Business information processing system, business information processing method, and business information processing program

Abstract

In situations where multiple business systems exist within an organization, this system provides a business information processing system that enables users to efficiently perform their tasks. [Solution] The business information processing system comprises a language model 11 trained to output responses when output instructions are input in natural language, a plurality of business support systems 13, and an API 14 which is an interface to the business support systems. When an output instruction or output request is input to the language model 11, the language model 11 selects the API 14 necessary for the output instruction or the output according to the output instruction and sends a request to the API 14.

Description

Technical Field

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

Background Art

[0002] In enterprises, a large number of types of business support systems such as attendance management systems, approval systems, expense management systems, contract document management systems, order systems, and email sending / receiving systems are used in business.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When there are a large number of business systems within an organization, when performing a certain business, users need to consider which business system to use and how to use it, which is laborious and time-consuming. In particular, when multiple types of business systems must be used to perform work, there are problems that it takes a lot of time to consider how to use multiple types of business systems and to operate multiple business systems.

[0005] An object of the present invention is to provide a business information processing system, a business information processing method, and a business information processing program that enable users to efficiently perform business operations in a situation where multiple business systems exist within an organization.

Means for Solving the Problems

[0006] To solve the above problems, the business information processing system according to an embodiment of the present invention comprises a language model trained to output a response when an output instruction is input, a plurality of business support systems, and an API which is an interface to the business support systems. When an output instruction is input to the language model, the language model selects the API necessary for outputting according to the output instruction and sends a request to the API.

[0007] The business support system corresponding to the API from which the request was sent may send output corresponding to the request to the language model, and the language model may determine the next action based on the output sent by the business support system.

[0008] The aforementioned business support system may include an approval system and an ordering system. In the approval system, if an approval request is approved, the language model may extract purchase information included in the approval request, input the extracted purchase information into an API which is the interface of the ordering system, and the ordering system may execute a purchase process corresponding to the purchase information input into the API.

[0009] The business support system may include an accounting system, and when the language model is instructed to perform a comparison between input information relating to a document and information related to said document, the language model may send a request to the accounting system's API that includes a request to output the input information relating to said document, and in accordance with the request, the accounting system may perform a comparison between the input information relating to the document output and information related to said document, and output the result of the comparison.

[0010] If the language model is instructed to compare input information relating to a document with information recorded in an image of the document, the language model may send a request to the accounting system's API that includes an instruction to output the input information relating to the document, and the accounting system may then compare the input information relating to the document output by the accounting system with the information recorded in an image of the document and output the comparison result.

[0011] The business support system may include an invoice management system, an approval system, and a payment processing system. When an invoice image is uploaded to the invoice management system, the language model sends a first request to the API of the approval system, which includes an output instruction to search for an approved approval document corresponding to the transaction recorded in the invoice image. If the approval system responds to the first request by indicating that an approved approval document for the transaction exists, the language model sends a second request to the API of the payment processing system, which includes an output instruction to perform payment processing corresponding to the transaction. Alternatively, the system may include a business support system function list in which the functions of the business support system are recorded. When an output instruction is input to the language model, the language model may refer to the business support system function list to select the API necessary for outputting according to the output instruction and send a request to that API.

[0012] A business information processing method according to an embodiment of the present invention comprises: an output instruction input step in which an output instruction is input to a language model that has been trained by machine learning to output a response when an output instruction is input, which is executed by a computer; a request transmission step in which the language model selects an API necessary for outputting according to the output instruction from a plurality of APIs and sends a request to the selected API; and a request execution step in which the business support system corresponding to the API to which the request was sent in the request transmission step executes the request and causes the business support system to output a response.

[0013] An embodiment of the present invention provides a business information processing program that causes a computer to perform the following steps: an output instruction input step of inputting an output instruction to a language model trained on machine learning to output a response when an output instruction is input; a request transmission step of the language model selecting an API necessary for outputting according to the output instruction from a plurality of APIs and sending a request to the selected API; and a request execution step of causing a business support system corresponding to the API to which the request was sent in the request transmission step to execute the request and cause the business support system to output a response. [Effects of the Invention]

[0014] In situations where multiple business systems exist within an organization, we can provide a business information processing system, a business information processing method, and a business information processing program that enable users to efficiently perform their tasks. [Brief explanation of the drawing]

[0015] [Figure 1] This figure shows the configuration of a business information processing system according to an embodiment of the present invention. [Figure 2] This diagram shows the configuration of the computers used in this system. [Figure 3] This figure shows a flowchart of information processing in an embodiment of the present invention. [Figure 4] This figure shows an example of a business support system function list. [Figure 5] This diagram shows the configuration of the business information processing system according to Example 1. [Figure 6] This figure shows the configuration of the business information processing system according to Example 2. [Figure 7] This figure shows the configuration of the business information processing system according to Example 3. [Figure 8] This diagram shows the configuration of the business information processing system according to Example 4. [Modes for carrying out the invention]

[0016] FIG. 1 is a diagram for explaining an embodiment of the present invention. A business information processing system according to an embodiment of the present invention includes a language model 11 that is machine-learned to output an answer when an output instruction is input in a natural language, a plurality of business support systems 13, and an API 14 that is an interface of the business support system. When an output instruction or an output request is input to the language model 11, the language model 11 selects an API 14 necessary for the output according to the output instruction or the output instruction, and transmits a request to the API 14.

[0017] A business support system is a system or application program for supporting specific operations such as accounting operations, personnel operations, production operations, logistics operations, sales operations, etc. in an organization. For example, there are expense application systems, attendance management systems, approval systems, order placement systems (including purchasing systems), contract document management systems, accounting systems, salary management systems, email sending and receiving systems, communication systems, patent information management systems, tax management systems, production management systems, sales support systems, entry and exit management systems, etc., but not limited to these. Business support systems include ERP (Enterprise Resource Planning) systems, integrated core business systems, core systems, and business management systems.

[0018] API is an abbreviation for Application Programming Interface. API refers to an interface for exchanging information between software, programs, and web services (inputting information from another software, etc. to software, etc.). Usually, a user inputs information or an output instruction to software, but for software equipped with an API, input from other software can also be input through the API.

[0019] A language model refers to a model that has been machine - learned using training data (teacher data) to output an answer according to an output instruction or output request in natural language. A language model can process not only natural language but also programming languages. A language model constructed with a large amount of data is called an LLM (Large Language Models). In this specification, a language model with a training data capacity of 1GB or more is referred to as an LLM (Large Language Model).

[0020] The information processing of this system is realized by a computer. Figure 2 is a diagram showing the configuration of a computer 20 used in the information processing of this system. The storage unit 21 is a device having a function of storing information. In this system, the language model 11 is stored in the storage unit 11. The storage unit 11 is composed of, for example, a hard disk, an SSD, a semiconductor memory, etc. The arithmetic unit 22 is a device for executing a program read from the storage unit 21. The output unit 24 is a device for outputting the information stored in the storage unit 21 in a recognizable manner to the user, and is composed of, for example, a display, a speaker, etc. The input unit 23 is a device for the user to input information to the computer 20, and is composed of, for example, a keyboard, a mouse, a microphone, a camera. The communication unit 25 is a device for performing input and output of information with a device outside the computer 20.

[0021] Figure 3 is a diagram showing a flowchart of the information processing in this embodiment. First, the user inputs an output instruction (request) to the LLM. Then, the LLM understands the user's request and determines an action that must be executed to achieve the request (ST1).

[0022] Next, the LLM determines which business system with an API to use to execute the decided action (ST2). Specifically, it refers to a business support system function list (see Figure 4), which records the functions of each business support system with an API, and identifies the business support system that can execute the action based on this list. Figure 4 shows List 4, which records the main functions of the business support systems. Although List 4 only lists the main functions of the business support systems, the actual business support system function list records the functions of the business support systems in detail and comprehensively.

[0023] Then, LLM creates a request containing keywords for executing the action, which is sent to the API corresponding to the available business support system in order to carry out the decided action (ST3). Subsequently, it sends the created request to the API and queries the API (ST4).

[0024] The business support system that receives the API inquiry outputs a response (also called an answer) and sends it to the LLM API (ST5). LLM reviews the response, and if the response contains errors or differs from the expected response, it returns to ST3 (ST6), corrects the request, and sends the corrected request to the API.

[0025] If the response is as expected, determine whether the requirement has been met (ST7). If the requirement has not been met, return to ST2 to perform the next action, and then perform the steps from ST2 to ST7 for the next action. In ST7, if the request is fulfilled, the process terminates and the response is output (ST8). [Examples]

[0026] Figure 5 shows the configuration of the business information processing system according to Example 1. In the approval system 531, the person in charge submits an approval request, and the approving authority performs approval or rejection operations in sequence. The approval system 531 has an API 541 for inputting and outputting information with external software. The ordering system 532 has a function to search whether the product you want to buy is available for purchase and a function to send order information to the supplier. The ordering system 542 also has an API 542 for inputting and outputting information with external software.

[0027] In the approval system 531, if the approval request includes the purchase of goods, once the approval request is granted, it is programmed to send a predetermined output instruction to the LLM API stating, "Based on this approval request, search for the target goods in the purchasing system and purchase them if they are found. Purchase only if the conditions of the approval request are met. If the target goods are not found, notify the applicant accordingly." When the above output instruction is input to the LLM 51 API 52, the LLM identifies the action required to produce the output according to the above instruction, and searches for a business support system available within the organization to perform that action by referring to the business support system function list (see Figure 4). As a result of the search, it identifies that the ordering system can perform the action and selects the ordering system 532 API 542 as the destination for the request.

[0028] Next, LLM51 creates a request to execute the action, which will be sent to API542 of the ordering system 532, in a format that API542 of the ordering system 532 can process. The content of the request is a concrete manifestation of the output instructions described above, but since API542 of the ordering system 532 cannot process natural language, it needs to be converted into a format that API542 can process.

[0029] When the created request is sent to API 542, the ordering system 532 executes the contents of the request. Specifically, it checks whether the desired purchase items listed in this approval document can be purchased in accordance with the conditions and internal rules stated in the approval document. If the purchase is possible, it executes the order process. If the purchase is not possible, it notifies the approval applicant of this fact via email or other communication tool.

[0030] According to this embodiment, the applicant does not need to perform any operations on the approval system or the ordering system. Once the approval is granted, the purchase process is automatically initiated, reducing the burden on the user's work and improving operational efficiency. Furthermore, the necessary purchase processes are reliably executed.

[0031] Furthermore, since output instructions for LLM are expressed in natural language, users can easily add new output instructions as needed for their work, even without programming knowledge. [Examples]

[0032] Figure 6 shows the configuration of the business information processing system according to Example 2. The accounting system 631 stores data related to expenses and expenditures. The invoice image management system 632 records expenditure data read by applying AI-OCR to invoice images. AI-OCR (Optical Character Recognition / Reader) can not only extract text information from images, but also extract text information corresponding to predetermined items.

[0033] If a user inputs an output instruction to LLM61 such as "Please tell me the patent-related expenses from January to September 2024" or "Please verify the patent-related expenses from January to September 2024," LLM61 will verify the content of the output instruction against the information recorded in the business support system function list (see Figure 4), identify that the accounting system 631 and the invoice image management system 632 are the appropriate business support systems to perform the action corresponding to the output instruction, and select APIs 641 and 642, which correspond to these business support systems, as the destinations for the request. Verification means comparing two or more types of information to confirm whether this information is correct. Then, requests are created for each API in a format that APIs 641 and 642 can process. The created requests are sent to APIs 641 and 642, respectively, and the accounting system 631 and the invoice image management system 632 execute the requests, respectively. As a result, if the accounting system 631 outputs a response stating, "Patent-related expenses for the period from January to September 2024 are 10 million yen," and the invoice image management system 632 outputs a response stating, "Patent-related expenses for the period from January to September 2024 are 10.2 million yen," the user can recognize that 200,000 yen worth of patent-related expenses may not have been entered into the accounting system, or that an incorrect invoice image may have been associated with the transaction and saved.

[0034] If a user were to investigate this matter by operating an accounting system and an invoice image management system, it would take several hours. However, with this embodiment, simply entering a natural language command into LLM, such as "Please tell me the patent-related expenses from January to September 2024," will instantly provide the answer, significantly improving work efficiency. Furthermore, even if the question is not anticipated and the answer process is not programmed into the system, an appropriate answer can be obtained instantly. Moreover, with this embodiment, even if a user does not know how to operate an accounting system or an invoice image management system, they can obtain the information they need by simply entering output instructions in natural language using the business support system.

[0035] This system can also efficiently perform various verification tasks. For example, if you want to check whether an employee's travel expense claim is appropriate, you can simply issue an output instruction to the LLM. The LLM will then identify the action required to respond according to that instruction, identify the expense claim system as the business support system capable of executing that action, and send a request to the expense claim system's API to obtain data on the employee's claimed route and claimed travel expenses. Subsequently, it will select the travel expense search system as the business support system capable of executing the action, and send a request to the travel expense search system's API to obtain data on the normal travel expenses for the claimed route. The LLM will then compare the travel expenses claimed by the employee with the normal travel expenses for the claimed route, determine whether the employee has claimed excessive travel expenses, and output the result. This allows you to instantly find out whether the employee's travel expense claim is appropriate simply by querying it in natural language. [Examples]

[0036] Figure 7 shows the configuration of the business information processing system according to Example 3. The email sending and receiving system 731 is an application program for sending and receiving emails and has an API 732 for inputting and outputting information with external software. The tax LLM 741 is an LLM specifically built to answer tax-related questions and has an API 742. The tax LLM 741 can only output answers to tax-related questions, but it can answer tax-related questions more accurately than a general-purpose LLM.

[0037] Suppose a user inputs an output instruction to LLM71 requesting it to answer a tax-related question. To respond to the output instruction, LLM71 creates two actions: Action 1, "Contact an external tax professional," and Action 2, "Contact a tax LLM." Then, as the business support system for executing Action 1, it refers to the business support system function list (see Figure 4) to identify the email sending / receiving system 731, and selects API 732 of the email sending / receiving system 731 as the destination for the request. As the content of the request, LLM71 creates the email text and email address of the appropriate external tax professional, and sends the request to API 732 of the email sending / receiving system 731. The email sending / receiving system 731 executes the request and sends an email to the external tax professional's email address inquiring about the matters to be asked.

[0038] Furthermore, LLM71 identifies Tax LLM741 as a business support system capable of executing Action 2 by referring to the business support system function list (see Figure 4), and selects API 741 of LLM741 as the destination for the request. LLM741 creates a natural language message to query Tax LLM741 as the content of the request, and sends the request to API 742 of Tax LLM741. Tax LLM741 executes the request and outputs an answer to the tax-related question.

[0039] In the example above, the sending of a request to API 732 of the email sending / receiving system 731 and the sending of a request to API 742 of the tax LLM 741 were performed simultaneously. However, if an appropriate response is not received from the tax LLM 741, the process of contacting a tax professional via email may be performed only then.

[0040] According to this embodiment, by simply inputting tax-related questions in natural language into the LLM, it is possible to execute multiple appropriate requests that fulfill the content of the output instructions. [Examples]

[0041] Figure 8 shows the configuration of the business information processing system according to Example 4. The invoice image management system 831 has the function of applying AI-OCR to the invoice image when invoice image information is uploaded, organizing, extracting, and saving the text data recorded in the invoice image. The payment processing system 833 has the function of creating payment information and sending the created payment information to a financial institution.

[0042] The invoice image management system 831 is pre-programmed to send an output instruction to LLM81 stating, "If payment has been approved, please request payment from the financial institution by the invoice due date," when invoice image information is uploaded and AI-OCR is applied to the invoice image information and it is determined that payment is required.

[0043] When the above output instruction is input to LLM81, it identifies action 3, "Check whether the approval request for the payment has been approved," as the action to execute the output instruction.

[0044] LLM81 identifies the approval system 832 as a business support system capable of executing action 3 by referring to the business support system function list (see Figure 4). It then creates a request corresponding to action 3 in a format that the approval system 832's API 842 can process, and sends the created request to the approval system 832's API 842. The approval system 832 executes the request, searches its database for an approved approval document corresponding to the payment information, notifies the user if no approved approval document exists, and sends a notification to LLM81's API 82 if an approved approval document exists.

[0045] Next, based on the information that there is an approved approval document, LLM81 identifies Action 4, "If the approval document for the payment has been approved, execute the payment process," as the next action. Then, LLM81 refers to the business support system function list (see Figure 4) as a business support system capable of executing Action 4 and identifies the payment processing system 833. Then, it creates a request corresponding to Action 4 in a format that API 843 of the payment processing system 833 can process and sends the request to API 843 of the payment processing system 833. The payment processing system 833 executes the request, creates payment request data, and sends the payment request data to the financial institution used by the company for payments by the payment deadline.

[0046] In this embodiment, the existence of an approved proposal document for the transaction was confirmed by sending a request to the approval system 832 for inquiry. However, it is also possible to send a request to the ordering system to confirm the existence of a purchase order for the transaction.

[0047] If invoice payment processing is done manually, it is labor-intensive and there is a risk of missed payments. However, with this embodiment, there is no burden on the user, and payments can be made reliably within the payment deadline.

[0048] Furthermore, since output instructions for LLM can be written in natural language, new output instructions can be easily added as needed for business operations, even without programming knowledge.

[0049] The embodiments of the present invention are not limited to the examples and embodiments described above, and the above examples and embodiments can be modified in various ways within the scope of the claims. [Explanation of Symbols]

[0050] 11 Language Models 12 API (LLM) 13 Business Support Systems 14 API 20 Computers 21 Memory section 22 Arithmetic section 23 Input section 24 Output section 25 Communications Department 4. Business Support System Function List 51 LLM 52 API (LLM) 531 Approval System 541, 542 API 531 Approval System 532 Ordering System 61 LLM 62 API (LLM) 631 Accounting System 632 Invoice Image Database 641, 642 API 71 LLM 72 API(LLM) 731 Email sending and receiving system 732, 742 API 81 LLM 82 API (LLM) 831 Invoice Image Management System 832 Approval System 833 Payment Processing System

Claims

1. A language model trained to output a response when an output instruction is input, Multiple business support systems, The system comprises an API which is an interface for the aforementioned business support system, When an output instruction is input to the language model, the language model selects the API necessary for the output according to the output instruction and sends a request to the API. Business information processing system.

2. The business support system corresponding to the API from which the request was sent sends the output corresponding to the request to the language model. The business information processing system according to claim 1, wherein the language model determines the next action based on the output transmitted by the business support system.

3. The aforementioned business support system includes an approval system and an ordering system. In the aforementioned approval system, when an approval request is approved, the language model extracts the purchase information included in the approval request, inputs the extracted purchase information into an API which is the interface of the ordering system, and the ordering system executes the purchase process corresponding to the purchase information input into the API. The business information processing system according to claim 1.

4. The aforementioned business support system includes an accounting system, When the language model is instructed to perform a comparison between input information relating to a document and information related to that document, the language model sends a request to the accounting system's API that includes a request to output the input information relating to the document, and the accounting system performs a comparison between the input information relating to the document output by the accounting system and the information related to that document, and outputs the result of the comparison. The business information processing system according to claim 1.

5. When the language model is instructed to compare input information relating to a document with information recorded in an image of the document, the language model sends a request to the accounting system's API that includes an instruction to output the input information relating to the document, and the accounting system, in accordance with the request, compares the input information relating to the document output by the accounting system with the information recorded in an image of the document and outputs the comparison result. The business information processing system according to claim 4.

6. The aforementioned business support system includes an invoice management system, an approval system, and a payment processing system. When an invoice image is uploaded to the invoice management system, the language model sends a first request to the API of the approval system, which includes an output instruction to search for approved approval documents corresponding to the transactions recorded in the invoice image. If the approval system outputs a response indicating that an approved approval document for the transaction exists in response to the first request, the language model sends a second request to the API of the payment processing system, which includes an output instruction to perform payment processing corresponding to the transaction. The business information processing system according to claim 1.

7. Executed by a computer, An output instruction input step involves inputting output instructions into a language model that has been trained to output a response when output instructions are input, and The language model selects an API from a plurality of APIs necessary for the output in accordance with the output instruction, and sends a request to the selected API in a request transmission step. The request execution step involves causing the business support system corresponding to the API to which the request was sent to execute the request and causing the business support system to output a response. A business information processing method having the following characteristics.

8. On the computer, An output instruction input step involves inputting output instructions into a language model that has been trained to output a response when output instructions are input, and The language model selects an API from a plurality of APIs necessary for the output in accordance with the output instruction, and sends a request to the selected API in a request transmission step. The request execution step involves causing the business support system corresponding to the API to which the request was sent to execute the request and causing the business support system to output a response. A business information processing program for executing tasks.

9. It includes a business support system function list that records the functions of the business support system, When an output instruction is input to the language model, the language model refers to the business support system function list to select the API necessary for the output according to the output instruction and sends a request to the API. The business information processing system according to claim 1.

Images