Information processing system

Abstract

This system treats employee icons in a human resources management system as virtual employees and provides an AI-powered information processing system for chat services. [Solution] An information processing system comprising a terminal used by a user, an existing human resources management system having a chat function between users, a terminal used by the operator providing the existing human resources management system, and a large-scale language model server, wherein a chatbot that receives and answers questions from users is installed on the existing human resources management system, and the system has a profile DB2 for registering the chatbot's basic information and a prompt DB3 for registering information related to creating prompts when the chatbot answers, and the chatbot has means for sending questions sent from the user terminal to the large-scale language model server 6, means for receiving answers from the large-scale language model server 6, and means for generating answers to send to the user terminal by referring to the prompt DB3.

Description

Technical Field

[0001] The present invention relates to an information processing system.

Background Art

[0002] In recent years, a personnel management system that manages personnel data within a company and functions in the cloud has been provided, enabling employees within the company to communicate with each other through a chat function. In addition, a chat service is also provided in which AI answers questions input by users in a natural dialogue format. For example, there is a chat service that transmits to a language model server a request to answer whether a document image violates company rules and obtains answer information (see Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the chat service of Patent Document 1, the answer information created by the language model server is sent to an operator terminal to confirm whether it is correct, and the information confirmed and corrected by the operator side is sent to the user terminal via a management server.

[0005] Therefore, there is a need for a system that utilizes a chat service that searches information accumulated within a company and automatically generates appropriate answers without operator confirmation. In addition, there is a need for a system that utilizes a chat service in a more natural dialogue format, as if obtaining answers from company employees.

[0006] Therefore, the present invention aims to provide an information processing system related to an AI-powered chat service using a human resources management system, where employee icons in the human resources management system are treated as virtual employees. [Means for solving the problem]

[0007] The present invention provides an information processing system comprising a user terminal used by a user, an existing human resources management system having a chat function between users, a terminal used by an operator that provides the existing human resources management system to users via a cloud server, and a large-scale language model server, wherein the existing human resources management system is equipped with a chatbot that receives and answers questions from users, and the storage unit of the existing human resources management system has a profile DB for registering basic information of the chatbot and a prompt DB for registering information related to the creation of prompts when the chatbot answers, and the chatbot has means for sending questions sent from the user terminal to the large-scale language model server, means for receiving answers from the large-scale language model server, and means for generating answers to send to the user terminal by referring to the prompt DB.

[0008] Furthermore, it is preferable that the chatbot is displayed in the same way as a user on the existing human resources management system, based on the aforementioned basic information.

[0009] Furthermore, it is preferable that the basic information be editable by the user terminal, and that the information regarding prompt creation registered in the prompt DB be updated based on the edited basic information.

[0010] Furthermore, it is preferable that the storage unit has a learning folder where document files are saved by the user terminal's operation, and that the document files are indexed and provided to the large-scale language model server.

[0011] Furthermore, it is preferable that the prompt creation information registered in the prompt DB can be edited to include business-related information, and that when business-related information is added, the user terminal can instruct the chatbot to generate a response related to the added business.

[0012] Furthermore, it is preferable that the storage unit further includes a document database for registering information related to the document file, and that the response from the large-scale language model server is generated by referencing the document database.

[0013] Furthermore, it is preferable that the answers generated by the information processing system are evaluated by judgment using a generation AI.

[0014] Furthermore, it is preferable that there be multiple chatbots, that they can be installed in each organization to which a user belongs, and that each user in each organization be able to access the chatbot.

[0015] Furthermore, multiple chatbots can be installed within a single organization to which a user belongs, and it is preferable that each chatbot generates responses based on the collaboration of multiple chatbots and sends them to the user's terminal. [Effects of the Invention]

[0016] According to the present invention, it is possible to provide an information processing system in a human resources management system that utilizes a chat service in which AI answers questions and consultations from employees within the company in a natural conversational format. [Brief explanation of the drawing]

[0017] [Figure 1] This diagram conceptually illustrates the configuration of a personnel management system related to an information processing system according to an embodiment of the present invention. [Figure 2] Similarly, this is a diagram that conceptually illustrates the dispatch of virtual staff. [Figure 3] Similarly, this is a diagram showing the registration flow for virtual staff on the company's side. [Figure 4] Also, it is a diagram showing a flow for generating a custom prompt. [Figure 5] Also, it is a conceptual diagram regarding the upload of document files to a learning folder. [Figure 6] Also, it is a diagram showing the user interface of a personnel management system. [Figure 7] Also, it is a diagram showing a flow for answer generation. [Figure 8] Also, it is a diagram of an example showing the difference in answer accuracy depending on the presence or absence of RAG. [Figure 9] Also, it is a diagram showing a flow of the OJT operation process for virtual staff. [Figure 10] Also, it is a conceptual diagram showing a method of feedback for virtual staff. [Figure 11] Also, it is a conceptual diagram showing a method of reflecting feedback results in a prompt DB and a reference document folder. [Figure 12] Also, it is a diagram showing a flow of setting a destination manager by a system administrator. [Figure 13] Also, it is a diagram showing a flow of setting a publication range by a system administrator. [Figure 14] Also, it is a diagram showing an example of generation of information related to document files in a learning folder. [Figure 15] Also, it is a diagram showing a flow for information generation regarding answer accuracy. [Figure 16] Also, it is a diagram showing a flow for information generation regarding answer accuracy. [Figure 17] Also, it is a diagram showing an example of a page for administrators. [Figure 18] Also, it is a diagram showing an example of collaboration of multiple virtual staff.

Embodiments for Carrying Out the Invention

[0018] Embodiments of the present invention will be described below with reference to the drawings. Figure 1 conceptually shows the configuration of the personnel management system 1 related to the information processing system of the present invention. The personnel management system 1 is provided on a cloud server. This personnel management system 1 has multiple chatbot programs equipped with AI functions, called virtual staff, registered in the domain of the operating management company. By registering these virtual staff in the domains of Company A and Company B in the personnel management system 1, it becomes possible to display them on the operation screen of the personnel management system 1 as if they were dispatched as employees of Company A and Company B. In the example in Figure 1, the pre-configured virtual staff A and B are dispatched to Company A, and virtual staff C and D are dispatched to Company B, as if they were dispatched from the operating management company. In the personnel management system 1, just as real employees of Company A and Company B can consult with and make requests to other real employees via chat, they can also consult with and make requests to virtual staff. Note that the domains of Company A and Company B are separated for security reasons so that communication between them is not possible.

[0019] The information processing system according to this embodiment comprises a personnel management system 1, an operating terminal used by the operating management company that provides and operates the personnel management system 1, user terminals used by multiple users who are employees of various companies (organizations) such as Company A and Company B, and a Large-Scale Language Model (LLM) server 6. The personnel management system 1 is provided on a server in the cloud and is communicated to the operating terminal, user terminals, and Large-Scale Language Model (LLM) server 6 via a network such as an internet connection. Each component of this information processing system will be described below.

[0020] The operating terminals include personal computers and other devices installed in the offices of the operating management company that manages the server for the personnel management system 1. Operators, who are employees of this operating management company, operate the operating terminals to send and receive information to and from the personnel management system 1. Specifically, the operators operate the operating terminals to pre-configure the initial settings for each virtual staff member, as described later.

[0021] User terminals include personal computers and other devices installed in companies (Company A, Company B) that wish to utilize the services of this form of personnel management system 1, and are operated by employees of these companies. In addition to the chat service with virtual staff according to the present invention, personnel management system 1 has a variety of known functions, such as the registration of employee personnel data. Operations performed by user terminals include not only the chat service with virtual staff mentioned above, but also operations such as setting assignments and access permissions by system administrators, and changing profiles, which will be described later.

[0022] The Large-Scale Language Model (LLM) Server 6 is designed to generate responses quickly to requests from multiple clients (in this embodiment, the servers of the Human Resources Management System 1), for example, using OpenAI. The Large-Scale Language Model (LLM) Server 6 generally uses a neural network-based model. This model learns from a vast amount of data and then makes predictions for new inputs. It hosts the model, calls it upon request, and returns it as an answer. It is also possible to reuse a model that has been trained once. In this embodiment, when the Large-Scale Language Model (LLM) Server 6 receives question information, it generates answer information to the question information through the processing described later and sends it back to the user terminal.

[0023] The server of the personnel management system 1 comprises a control unit, a memory unit, and a communication unit. The control unit is, for example, a CPU (Central Processing Unit), and by executing programs stored in the memory unit, it performs functions such as receiving questions from user terminals, sending them to the Large-Scale Language Model (LLM) server 6, and obtaining answers from the Large-Scale Language Model (LLM) server 6.

[0024] Furthermore, the program executed by the control unit is not limited to one stored in the memory unit. The program executed by the control unit may be one transmitted to the control unit from an external device via the communication unit, one stored on a storage medium such as a USB memory stick provided separately on the server, or one stored on a cloud server separate from the server.

[0025] The memory unit consists of an HDD, RAM, ROM, and SSD, and various information related to the personnel management system 1 is stored in the memory unit.

[0026] The communications unit is connected to the control unit in a way that allows it to communicate with the operator's terminal, user terminals, large-scale language model (LLM) server, etc., and to send and receive signals.

[0027] Next, the concept of dispatching virtual staff to each company (Company A, Company B) will be explained based on Figure 2. The operating management company sets up a chatbot program equipped with AI functionality to act as a virtual employee, through operations from the operating terminal. To make them appear as if they were real staff that could be dispatched, each virtual staff member is assigned a name, gender, area of ​​expertise such as HR, sales, or finance, rank such as manager or associate, personality, work style, and an appropriate appearance icon. This information is stored in Profile DB (database) 2, which is stored in the memory of the server of the personnel management system 1. Based on the information stored in Profile DB 2, prompts are generated to be registered in Prompt DB 3, which will be described later, and the chatbot acting as the virtual staff member refers to Prompt DB 3 when generating responses.

[0028] Then, through user terminal operations, each company (Company A, Company B) selects the desired virtual staff, sets their assignment location as described later, and registers them in the environment of each company (Company A, Company B) in the Human Resources Management System 1. From the user terminals of each company (Company A, Company B), the virtual staff will be displayed in the Human Resources Management System 1 as if a new staff member had been dispatched.

[0029] In the example shown in Figure 2, virtual staff member A, who is well-versed in human resources, is registered to be assigned to the Human Resources Department of Company A; virtual staff member C, who is well-versed in sales, is registered to be assigned to the Sales Department of Company B; and virtual staff member D, who is well-versed in finance, is registered to be assigned to the Finance Department of Company B.

[0030] The process then proceeds to the registration of virtual staff by the system administrator of the company (Company A). This process will be explained using Figures 3 to 5. First, on the administrator interface screen of the personnel management system 1, the desired virtual staff member is selected from the pre-configured virtual staff (Step S1).

[0031] Next, the system administrator at Company A determines the assignment location and sets up the administrator for that location (see Figure 12). Furthermore, they set the scope of access for real employees who can access the virtual staff (see Figure 13). In this way, access permissions for real employees are set (Step S2). Access may be set to be available to all employees within the company, or it may be set to be available only to employees in the assigned department.

[0032] Next, the profile information pre-registered in the profile DB2 can be modified (step S3). This generates a custom prompt that reflects the persona and is registered in the prompt DB3, using the basic prompt that is pre-initialized in the prompt DB3 and the profile information stored in the profile DB2 (step S4: see Figure 4). As a result, when the chatbot, which is a virtual staff member, generates an answer, it refers to the custom prompt registered in the prompt DB3 to ensure that the wording of the answer conforms to the desired content (see Figure 4).

[0033] Next, the assigned manager uploads document files—the documents that the virtual staff member should refer to when generating answers—to a learning folder 5, which is set up for each virtual staff member (Step S5). Similarly, the learning folder 5 is stored in the storage of the server of the personnel management system 1. The document files include various documents such as contracts, approval documents, meeting minutes, invoices, receipts, purchase orders, delivery slips, presentation materials, and company regulations, and support various data formats.

[0034] Furthermore, the administrator at the assigned workplace can continue to upload document files to learning folder 5 at any time after the virtual staff member has been assigned, enabling them to generate more accurate answers. Learning folder 5 is assigned to each virtual staff member, and it can be configured so that virtual staff members assigned (registered) to other companies cannot access it, and even virtual staff members assigned (registered) to the same company cannot access it (see Figure 5).

[0035] Once the upload of the document files to learning folder 5 is complete (step S5), the virtual staff will be virtually assigned and begin operation within the company (step S6).

[0036] After the system is operational, users will be able to chat with virtual staff, who are chatbots, via a user interface as shown in Figure 6. The user list screen of the personnel management system 1 displayed on the user's terminal will show icons for virtual staff members, along with icons for real employees belonging to that company. Clicking on an icon will take the user to the virtual staff details screen, where detailed profile information will be displayed, along with a button to launch the dialogue application (chat application).

[0037] When the user activates the launch button for the conversational app (chat app), the app's user interface is displayed. When the user enters a question or request for advice from their device, a virtual employee generates and displays an answer, providing a chat service that feels like receiving an answer from a real employee in a natural conversational format.

[0038] Here, in generating an answer, the control unit of the Human Resources Management System 1 server receives a question from the user terminal and sends the question to the Large-Scale Language Model (LLM) server 6 via the communication unit. The flow up to answer generation is shown in Figure 7.

[0039] By constructing an architecture where the training folder is a RAG (Retrieval-Augmented Generation), higher search accuracy can be achieved. An index is created from the document files in training folder 5 and provided to the Large-Scale Language Model (LLM) server 6.

[0040] The Large-Scale Language Model (LLM) server 6 generates responses based on indexed information in addition to information searched on the internet. The control unit of the Human Resources Management System 1 server receives the responses from the Large-Scale Language Model (LLM) server 6 and sends them to the user terminal.

[0041] If the RAG architecture is not built, the accuracy of the answers will decrease, as shown in Figure 8. In other words, if there is an inquiry about the term "QBU," which is used colloquially within a company, there is no information available on the internet, etc., so if there is no document folder in the learning folder 5, the Large-Scale Language Model (LLM) server 6 cannot generate a specific answer. On the other hand, if there is a document folder, a specific answer can be generated.

[0042] Furthermore, just like with actual employees, it is possible to conduct on-the-job training (OJT) in advance to improve the accuracy of virtual staff responses. In other words, as shown in Figure 9, by starting with OJT and collecting and analyzing the feedback from within the company, it becomes possible to readjust the data stored in the prompt DB3 and learning folder 5, making it easier to obtain responses that meet the needs of each company.

[0043] As shown in Figure 10, after an employee finishes chatting with a virtual staff member using the dialogue app, they input a rating of "Good" or "Bad" by selecting a thumbs-up or thumbs-down icon. If "Bad" is selected, a form is displayed for further input of specific reasons. The entered rating results and reasons are stored in the feedback DB7 in the server memory of the personnel management system 1, and an analysis report is generated. The analysis report is sent to the administrator.

[0044] As shown in Figure 11, administrators who receive the analysis report can improve the response generation process of the Large-Scale Language Model (LLM) server 6 after the on-the-job training (OJT) phase by modifying custom prompts in the prompt DB3 or correcting document data in the learning folder 5 based on the suggested countermeasures in the analysis report.

[0045] Furthermore, by changing the usage fee for virtual staff from the company based on factors such as creativity and the number of cited documents, it is possible to operate in a manner that mimics an actual dispatch system where compensation varies depending on skills and abilities.

[0046] Furthermore, in order to enable virtual staff to answer questions about a wider range of tasks, editing such as adding (expanding) the tasks they are responsible for can be made. In other words, the basic prompts that are initially set in the prompt DB3 may be editable regarding the virtual staff's tasks. For example, if a new task is to be added to a virtual staff member, it can be added to that virtual staff member's prompt DB by an operator on the operating side via an operating terminal, for example, through the administrator interface on the operating side.

[0047] Specifically, virtual staff members in sales roles can be given additional sales-related tasks, such as scanning business cards, gathering customer-related news, and researching competitors for the company's products. Similarly, virtual staff members in human resources and general affairs roles can be given additional tasks, such as gathering information on changes in labor laws and purchasing business cards and supplies.

[0048] On the user terminal, information regarding the addition of a task can be received, for example, by displaying it along with information about the virtual staff, and the user can instruct the virtual staff to start the desired task from among the added tasks. Once instructed to start a task, the virtual staff may prompt the user, via a display on the user terminal, to upload the company's document files related to the task instructed to start to the learning folder.

[0049] By enabling virtual staff to perform additional tasks in specialized areas such as general affairs and sales, the number of tasks that can be solved by AI can be increased. Initially, it may be difficult to concretely imagine work scenarios utilizing AI, but by expanding the tasks that can be automated by AI-based virtual staff, the benefits of using virtual staff can be further enhanced.

[0050] Furthermore, for document files uploaded to the learning folder 5, information related to the document files may be automatically generated to facilitate learning by the Large-Scale Language Model (LLM), and hallucination may be suppressed to further improve response accuracy. Examples of information generation processes related to document files include text structuring, chunking, and metadata generation, and for example, one or more of these may be performed. Figure 14 shows an example of text structuring, chunking, and metadata generation. The generated structured text, chunks, and metadata may be stored in a document database provided in the storage unit of the server of the personnel management system 1. In addition, the above-mentioned text structuring, chunking, and metadata generation may be performed by executing a program for these purposes, or by a generation AI configured to function on an operating terminal or the like.

[0051] The Large-Scale Language Model (LLM) searches for document files in the training folder 5 by referencing, for example, at least one of the structured sentences, chunks, and metadata stored in the document database, and uses the retrieved document files to generate answers.

[0052] The document file names in the learning folder 5, provided to the Large-Scale Language Model (LLM) server 6 for answer generation, can be displayed on the user terminal along with the answers. Furthermore, the number of citations for each document file may be stored in a citation count database located in the storage unit of the personnel management system 1 server.

[0053] Furthermore, in the information processing system of the present invention, in order to enable a more objective assessment of answer accuracy and contribute to improving answer accuracy, an evaluation of answer accuracy may be performed. This evaluation can be performed for each document file in the learning folder 5. Specifically, it can be performed as shown in the flow chart in Figures 15 and 16.

[0054] First, the assigned manager instructs a designated generating AI to generate questions (Figure 15). The content of the questions is not particularly limited, but for example, they could be common questions that are typical in a particular field. The answers to these questions may also be generated based on document files stored in the learning folder 5. The questions generated by the generating AI are directed to the virtual staff of the information processing system of the present invention.

[0055] Furthermore, the answers obtained by the virtual staff of the information processing system of the present invention can be evaluated by judgment using a predetermined generating AI. Figure 16 shows an example of judgment and evaluation. In the judgment and evaluation of this example, the judgment is made and a score is calculated based on the answers of the information processing system of the present invention to questions generated by a predetermined generating AI. The generating AI that generates, judges, and evaluates the questions is not particularly limited and can be, for example, a generating AI configured to function on an operating terminal. The generating AI that generates, judges, and evaluates the questions may be the same or different generating AIs. Furthermore, the evaluation of response accuracy, as illustrated in Figure 16, may be performed for each document file in the learning folder 5. In this case, a score can be calculated for each document file.

[0056] On the administrator page (Figure 17) for managing the uploading and deletion of document files to and from learning folder 5, the number of citations can be displayed based on the number of citations for each document file stored in the citation count DB (item enclosed in box 20a). The administrator page may also display words related to the document file included in the created metadata (item enclosed in box 20b). Furthermore, if a score regarding the accuracy of each document file is calculated, the administrator page may display a score based on that score (item enclosed in box 20c). This score may be displayed as is, or it may be simplified to low, medium, and high.

[0057] Furthermore, if there are multiple virtual staff members with different duties within the same company, they may collaborate to process inquiries. The duties of each virtual staff member may be stored in a virtual staff database located in the storage section of the server of the personnel management system 1. When a virtual staff member receives instructions from a user, if the instructions include processing outside their scope of duties, they refer to the virtual staff database to search for a virtual staff member who can handle the request. If a suitable virtual staff member exists, they are instructed to perform the processing outside their scope of duties, and the result is notified to the user. Figure 18 shows an example of when multiple virtual staff members collaborate to process a request.

[0058] It should be noted that the present invention is not limited to the embodiments described above, and various modifications and alterations are possible by those skilled in the art within the scope of the technical thinking disclosed herein, and various variations are included. Furthermore, the embodiments described above are explained in detail for the purpose of clearly illustrating the present invention, and are not necessarily limited to those having all the configurations described. [Explanation of symbols]

[0059] 1…Personnel management system, 2…Profile DB, 3…Prompt DB, 5…Learning folder, 6…Large-scale language model (LLM) server, 7…Feedback DB

Claims

1. An information processing system comprising a user terminal used by users, an existing human resources management system having a chat function between users, a terminal used by the operator that provides the existing human resources management system to users via a cloud server, and a large-scale language model server, wherein a chatbot that receives and answers questions from users is installed on the existing human resources management system, The storage unit of the aforementioned existing human resources management system includes a profile DB for registering basic information of the chatbot and a prompt DB for registering information related to the creation of prompts when the chatbot responds. The chatbot includes means for sending questions from a user terminal to a large-scale language model server, means for receiving answers from the large-scale language model server, and means for generating answers to send to the user terminal by referring to the prompt DB. An information processing system characterized by the following:

2. The information processing system according to claim 1, characterized in that the chatbot is displayed as a user on the existing human resources management system based on the aforementioned basic information.

3. The information processing system according to claim 2, characterized in that the basic information can be edited by the user terminal, and the information related to prompt creation registered in the prompt DB is updated based on the edited basic information.

4. The information processing system according to claim 3, characterized in that the storage unit has a learning folder where document files are saved by the operation of the user terminal, the document files are indexed, and provided to the large-scale language model server.

5. The information processing system according to claim 3, characterized in that the information related to prompt creation registered in the prompt DB is editable for business-related information, and when business-related information is added, the chatbot can be instructed from the user terminal to generate a response for the added business.

6. The information processing system according to claim 4, further comprising a document database in the storage unit for registering information related to the document file, wherein the response from the large-scale language model server is generated via referencing the document database.

7. The information processing system according to claim 6, characterized in that the response generated by the information processing system is evaluated by a determination using a generating AI.

8. The information processing system according to any one of claims 1 to 7, characterized in that there are multiple chatbots, which can be installed for each organization to which a user belongs, and each user in each organization can be configured to access the chatbot.

9. The information processing system according to any one of claims 1 to 7, characterized in that multiple chatbots can be installed in one organization to which a user belongs, and one chatbot generates responses based on the collaboration of multiple chatbots to send to the user's terminal.

Images