parts, and an LLM index value calculation unit that calculates LLM index values ​​that evaluate the company information on the multiple companies extracted by the company information extraction unit using a large-scale language model. a rule-based index calculation unit that calculates rule-based indexes that evaluate the company information about the plurality of companies extracted by the company information extraction unit; The LLM index value and the rule-based index value and a ranking unit that ranks the plurality of companies based on the ranking.

[0011] The business formation method of the present invention includes the steps of: a business component decomposition unit receiving data of a business plan and decomposing the data into elements that constitute the business plan; if the elements decomposed by the business component decomposition unit are parts, a company information extraction unit extracting company information on a plurality of companies that produce the parts; and an LLM index value calculation unit calculating, using a large-scale language model, LLM index values ​​that evaluate the company information on the plurality of companies extracted by the company information extraction unit. a step of calculating rule-based index values ​​by a rule-based index value calculation unit evaluating the company information about the plurality of companies extracted by the company information extraction unit; The LLM index value and the rule-based index value and a step in which a ranking unit ranks the plurality of companies based on the ranking. Other means will be described in the detailed description of the invention. [Effects of the Invention]

[0012] According to the present invention, it is possible to find the best business partner for each component of the hardware based on the concept of the hardware product business. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a configuration diagram of a business formation system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram illustrating the physical configuration of a business composition server. [Figure 3] This is the input screen for the Business Model Canvas. [Figure 4] This is a customer candidate list screen. [Figure 5] This is an overview of the Business Model Canvas. [Figure 6] Figure 1 shows an example prompt for creating a business proposal that includes a Business Model Canvas. [Figure 7] FIG. 10 is a diagram illustrating an example of data of a business plan. [Figure 8] FIG. 10 illustrates a prompt to decompose a business into components. [Figure 9] FIG. 10 is a diagram showing data decomposed into elements. [Figure 10] FIG. 10 illustrates a prompt to list elements. [Figure 11] FIG. 10 is a diagram showing data in which each element is listed in a table format. [Figure 12] FIG. 10 illustrates a prompt to extract company information. [Figure 13] FIG. 10 is a diagram showing data in which company information is added to an element table. [Figure 14] FIG. 10 illustrates a prompt to extract the original value. [Figure 15] FIG. 10 is a diagram showing data listing company overviews. [Figure 16A] FIG. 10 is a diagram showing a prompt for instructing to score the reliability as an index value of the company information. [Figure 16B] FIG. 10 is a diagram showing a prompt for instructing to score safety as an index value of company information. [Figure 16C] FIG. 10 is a diagram showing a prompt for instructing to score quality as an index value of company information. [Figure 16D] FIG. 10 is a diagram showing a prompt for instructing to score delivery time as an index value of company information. [Figure 16E] FIG. 10 is a diagram showing a prompt for instructing to score costs as an index value of company information. [Figure 17] FIG. 10 is a diagram showing data obtained by scoring index values ​​of company information. [Figure 18] 10 is a flowchart for scoring index values ​​of company information on a rule basis. [Figure 19] FIG. 10 illustrates a prompt to rank companies by score. [Figure 20] FIG. 10 is a diagram showing data in which companies are ranked in order of score. [Figure 21] FIG. 10 illustrates a prompt to generate a candidate business partner list. [Figure 22] FIG. 10 is a diagram illustrating an example of a candidate supplier list. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this embodiment, an assumed use case of a business owner considering a new business and a business formation system is described.

[0015] FIG. 1 is a configuration diagram of a business formation system 5 according to this embodiment. The business formation system 5 is configured by connecting a user terminal 1, a cloud-based business formation server 2, a large-scale language model server 31, and a transaction information database 33 in a manner that allows communication. The business formation system 5 digs deep into a business plan 42, embeds the company's strengths into the business, and filters out competing companies. In the business development action, the business formation system 5 investigates business partner companies and supports the selection of suitable business partner companies.

[0016] The terminal 1 is, for example, a computer on which a user can run a spreadsheet program or the like, and is configured to include an input unit 11 and a display unit 12. The input unit 11 is, for example, a mouse or a keyboard. A user inputs business model canvas information 41 and indicators 40 via the input unit 11. The display unit 12 is, for example, an LCD display, and displays the candidate business partner list 49 output by the business formation server 2.

[0017] The large-scale language model server 31 is a server equipped with LLMs (Large Language Models), and responds to user inquiries interactively on a web browser screen, as well as to inquiries from other computers by executing requests to API (Application Program Interface) endpoints. Here, the business formation system 5 receives support from the large-scale language models by executing requests to the API endpoints. The company information 32 is information acquired from, for example, the official homepage of each company, as well as a database that compiles information about each company. The transaction information database 33 is a database that stores transaction records between companies.

[0018] The business composition server 2 is configured to include a preprocessing unit 21, a business component decomposition unit 22, a listing unit 23, a company information extraction unit 24, an LLM original value extraction unit 25a, an LLM index value calculation unit 26a, a rule-based original value extraction unit 25b, a rule-based index value calculation unit 26b, a ranking unit 27, and a candidate business partner list output unit 28.

[0019] The preprocessing unit 21 accepts input of business model canvas information 41, converts it into Markdown format, and generates business plan 42 data. The constraints expressed in the business model canvas are treated as constraint conditions. An example of a constraint condition is that only companies that have obtained ISO 9001 certification are eligible. The preprocessing unit 21 converts the data into business plan 42 data using a large-scale language model, or may convert the data into business plan 42 data based on rules, and is not limited to this.

[0020] The business component decomposition unit 22 decomposes the business plan 42 created by the preprocessing unit 21 into its constituent elements or component groups. The decomposed components have a hierarchical structure, making it possible to expand into multiple layers. The business component decomposition unit 22 decomposes the business plan 42 into its constituent elements or component groups using a large-scale language model, or may decompose the business plan 42 into its constituent elements or component groups based on rules, and is not limited to this.

[0021] The listing unit 23 lists the elements decomposed by the business component decomposition unit 22 while maintaining the parent-child relationships of the elements. In other words, the listing unit 23 lists the structures between the elements. The listing unit 23 may list the elements using a large-scale language model or may list the elements on a rule-based basis, and is not limited thereto.

[0022] If the elements listed by the listing unit 23 are parts, the company information extraction unit 24 extracts company information about multiple companies that produce each part. The company information extraction unit 24 extracts company information about companies that produce parts by referencing company information 32, which is the result of a web search, and a transaction information database 33, which stores inter-company transaction records. The company information extraction unit 24 extracts company information using a large-scale language model, but may also extract company information on a rule-based basis, and is not limited to this.

[0023] The LLM original value extraction unit 25a extracts, in text form, the company information, such as sales amount, extracted by the company information extraction unit 24. This makes it possible to embed company information in the prompt.

[0024] The LLM index value calculation unit 26a causes the large-scale language model server 31 to calculate a score for each company based on the given indexes, based on the text information extracted by the LLM original value extraction unit 25a, and outputs company score list data 47a. The company score list data 47a is an index value calculated using a large-scale language model, and is therefore an LLM index value. The types of indexes calculated by the LLM index value calculation unit 26a can also include safety, quality, delivery time, and cost, in addition to reliability.

[0025] An example of calculating the reliability score could be based on sales revenue: if sales revenue is less than 1 trillion yen, the score would be 1, and if sales revenue is between 1 trillion yen and 10 trillion yen, the score would be 2.

[0026] The rule-based original value extraction unit 25b acquires, from among the company information extracted by the company information extraction unit 24, items to be used by the rule-based index value calculation unit 26b. Based on the information acquired by the rule-based original value extraction unit 25b, the rule-based index value calculation unit 26b scores companies using a predetermined logic based on the given indexes and outputs company score list data 47b. The types of indexes calculated by the rule-based index value calculation unit 26b are the same as the types of indexes calculated by the LLM index value calculation unit 26a, and it is also possible to use safety, quality, delivery time, cost, etc. in addition to reliability. Since the rule-based index value calculation unit 26b calculates index values ​​on a rule basis, the company score list data 47b are rule-based index values. The rule-based index value calculation unit 26b normalizes each element, such as the number of transactions and total amount, within the transaction information, and then calculates the reliability score S1 using the following formula (1): S1=X1×K1+X2×K2+X3×K3+X4×K4+X5×K5-Y1×L1…(1) Here, Score: S Original price. Number of transactions: X1 Weighting coefficient for original price and number of transactions: K1 Original price.Total transaction amount:X2 Weighting factor for original price and total transaction amount: K2 Original price. Transaction quantity: X3 Weighting factor for original price and trading volume: K3 Original price. Number of consecutive trading years: X4 Original price. Weighting factor for consecutive trading years: K4 Number of times a company appears in the parts list: X5 Weighting factor for the number of times a company appears in the parts list: K5 Risk factors (payment, delivery delays, etc.): Y1 Risk factor weighting coefficient: L1

[0027] The rule-based index value calculation unit 26b normalizes each element such as the number of transactions and the total amount within the transaction information, and then calculates the safety score S2 using the following formula (2). S2=X 11 ×K 11 +X 12 ×K 12 …(2) Here, Logic value of ISO27001 etc. certification: X 11 Weighting factor for logical value of ISO27001 etc. acquisition: K 11 The logical value is 1 when true and 0 when false. Number of years of business: X 12 Weighting factor for number of years of business: K 12

[0028] The rule-based index value calculation unit 26b normalizes each element such as the number of transactions and the total amount within the transaction information, and then calculates the quality score S3 using the following formula (3). S3=X 21 ×K 21 …(3) Here, Logical value of ISO9001 etc. certification: X 21 Weighting coefficient of logical value for ISO9001 etc. acquisition: K 21

[0029] The rule-based index value calculation unit 26b normalizes each element such as the number of transactions and the total amount within the transaction information, and then calculates the quality score S4 using the following formula (4). S4=X 31 ×K 31 +X 32 ×K 32 … (4) Here, On-time delivery rate: X 31 Weighting factor for on-time delivery rate: K 31 Delivery delay: X days 32 Weighting coefficient for delivery delay days: K 32

[0030] The rule-based index value calculation unit 26b normalizes each element such as the number of transactions and the total amount within the transaction information, and then calculates the cost score S5 using the following formula (5). S5=X 41 ×K 41 … (5) Here, Market deviation of part unit price: X 41 Weighting coefficient for part unit price: K 41

[0031] The ranking unit 27 ranks each company based on the LLM index value calculated by the LLM index value calculation unit 26a and the rule-based index value calculated by the rule-based index value calculation unit 26b. For example, the ranking unit 27 normalizes the LLM index value and the rule-based index value, calculates the average value, and then ranks the companies based on the average value, but the specific method of calculating the index value is not limited to this. This ranking is expressed in output data and serves as reference information for those considering new business ventures when selecting business partners. The ranking unit 27 may rank each company using a large-scale language model or may rank each company based on rules, and is not limited to this.

[0032] The candidate supplier list output unit 28 outputs the results of the ranking unit 27 as a candidate supplier list 49. The candidate supplier list output unit 28 associates the multiple companies ranked by the ranking unit 27 with the structure between each element listed by the listing unit 23, and outputs the list as a candidate supplier list for parts related to the business plan 42. The candidate supplier list output unit 28 may output the candidate supplier list using a large-scale language model, or may output the candidate supplier list on a rule-based basis, and is not limited to this.

[0033] FIG. 2 is a diagram showing the physical configuration of the business composition server 2. As shown in FIG. The business formation server 2 includes a CPU (Central Processing Unit) 201 , a ROM (Read Only Memory) 202 , and a RAM (Random Access Memory) 203 .

[0034] The CPU 201 executes a program (not shown) to realize the functions of the business formation server 2. The ROM 202 is a non-volatile memory and stores, for example, a BIOS (Basic I / O System). The RAM 203 is a volatile memory and stores data temporarily held by the program executed by the CPU 201.

[0035] The business formation server 2 further includes a storage unit 204 , a display unit 205 , an input unit 206 , and a communication interface 207 , which are interconnected by a bus 208 . The storage unit 204 is a large-capacity storage unit such as a hard disk or SSD (Solid State Drive), and stores programs, data, and the like.

[0036] The display unit 205 is, for example, a liquid crystal display, and displays characters, figures, images, etc. The display unit 205 may be detachable. The input unit 206 is, for example, a transparent touch panel superimposed on the liquid crystal display, and accepts user input operations. The communication interface 207 communicates with, for example, a cloud server. The communication interface 207 may support any communication format, such as a wired NIC (Network Interface Card), a wireless NIC, or a fifth-generation communication module.

[0037] FIG. 3 shows the Business Model Canvas input screen 10. The business model canvas input screen 10 is realized as, for example, a worksheet of a spreadsheet software, and displays cells 100 to 109, an indicator combo box 1011, and a business idea creation button 1012. Here, the business model canvas input screen 10 is provided on an input sheet.

[0038] In cell 100, the new business idea is entered as text. In cell 101, information about the main partner (KP) is entered as text. In cell 102, information on the main activity (KA) is entered as text. In cell 103, information on the main resource (KP) is entered as text. In cell 104, value proposition (VP) information is entered as text. In cell 105, customer relationship (CR) information is entered as text. In cell 106, customer segment (CS) information is entered as text. In cell 107, channel (CR) information is entered as text. In cell 108, information on the cost structure (C$) is entered as text. Cell 109 contains revenue stream (RS) information entered as text. In the index combo box 1011, the type of index for evaluating a company is input in a selectable manner. When the business idea creation button 1012 is clicked, the entered business model canvas information 41 is sent to the business composition server 2, and a series of processes begins.

[0039] FIG. 4 shows the supplier candidate list screen 19. The supplier candidate list screen 19 is realized as, for example, a spreadsheet in a spreadsheet software, and displays the supplier candidate list 49 in a table format. This enables the user to find the optimal supplier companies for each component related to the hardware based on the concept of the hardware product business.

[0040] Figure 5 shows an overview of the Business Model Canvas. The Business Model Canvas is a framework or template for visualizing a business model. It is used in situations such as confirming and verifying new businesses and conceptualizing your own business model.

[0041] By utilizing this Business Model Canvas, you can clarify the shape of your new business model and view it objectively. Also, by utilizing it in your existing business, you can gain a deeper understanding of your company's strengths and weaknesses, the business environment in which it operates, and your company's position.

[0042] The Business Model Canvas consists of the following nine building blocks:

[0043] "Customer Segment" is a block where you fill in information about your potential customers. The "Value Proposition" section is a block where you can write down the specific details of your product or service, as well as everything else that customers value, such as price, brand, customer support, and location. "Channel" is a block where you enter the channel through which you sell your products or services.

[0044] "Customer Relationships" is a block where you can write down specific ways to build long-term relationships with people who have become your customers. "Revenue flow" is literally a block where you write down the flow of money from sales, etc.

[0045] "Major Resources" is a block where you can write down the resources required to run your business. The "Major Activities" section is a block where you can write down tasks, which are activities necessary for developing your business.

[0046] The "Major Partners" block is where you can write down the partners with whom you collaborate in developing your business. The "Cost Structure" block is where you enter the costs required to develop your business.

[0047] <<Operation of the pre-processing section>> 6 and 7, a case will be described in which the preprocessing unit 21 executes a request to the API endpoint of the large-scale language model server 31 and performs preprocessing to convert the text into Markdown notation with the support of an LLM. Note that the preprocessing unit 21 is not limited to implementations with the support of an LLM, and may perform processing based on rules, for example.

[0048] FIG. 6 illustrates a prompt 211 for creating a business proposal 42 that includes a business model canvas. The preprocessing unit 21 sends a request in JSON (JavaScript Object Notation) format including this prompt 211 to the API endpoint of the large-scale language model server 31. The prompt 211 is transcribed below.

[0049] prompt # Task Create a Business Model Canvas in Markdown format using the following constraints: # Constraints - Display the business model canvas of the new business idea described in cells 100 to 109 of the input sheet in Markdown format. # Cell 100 ... # Cell 101 ... ... # Cell 109 ...

[0050] The preprocessing unit 21 sends a JSON-formatted request including this prompt 211 to the API endpoint of the large-scale language model server 31. The JSON-formatted request is shown below.

[0051] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 211 string)} ], "temperature": 0.7 }

[0052] FIG. 7 is a diagram showing an example of data for the business plan 42. The preprocessing unit 21 sends a JSON-formatted request including a prompt 211 to the API endpoint of the large-scale language model server 31, and then receives a JSON-formatted response, a business plan 42. An example of the JSON format is shown below.

[0053] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (Proposal 42 text) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": … } }

[0054] Below is a portion of the Markdown text written in Business Proposal 42. Figure 7 shows the result of displaying the Markdown text in a Markdown viewer.

[0055] # Flying cars ## KP - Highly skilled technical team - Manufacturing Facilities and Supply Chain ## KA - Research and Development - Manufacturing and Quality Control ## VP - Time saving - A new travel experience

[0056] <<Operation of the Business Component Decomposition Unit>> 8 and 9 below, a case will be described in which the business component decomposition unit 22 calls the large-scale language model server 31 and decomposes the business into its constituent elements with the support of the LLM. Note that the business component decomposition unit 22 is not limited to implementations supported by the LLM, and may, for example, perform rule-based processing. For the business component decomposition unit 22 to perform rule-based processing, it can, for example, perform morphological analysis of the business plan, derive a hypothetical product from the results of the morphological analysis, and refer to a database showing the hypothetical product and its parts.

[0057] FIG. 8 illustrates a prompt 221 for decomposing a business into components. The business component decomposition unit 22 sends a request in JSON format including this prompt 221 to the API endpoint of the large-scale language model server 31. The prompt 221 is transcribed below.

[0058] prompt # Task Please factorize the following constraints: # Constraints - Items are "Parts Category", "Required Parts", and "Detailed Parts". - The granularity of "detailed parts" is as follows: Example: Detailed parts of the frame: nuts, bolts, etc.

[0059] The business component decomposition unit 22 sends a JSON-formatted request including this prompt 221 to the API endpoint of the large-scale language model server 31. The JSON-formatted request is shown below. Experimental results show that a more appropriate bill of materials can be generated by inputting business model canvas information in addition to the new business idea, rather than simply inputting the new business idea.

[0060] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 211 string)} {"role": "assistant", "content": (Proposal 42 text)} {"role": "user", "content": (Prompt 221 string)} ], "temperature": 0.7 }

[0061] FIG. 9 is a diagram showing data 43 decomposed into elements. The business component decomposition unit 22 sends a JSON format request including a prompt 221 to the API endpoint of the large-scale language model server 31, and then receives data 43 included in the JSON format response. An example of a JSON format response is shown below.

[0062] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (string of data 43) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": … } }

[0063] The Markdown text written in this data 43 is listed below.

[0064] # Below is an example of the decomposition of elements in the development of a flying car. # Part category, required parts, detailed parts ## 1. Frame - Required parts: Main frame, subframe, stiffeners - Detail parts: nuts, bolts, rivets, welding materials ## 2. Engine / Motor - Required parts: engine block, pistons, crankshaft, electric motor - Detail parts: gaskets, bearings, seals, wire harnesses

[0065] Figure 9 shows the result of displaying Markdown text in a Markdown viewer.

[0066] <<Listing section operation>> 8 and 9 below, a case will be described in which the listing unit 23 sends a prompt to the API endpoint of the large-scale language model server 31 to list the elements that make up a business with the support of an LLM. Note that the listing unit 23 is not limited to implementations supported by an LLM, and may perform rule-based processing, for example. The method by which the listing unit 23 performs rule-based processing involves determining which item each line of text written in Markdown corresponds to and listing the items.

[0067] FIG. 10 shows a prompt 231 for listing elements. The listing unit 23 sends a request in JSON format including this prompt 231 to the API endpoint of the large-scale language model server 31. This prompt 231 is transcribed below.

[0068] prompt # Task Please display the following constraints in a table format: # Constraints - The items in the table are, from left to right, "Parts Category," "Necessary Parts," and "Detailed Parts." - Each detailed part should be listed on a separate line.

[0069] The listing unit 23 sends a JSON-formatted request including this prompt 231 and past system and user messages in a message array to the API endpoint of the large-scale language model server 31. The JSON-formatted request is shown below.

[0070] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 211 string)} {"role": "assistant", "content": (Proposal 42 text)} {"role": "user", "content": (Prompt 221 string)} {"role": "assistant", "content": (string from data 43)} {"role": "user", "content": (Prompt 231 string)} ], "temperature": 0.7 }

[0071] FIG. 11 is a diagram showing data 44 in which each element is listed in a table format. The listing unit 23 sends a JSON-formatted request including a prompt 231 to the API endpoint of the large-scale language model server 31, and then receives data 44 included in the JSON-formatted response. An example of a JSON-formatted response is shown below.

[0072] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (string of data 44) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": …} }

[0073] The Markdown text written in this data 44 is listed below.

[0074] # Below is a tabular breakdown of the elements involved in developing a flying car. | Parts Category | Required Parts | Part Details | |----------------|--------------|----------| | Frame | Main frame | Nut | | | | Bolt | | | | Rivet | | | | Welding materials | | | Subframe | Nut | | | | Bolt | | | | Rivet | | | | Welding materials |

[0075] Figure 11 shows the result of displaying text written in Markdown notation in a Markdown viewer. Figure 11 is a parts list that includes a parts category column, a required parts column, a detailed parts column, and a manufacturing company column.

[0076] <<Operation of the Company Information Extraction Unit>> 12 and 13, a case will be described in which the company information extraction unit 24 calls the large-scale language model server 31 and extracts company information with the support of an LLM. Note that the company information extraction unit 24 is not limited to implementations supported by an LLM, and may perform rule-based processing, for example. Possible methods for the company information extraction unit 24 to perform rule-based processing include, for example, performing a web search for each company's name and obtaining company information from the top search results, and accessing a transaction information database to obtain company transaction information.

[0077] FIG. 12 shows a prompt 241 for instructing the user to extract company information. The company information extraction unit 24 sends a request in JSON format including this prompt 241 to the API endpoint of the large-scale language model server 31. The prompt 241 is transcribed below.

[0078] prompt # Task Please add manufacturers to the bill of materials using the following constraints. Then, create a table of the official websites of each manufacturer. # Constraints - Add the manufacturer of each "required part" to the column to the right of "required part". - Add the production company for each "Detailed Part" in the column to the right of "Detailed Part". - Add companies that produce each part

[0079] The company information extraction unit 24 sends a JSON-formatted request including this prompt 241 and past system and user messages in a message array to the API endpoint of the large-scale language model server 31. The JSON-formatted request is shown below.

[0080] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 211 string)} {"role": "assistant", "content": (Proposal 42 text)} {"role": "user", "content": (Prompt 221 string)} {"role": "assistant", "content": (string from data 43)} {"role": "user", "content": (Prompt 231 string)} {"role": "assistant", "content": (string from data 44)} {"role": "user", "content": (Prompt 241 string)} ], "temperature": 0.7 }

[0081] FIG. 13 is a diagram showing data 45 in which company information is added to the element table. The company information extraction unit 24 sends a JSON-formatted request including a prompt 241 to the API endpoint of the large-scale language model server 31, and then receives data 45 included in the JSON-formatted response. An example of a JSON-formatted response is shown below.

[0082] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (string of data 45) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": … } }

[0083] The Markdown text written in this data 45 is listed below.

[0084] # Below is a tabular breakdown of the elements involved in developing a flying car. | Parts Category | Required Parts | Manufacturers | Part Details | Manufacturers | |--------------|------------|------------------|----------|------------------| | Frame | Main Frame | Company A, Company B, Company C | Nut | Company D, Company F, Company E | | | | | Bolts | G Company, H Company, I Company | | | | | Rivets | Company J, Company K, Company L | | | | | Welding materials | Company M, Company N, Company O | | | Subframe | Company A, Company B, Company C | Nut | Company D, Company E, Company F | | | | | Bolts | G Company, H Company, I Company | | | | | Rivets | Company J, Company K, Company L | | | | | Welding materials | Company M, Company N, Company O | # Below is information about the manufacturing companies and their official websites. | Manufacturers | Official website information | |----------|----------------------| | Company A | | | Company B | ... | | Company C | ... | | … | ··· |

[0085] Fig. 13 shows the result of displaying text written in Markdown notation in a Markdown viewer. The parts list in Fig. 13 includes a parts category column, a required parts column, a production company column, a detailed parts column, and a production company column.

[0086] <<Operation of the LLM element value extraction unit>> A case where the LLM element value extraction unit 25a calls the large-scale language model server 31 and extracts element values ​​with the support of LLM will be described below with reference to FIGS.

[0087] FIG. 14 shows a prompt 251 for instructing to extract the original value. The LLM original value extraction unit 25a sends a request in JSON format including this prompt 251 to the API endpoint of the large-scale language model server 31. The prompt 251 is transcribed below.

[0088] prompt # Task Please list the outline of the manufacturing companies based on the following constraints. # Constraints - All companies listed in the "Production Company" column are included - Extract the contents of the official website - Summary includes sales revenue

[0089] The LLM original value extraction unit 25a sends a JSON format request including this prompt 251 and past system and user messages in a message array to the large-scale language model server 31. The JSON format request is shown below.

[0090] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 211 string)} {"role": "assistant", "content": (Proposal 42 text)} {"role": "user", "content": (Prompt 221 string)} {"role": "assistant", "content": (string from data 43)} {"role": "user", "content": (Prompt 231 string)} {"role": "assistant", "content": (string from data 44)} {"role": "user", "content": (Prompt 241 string)} {"role": "assistant", "content": (string from data 45)} {"role": "user", "content": (Prompt 251 string)} ], "temperature": 0.7 }

[0091] FIG. 15 is a diagram showing an example of data listing company overviews. The LLM element value extraction unit 25a receives the data 46 included in the response in JSON format after sending the prompt 251 to the large-scale language model server 31. An example of a response in JSON format is shown below.

[0092] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (string from data 46) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": … } }

[0093] Below is a portion of the Markdown text written in Data 46. Data 46 actually contains an overview of all manufacturing companies from Company A to Company O.

[0094] # Below is a list of the designated production companies based on the information on their official websites. ## 1. Company A overview:… Sales:… ## 2.Company B overview:… Sales:… ## 3.Company C overview:… Sales:…

[0095] Figure 15 shows the result of displaying text written in Markdown notation in a Markdown viewer.

[0096] <<Operation of the LLM index value calculation unit>> 16A to 16E and 17, a case will be described in which the LLM index value calculation unit 26a calls the large-scale language model server 31 and extracts index values ​​with the support of LLM. Note that the LLM index value calculation unit 26a is not limited to implementations with the support of LLM, and may perform processing based on rules, for example.

[0097] FIG. 16A is a diagram showing a prompt 261a that instructs the user to score the reliability as an index value of the company information. If the index type is reliability, the LLM index value calculation unit 26a sends this prompt 261a via the API to the large-scale language model server 31. The prompt 261a is transcribed below.

[0098] prompt # Task Please assign a score to the manufacturer based on the following reliability constraints: # Constraints Reliability - A score will be given if the sales amount is as follows: - Over 30 trillion yen: 5 - 10 trillion yen to 30 trillion yen: 4 - Less than 10 trillion yen: 3 - Add 1 to your score in the following cases: - Working on SDGs

[0099] The prompt 261a is for when the type of index is reliability. When the type of index is another type, the LLM index value calculation unit 26a selects another prompt according to the type of index.

[0100] FIG. 16B is a diagram showing a prompt 261b that instructs the user to score safety as an index value of the company information. If the index type is safety, the LLM index value calculation unit 26a sends a JSON-formatted request including this prompt 261b to the API endpoint of the large-scale language model server 31. The prompt 261b is transcribed below.

[0101] prompt # Task Please score the manufacturer based on the following safety constraints: # Constraints ## Safety If you have ISO27001 certification, add 1 to your score. If the number of years of trading is less than X, subtract 1 If the number of years of trading is X or more, add 1

[0102] FIG. 16C shows a prompt 261c that instructs the user to score the quality as an index value of the company information. If the index type is quality, the LLM index value calculation unit 26a sends a JSON-formatted request including this prompt 261c to the API endpoint of the large-scale language model server 31. The prompt 261c is transcribed below.

[0103] prompt # Task Please score the manufacturer based on the following quality constraints: # Constraints ## Quality - If there is a history of defective products, the score will be reduced by 1.

[0104] FIG. 16D is a diagram showing a prompt 261d for instructing to score delivery time as an index value of the company information. If the index type is delivery time, the LLM index value calculation unit 26a sends this prompt 261d via the API to the large-scale language model server 31. The prompt 261d is transcribed below.

[0105] prompt # Task Please score the production company based on the following delivery time constraints: # Constraints ## deadline - If there is a history of delayed delivery, subtract 1 from the score.

[0106] FIG. 16E is a diagram showing a prompt 261e that instructs the user to score costs as index values ​​for company information. If the index type is cost, the LLM index value calculation unit 26a sends this prompt 261e via the API to the large-scale language model server 31. The prompt 261e is transcribed below.

[0107] prompt # Task Please score the production company based on the following cost constraints: # Constraints Cost - If the part unit price is below, a score is given. - XX yen or more: 1 - Between YY yen and XX yen: 2 - Less than YY yen: 3

[0108] The LLM index value calculation unit 26a sends a JSON-formatted request including any of these prompts 261a to 261e and past system and user messages in a message array to the API endpoint of the large-scale language model server 31. The JSON-formatted request is shown below.

[0109] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 211 string)} {"role": "assistant", "content": (Proposal 42 text)} {"role": "user", "content": (Prompt 221 string)} {"role": "assistant", "content": (string from data 43)} {"role": "user", "content": (Prompt 231 string)} {"role": "assistant", "content": (string from data 44)} {"role": "user", "content": (Prompt 241 string)} {"role": "assistant", "content": (string from data 45)} {"role": "user", "content": (Prompt 251 string)} {"role": "assistant", "content": (string from data 46)} {"role": "user", "content": (any string from prompts 261a to 261e)} ], "temperature": 0.7 }

[0110] FIG. 17 is a diagram showing an example of data in which index values ​​of company information are scored. The LLM original value extraction unit 25a receives the company score list data 47a included in the JSON format response after sending the prompt 261 to the large-scale language model server 31. An example of the JSON format response is shown below.

[0111] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (string of data 47) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": … } }

[0112] A portion of the Markdown text of this company score list data 47a is listed below.

[0113] | Company name | Sales revenue | SDG initiatives | Base score | SDG score | Total score | |--------|--------|----------------|------------|------------|------------| | Company A | ... | Yes | 5 | 1 | 6 | | Company B | ... | Yes | 4 | 1 | 5 | | Company C | ... | Yes | 3 | 1 | 4 |

[0114] Note that Figure 17 shows the result of displaying Markdown-written text in a Markdown viewer.

[0115] <<Operations of the Rule-Based Source Value Extraction Unit and the Rule-Based Index Value Calculation Unit>> FIG. 18 is a flowchart showing a rule-based scoring of index values ​​of company information. The rule-based original value extraction unit 25b determines the type of the index (step S10). If the index is reliability, the process proceeds to step S11; if the index is safety, the process proceeds to step S14; if the index is quality, the process proceeds to step S17; if the index is delivery date, the process proceeds to step S20; and if the index is cost, the process proceeds to step S23.

[0116] In step S11, the rule-based original value extractor 25b extracts items to be used in calculating the reliability. Then, the rule-based index value calculator 26b normalizes the item values ​​in the transaction information (step S12), calculates the reliability score based on the above-mentioned formula (1) (step S13), and ends the processing of FIG.

[0117] In step S14, the rule-based original value extractor 25b extracts items to be used in calculating safety. Then, the rule-based index value calculator 26b normalizes the item values ​​in the transaction information (step S15), calculates the safety score based on the above-mentioned formula (2) (step S16), and ends the processing in FIG.

[0118] In step S17, the rule-based original value extractor 25b extracts items to be used in calculating the quality. Then, the rule-based index value calculator 26b normalizes the values ​​of the items in the transaction information (step S18), calculates the quality score based on the above-mentioned formula (3) (step S19), and ends the processing in FIG.

[0119] In step S20, the rule-based original value extractor 25b extracts items to be used in calculating the delivery time. Then, the rule-based index value calculator 26b normalizes the values ​​of the items in the transaction information (step S21), calculates the score of the delivery time based on the above-mentioned formula (4) (step S22), and ends the processing in FIG.

[0120] In step S23, the rule-based original value extractor 25b extracts items to be used in calculating costs. Then, the rule-based index value calculator 26b normalizes the item values ​​in the transaction information (step S24), calculates the cost score based on the above-mentioned formula (5) (step S25), and ends the processing in FIG.

[0121] <<Operation of the ranking part>> 18 and 19, a case will be described in which the ranking unit 27 calls the large-scale language model server 31 and ranks each company with the support of LLM. By comparing the company score list data 47a calculated with the support of LLM with the company score list data 47b calculated based on rules, the ranking unit 27 can calculate company scores that take non-numerical indicators into consideration, and can mitigate erroneous output due to hallucination in LLM with the rule-based output. The ranking unit 27 may rank each company using only the company score list data 47a calculated with the support of LLM. The ranking unit 27 is not limited to implementations supported by LLM, and may perform processing on a rule-based basis, for example.

[0122] FIG. 19 shows a prompt 271 for ranking companies by score. The ranking unit 27 sends this prompt 271 via the API to the large-scale language model server 31. The prompt 271 is transcribed below.

[0123] prompt # Task Please sort the manufacturers by score according to the following constraints: # Constraints - Open to all businesses - Prorate scores between LLM and Rules-Based tables # LLM table ... # Rule-based table ...

[0124] The ranking unit 27 sends a JSON-formatted request including this prompt 271 to the API endpoint of the large-scale language model server 31. The JSON-formatted request is shown below.

[0125] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 271 string)} ], "temperature": 0.7 }

[0126] FIG. 20 is a diagram showing ranking data 48 in which companies are ranked in order of score. After sending the prompt 271 to the large-scale language model server 31, the ranking unit 27 receives the ranking data 48 contained in the response in JSON format. An example of the response in JSON format is shown below.

[0127] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (string of ranking data 48) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": … } }

[0128] Below is a portion of the Markdown text written in this ranking data 48. In the actual ranking data 48, the total scores of all companies from A to O are calculated and ranked.

[0129] | Company name | Sales revenue | SDG initiatives | Base score | SDG score | Total score | |--------|--------|----------------|------------|------------|------------| | Company A | ... | Yes | 5 | 1 | 6 | | Company C | ... | Yes | 3 | 2 | 5 | | Company B | ... | Yes | 3 | 1 | 4 |

[0130] Figure 20 shows the results of displaying Markdown text in a Markdown viewer. The ranking data in Figure 20 includes a company name column, a sales amount column, a SDG initiative column, a basic score column, an SDG score column, and a total score column.

[0131] <<Operation of the supplier candidate list output section>> 21 and 22, a case will be described in which the supplier candidate list output unit 28 calls an API endpoint of the large-scale language model server 31, creates a supplier candidate list 49 with the support of an LLM, and outputs the list. Note that the supplier candidate list output unit 28 is not limited to implementations with the support of an LLM, and may perform processing on a rule basis, for example.

[0132] FIG. 21 is a diagram showing a prompt for instructing the creation of a supplier candidate list 49. As shown in FIG. The supplier candidate list output unit 28 transmits this prompt 281 via the API to the large-scale language model server 31. The prompt 281 is transcribed below.

[0133] prompt # Task Please update the order of the manufacturers in the bill of materials according to the following constraints: # Constraints - Display from left to right in descending order of score, separated by ","

[0134] The supplier candidate list output unit 28 sends a JSON format request including this prompt 281 to the API endpoint of the large-scale language model server 31. The JSON format request is shown below.

[0135] { "model": "gpt-4o", "messages": [ {"role": "system", "content": "You are a skilled business formation agent."}, {"role": "user", "content": (Prompt 211 string)} {"role": "assistant", "content": (Proposal 42 text)} {"role": "user", "content": (Prompt 221 string)} {"role": "assistant", "content": (string from data 43)} {"role": "user", "content": (Prompt 231 string)} {"role": "assistant", "content": (string from data 44)} {"role": "user", "content": (Prompt 241 string)} {"role": "assistant", "content": (string from data 45)} {"role": "user", "content": (Prompt 271 string)} {"role": "assistant", "content": (string of ranking data 48)} {"role": "user", "content": (Prompt 281 string)} ], "temperature": 0.7 }

[0136] FIG. 22 is a diagram showing an example of the supplier candidate list 49. As shown in FIG. The candidate supplier list output unit 28 sends a JSON-formatted request including a prompt 281 to the API endpoint of the large-scale language model server 31, and then receives a candidate supplier list 49. An example of a JSON-formatted response is shown below.

[0137] { "id": …, "object": "chat.completion", "created": …, "model": "gpt-4o", "choices": [ { "index": 0, "message": { "role": "assistant", "content": (string of customer candidate list 49) }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": …, "completion_tokens": …, "total_tokens": … } }

[0138] The text written in Markdown notation in this business partner candidate list 49 is shown below.

[0139] | Parts Category | Required Parts | Manufacturers | Part Details | Manufacturers | |--------------|------------|------------------|----------|------------------| | Frame | Main Frame | Company A, Company C, Company B | Nut | Company D, Company F, Company E | | | | | Bolts | G Company, H Company, I Company | | | | | Rivets | Company J, Company K, Company L | | | | | Welding materials | Company M, Company N, Company O | | | Subframe | Company A, Company C, Company B | Nut | Company D, Company E, Company F | | | | | Bolts | G Company, H Company, I Company | | | | | Rivets | Company J, Company K, Company L | | | | | Welding materials | Company M, Company N, Company O |

[0140] Note that Fig. 22 shows the result of displaying text written in Markdown notation in a Markdown viewer. The supplier candidate list 49 in Fig. 22 includes a part category column, a required part column, a production company column, a detailed part column, and a production company column.

[0141] The business formation system of the present invention can help you find the optimal business partners based on your business concept, significantly reducing the time it takes to find the business partners you need for your business. For example, if you need to collaborate with 10 business partners, doing it manually would take at least 10 weeks, assuming one week per company. With the business formation system of the present invention, it is expected that this can be completed in one day.

[0142] The configuration and effects of the present invention will be described below.

[0143] [1] a business component decomposition unit (22) that receives data on a business plan and decomposes the business plan into elements or parts that constitute the business plan; If the elements decomposed by the business component decomposition unit (22) are parts, a company information extraction unit (24) extracts company information on a plurality of companies that produce the parts; an LLM index value calculation unit (26a) that calculates, by a large-scale language model (large-scale language model server 31), LLM index values ​​that evaluate the company information about the plurality of companies extracted by the company information extraction unit (24); a ranking unit (27) that ranks the plurality of companies based on the LLM index value; A business formation system (5) comprising:

[0144] This allows users to narrow down the list of companies that produce each part that is broken down into products related to that business simply by entering the business plan.

[0145] [2] a rule-based index calculation unit (26b) that calculates rule-based indexes by evaluating the company information about the plurality of companies extracted by the company information extraction unit (24), The ranking unit (27) ranks the plurality of companies based on the LLM index value and the rule-based index value. A business formation system (5) according to claim 1.

[0146] This makes it possible to mitigate errors even if hallucinations are included in the answers of large-scale language models.

[0147] [3] the company information extraction unit (24) extracts company information relating to a plurality of companies that produce the parts from transaction information with each company; A business formation system (5) according to claim 1.

[0148] This allows us to evaluate each company based on actual transaction information.

[0149] [4] the company information extraction unit (24) extracts company information relating to a plurality of companies that manufacture the parts from the homepages of the companies; A business formation system (5) according to claim 1.

[0150] This allows you to evaluate each company based on information on their official website, for example.

[0151] [5] The system further includes a listing unit (23) that lists the structures between the elements decomposed by the business component decomposition unit (22), A business formation system (5) according to claim 1.

[0152] This makes it possible to clearly show the hierarchical structure of parts and the relationship between parts and the companies that produce them.

[0153] [6] a supplier candidate list output unit (28) that associates the plurality of companies ranked by the ranking unit (27) with the structure between each element listed by the listing unit (23) and outputs the result as a supplier candidate list (49) for parts related to the business plan; A business formation system (5) according to claim 5.

[0154] This makes it possible to output a list of potential business partners that clearly shows the relationship between the parts and the companies that produce them.

[0155] [7] The system further includes a pre-processing unit (21) that receives business model canvas information (41) and converts it into data for the business plan (42). A business formation system (5) according to claim 1.

[0156] This allows you to convert the Business Model Canvas information into a Markdown-formatted business proposal that is easy for large-scale language models to understand.

[0157] [8] The business component decomposition unit (22) decomposes the business plan into components that constitute the business plan using a large-scale language model. A business formation system (5) according to claim 1.

[0158] This allows a business plan to be broken down into its constituent elements simply by prompting a large-scale language model, without the need for a rule-based approach.

[0159] [9] The business component decomposition unit (22) decomposes the business plan into components that constitute the business plan on a rule-based basis. A business formation system (5) according to claim 1.

[0160] This allows us to avoid incorrect answers caused by hallucination in large-scale language models and break down the business proposal into its constituent elements.

[0161]

[10] If the element decomposed by the business component decomposition unit (22) is a part, the company information extraction unit (24) extracts company information about a plurality of companies that produce the part using a large-scale language model. A business formation system (5) according to claim 1.

[0162] This makes it possible to extract company information about multiple companies that produce parts simply by prompting a large-scale language model, without having to create a rule-based model.

[0163]

[11] If the element decomposed by the business component decomposition unit (22) is a part, the company information extraction unit (24) extracts company information on a plurality of companies that produce the part on a rule-based basis. A business formation system (5) according to claim 1.

[0164] This makes it possible to avoid incorrect answers due to hallucination in large-scale language models and extract company information about multiple companies.

[0165]

[12] A step in which a business component decomposition unit (22) receives data of a business plan (42) and decomposes the data into components constituting the business plan; If the elements decomposed by the business component decomposition unit (22) are parts, a company information extraction unit (24) extracts company information on a plurality of companies that produce the parts; an LLM index value calculation unit (26a) calculating an LLM index value that evaluates each of the pieces of company information about the plurality of companies extracted by the company information extraction unit (24) using a large-scale language model; A step in which a ranking unit (27) ranks the plurality of companies based on the LLM index value; A business formation method comprising:

[0166] This allows users to narrow down the list of companies that produce each part that is broken down into products related to that business simply by entering the business plan.

[0167] <<Variation>> The present invention is not limited to the above-described embodiments and includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and are not necessarily limited to those including all of the described configurations. It is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is also possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0168] The above-described configurations, functions, processing units, processing means, etc. may be realized in part or in whole by hardware such as an integrated circuit. The above-described configurations, functions, etc. may be realized by software by a processor interpreting and executing a program that realizes each function. Information such as the programs, tables, and files that realize each function can be stored in a storage device such as a memory, a hard disk, or an SSD (Solid State Drive), or on a storage medium such as a flash memory card or a DVD (Digital Versatile Disk).

[0169] In each embodiment, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are interconnected. [Explanation of symbols]

[0170] 5. Business formation system 1 device 2. Business formation server 31 Large-scale language model server 33 Transaction Information Database 42 Business proposal 11 Input section 12 Display section 41 Business Model Canvas Information 40 indicators 49 Business Partner Candidate List 32 Company information 21 Pretreatment section 22 Business Component Decomposition Department 23 Listing Department 24 Corporate Information Extraction Department 25a LLM original value extraction unit 26a LLM index calculation section 25b Rule-based original value extraction unit 26b Rule-based index value calculation unit 27 Ranking Section 28. Supplier candidate list output section 47a Company Score List Data (LLM Index Value) 47b Company Score List Data (Rule-Based Index Values) 201 CPU 202 ROM 203 RAM 204 Storage section 205 Display section 206 Input section 207 Communication Interface 208 Bus 10 Business Model Canvas Input Screen 100 cells 101 cells 102 cells 103 cells 104 cells 105 cells 106 cells 107 cells 108 cells 109 cells 1011 Index Combo Box 1012 Business idea creation button 19 Customer candidate list screen 211 prompt

Claims

1. a business component decomposition unit that receives data of a business plan and decomposes the business plan into elements or parts that constitute the business plan; a company information extraction unit that extracts company information about a plurality of companies that produce the parts if the elements decomposed by the business component decomposition unit are parts; an LLM index value calculation unit that calculates, by a large-scale language model, an LLM index value that evaluates each of the pieces of company information about the plurality of companies extracted by the company information extraction unit; a rule-based index calculation unit that calculates rule-based indexes that evaluate the company information about the plurality of companies extracted by the company information extraction unit; a ranking unit that ranks the plurality of companies based on the LLM index value and the rule-based index value; A business formation system comprising:

2. A preprocessing unit that receives business model canvas information and converts it into business plan data; a business component decomposition unit that receives data of the business plan and decomposes the business plan into elements or parts that constitute the business plan; a company information extraction unit that extracts company information about a plurality of companies that produce the parts if the elements decomposed by the business component decomposition unit are parts; an LLM index value calculation unit that calculates, by a large-scale language model, an LLM index value that evaluates each of the pieces of company information about the plurality of companies extracted by the company information extraction unit; a ranking unit that ranks the plurality of companies based on the LLM index value; A business formation system comprising:

3. a business component decomposition unit that receives data of a business plan and decomposes the business plan into elements or parts that make up the business plan on a rule-based basis; a company information extraction unit that extracts company information about a plurality of companies that produce the parts if the elements decomposed by the business component decomposition unit are parts; an LLM index value calculation unit that calculates, by a large-scale language model, an LLM index value that evaluates each of the pieces of company information about the plurality of companies extracted by the company information extraction unit; a ranking unit that ranks the plurality of companies based on the LLM index value; A business formation system comprising:

4. A business component decomposition unit that receives data on a business plan and decomposes the business plan into elements or parts that make up the business plan; a company information extraction unit that, if the elements decomposed by the business component decomposition unit are parts, extracts company information on a rule-based basis about a plurality of companies that produce the parts; an LLM index value calculation unit that calculates, by a large-scale language model, an LLM index value that evaluates each of the pieces of company information related to the plurality of companies extracted by the company information extraction unit; a ranking unit that ranks the plurality of companies based on the LLM index value; A business formation system comprising:

5. a business component decomposition unit that receives data of a business plan and decomposes the business plan into elements or parts that constitute the business plan; a company information extraction unit that extracts company information about a plurality of companies that produce the parts if the elements decomposed by the business component decomposition unit are parts; an LLM index value calculation unit that calculates, by a large-scale language model, an LLM index value that evaluates each of the pieces of company information related to the plurality of companies extracted by the company information extraction unit; a ranking unit that ranks the plurality of companies based on the LLM index value; a listing unit that lists the structures between the elements decomposed by the business component decomposition unit; A business formation system comprising:

6. a supplier candidate list output unit that associates the plurality of companies ranked by the ranking unit with the structure between each element listed by the listing unit and outputs the result as a supplier candidate list for parts related to the business plan; The business formation system according to claim 5 .

7. the company information extraction unit extracts company information relating to a plurality of companies that produce the parts from transaction information with each company; 6. The business formation system according to claim 1, wherein:

8. the company information extraction unit extracts company information relating to a plurality of companies that produce the parts from the homepages of the companies; 6. The business formation system according to claim 1, wherein:

9. the business component decomposition unit decomposes the business plan into components that constitute the business plan using a large-scale language model; 6. The business formation system according to claim 1, wherein:

10. the company information extraction unit, if the elements decomposed by the business component decomposition unit are parts, extracts company information about a plurality of companies that produce the parts using a large-scale language model; 6. The business formation system according to claim 1, wherein:

11. a step in which a business component decomposition unit receives data of a business plan and decomposes the data into components constituting the business plan; If the elements decomposed by the business component decomposition unit are parts, a company information extraction unit extracts company information on a plurality of companies that produce the parts; an LLM index value calculation unit calculating an LLM index value by using a large-scale language model, the LLM index value evaluating each of the pieces of company information related to the plurality of companies extracted by the company information extraction unit; a step of calculating rule-based index values ​​by a rule-based index value calculation unit evaluating the company information about the plurality of companies extracted by the company information extraction unit; a ranking unit ranking the plurality of companies based on the LLM index value and the rule-based index value; A business formation method comprising:

12. A preprocessing unit that receives business model canvas information and converts it into business plan data; a step in which a business component decomposition unit receives data of the business plan and decomposes the data into components constituting the business plan; If the elements decomposed by the business component decomposition unit are parts, a company information extraction unit extracts company information on a plurality of companies that produce the parts; an LLM index value calculation unit calculating an LLM index value by using a large-scale language model, the LLM index value evaluating each of the pieces of company information related to the plurality of companies extracted by the company information extraction unit; a ranking unit ranking the plurality of companies based on the LLM index value; A business formation method comprising:

Images