Chatbot system
The chatbot system integrates automated and human responses through a monitoring server to address the challenge of inadequate chatbot responses, enhancing efficiency and service quality by enabling operator intervention.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- SOLBRAIN CO LTD
- Filing Date
- 2025-04-24
- Publication Date
- 2026-05-20
AI Technical Summary
Existing chatbot systems struggle to provide high-quality customer service when they are unable to adequately respond to user inquiries, necessitating a seamless integration of automated and human responses.
A chatbot system comprising a chat server, automated response server, web server, and chat monitoring server that determines the need for human intervention, temporarily suspending automated responses and enabling operators to intervene, thereby facilitating efficient human support.
Enhances the efficiency and quality of service by allowing operators to seamlessly transition between automated and human responses, ensuring high-quality customer interactions.
Smart Images
Figure 0007862813000001_ABST
Abstract
Description
Technical Field
[0005] , , ,
[0006] , , system , , , ,
[0001] The present invention relates to a chatbot system .
Background Art
[0002] In recent years, with the development of AI and NLP technologies, an automatic response system called a chatbot has been put into practical use for handling customer inquiries in enterprises (see, for example, Patent Document 1). By analyzing the customer's inquiry intention and automatically generating an appropriate answer, this technology has realized an efficient hybrid customer response system that can immediately respond to routine inquiries and cooperate with human operators.
Prior Art Documents
[0007] According to the present invention, if a situation arises where the chatbot is unable to provide a satisfactory answer, an operator can intervene in the chat to facilitate human support. This enables both the efficiency of the chatbot and the high quality of service provided by the operator. [Brief explanation of the drawing]
[0008] [Figure 1] This is a schematic diagram of a chatbot system according to one embodiment of the present invention. [Figure 2] This is a schematic diagram illustrating an example of the transition in the chat response state. [Figure 3] This figure shows an example of an operator's terminal screen during chat monitoring. [Figure 4] This figure shows an example of an operator's terminal screen while an operator is intervening in a chat. [Modes for carrying out the invention]
[0009] The embodiments of the present invention will be described in detail below, with reference to the drawings as appropriate. However, unnecessary details may be omitted. For example, detailed explanations of already well-known matters and redundant explanations of substantially identical configurations may be omitted. This is to avoid the following explanation becoming unnecessarily verbose and to facilitate understanding by those skilled in the art. The inventors provide the accompanying drawings and the following explanation so that those skilled in the art can fully understand the present invention, and do not intend to limit the subject matter described in the claims by means of these.
[0010] <<Embodiment>> Figure 1 is a schematic diagram of a chatbot system according to one embodiment of the present invention. In this embodiment, the chatbot system 100 receives various inquiries from users via chat and the bot automatically provides answers. Such a chatbot system 100 can accept inquiries from users 24 hours a day, 365 days a year and provide appropriate answers.
[0011] The chatbot system 100 includes, as its main components, a chat server 10, an automated response server 20, a web server 30, and a chat monitoring server 40. Each of these servers can be composed of one or more computer devices having a processor, memory, storage, and interfaces.
[0012] Each server is connected via a network (not shown in the diagram) and communicates with each other through an API (Application Programming Interface) to form the chatbot system 100. Multiple servers may also be implemented on the same computer device.
[0013] The network is a comprehensive network that includes public networks such as the internet and carrier communication networks, as well as private networks such as WANs and LANs. Furthermore, one or more user terminals 200, one or more operator terminals 300, and external databases 400 can be connected to the network.
[0014] The main role of the chat server 10 is to communicate with the browser or chat messaging application of the user terminal 200, receive chat messages of user queries from the user terminal 200, and send reply chat messages to the user terminal 200. The user terminal 200 can be a personal computer, smartphone, tablet, etc. The chat server 10 can communicate with multiple user terminals 200 and conduct individual chats with them.
[0015] The chat server 10 can record chat messages for each chat in internal or external storage. Specifically, the chat server 10 records chat messages along with user session information, associating them with the chat ID. The chat server 10 can provide the recorded chat messages in response to requests from external sources. For example, as described later, the chat server 10 can provide chat messages for any chat to the web server 30 in response to a request from the web server 30.
[0016] The automated response server 20 receives user queries from the chat server 10, automatically generates responses, and returns them to the chat server 10. Specifically, the automated response server 20 includes an LLM (Large Language Model) engine 21 and a RAG (Retrieval Augmented Generation) component 22. The LLM engine 22 and the RAG component 21 may each be configured on separate servers or computer devices. In that case, they cooperate with each other via an API to function as the automated response server 20.
[0017] The LLM engine 21 processes the user queries transferred from the chat server 10 and generates responses in cooperation with the RAG component 22 as necessary. More specifically, the RAG component 22 performs a vector search on the external database 400 to collect information related to the user query and synthesize the context. The external database 400 includes in-house manuals, CRM (Customer Relationship Management) systems, web pages, scripts, and the like. The LLM engine 21 provides functions for inference processing and response generation, and integrates the user query with the context collected, retrieved, and synthesized by the RAG component 22 from the external database 400 to generate a response. In addition, in order to understand the context of the chat and automatically generate appropriate responses, the LLM engine 21 temporarily stores a series of user queries.
[0018] The main role of the web server 30 is to display each ongoing chat on the chat server 10 on the operator terminal 300 and receive responses from the operator terminal 300 and transfer them to the chat server 10. That is, the web server 30 provides a GUI for managing the chats in the chatbot system 100 on the operator terminal 300. The operator terminal 300 is a personal computer, smartphone, tablet terminal, or the like. The web server 30 can communicate with a plurality of operator terminals 300 and provide a chat management GUI individually.
[0019] The chat monitoring server 40 monitors each ongoing chat on the chat server 10 in real time and provides a function to determine whether human intervention is required for each chat based on the content of the chat message. Whether human intervention is required for a chat can be determined according to pre-defined rules. For example, the chat monitoring server 40 performs morphological analysis on the chat message. If it contains keywords related to the price of a product or service, quotation or payment, keywords used when the user is dissatisfied, or the tone at the end of the message, it can be determined that human intervention in the chat is necessary.
[0020] Alternatively, the necessity of manned response for chat can be determined using a neural network that has learned examples of chat messages. In this case, when the chat monitoring server 40 inputs the chat messages of each chat into the neural network, the neural network estimates the necessity of manned response for the chat and outputs a numerical value, and if the numerical value exceeds a threshold value, it can be determined that manned response is necessary.
[0021] The chat monitoring server 40 further provides a function to the chat server 10 for the chat determined to require manned response to suspend the automatic response by the automatic answer server 20 and transmit the answer from the operator terminal 300 transferred from the Web server 30 to the user terminal 200. That is, the chat monitoring server 40 automatically determines the necessity of manned response for chat, and when it determines that manned response is necessary, it interrupts the automatic answer of the bot to enable manned answer by the operator. As a result, basically, the operator leaves the chat response to the automatic answer of the bot and monitors it, and only intervenes in the chat determined by the chat monitoring server 40 to require manned response. As will be described later, the operator can intervene in the chat and switch to manned answer if it is determined necessary even during the automatic answer of the bot.
[0022] ≪Chat Response State Transition≫ In the chatbot system 100, the chat response states are (a) during automatic answer by the bot, (b) waiting for response by the operator as manned response is required, and (c) during manned answer by the operator.
[0023] Figure 2 is a schematic diagram showing an example of the transition of chat response states. Let's assume that there are currently three chats A, B, and C in progress. When the automated response server 20 is currently providing automated responses to chats A, B, and C, the state of chats A, B, and C is (a) Bot response in progress (AI response in progress). Now, let's assume that the chat monitoring server 40 determines that human response is required for chats B and C, and they are placed in a queue waiting for operator assistance. At this time, the state of chats B and C transitions to (b) Waiting for operator response (response required). When an operator intervenes and begins providing a human response to chats B and C, the state of chats B and C transitions to (c) Operator response in progress (OP response in progress). Finally, when the human response by the operator for chats B and C is completed and the automated response returns to the bot, the state of chats B and C transitions to (a) Bot response in progress (AI response in progress).
[0024] <<Example of operator terminal screen>> Figure 3 shows an example of an operator terminal screen during chat monitoring. For example, the left pane 301 of the operator terminal 300 screen displays a list of currently active chat screens. In this example, four chat screens are displayed in a 2x2 grid within pane 301. If there are too many active chats to display all of them within pane 301, the hidden chat screens can be displayed within pane 301 by using the up / down or left / right scroll bars. Additionally, the message history of each chat can be viewed by using the up / down scroll bar on each chat screen.
[0025] Each chat screen displays the status of that chat. For example, "AI handling" is displayed when an automated response is being handled by a bot, "Response required" is displayed when a response from an operator is awaiting, and "Operator handling" is displayed in an appropriate location on each chat screen (for example, the upper right corner of the chat screen) when a human operator is providing a response. Operators can see the status of a chat to know what its current status is.
[0026] For chats where a bot is automatically responding, a "Stop Generation" button 303 may be displayed in an appropriate location on the chat screen (for example, the lower right corner of the chat screen). Clicking this button 303 forces the chat's response status to change from (a) bot-assisted to (c) operator-assisted. In other words, when the "Stop Generation" button 303 is clicked, the web server 10 notifies the chat server 10 that an operator has intervened in the chat, and the chat server 10 temporarily suspends the automatic response by the automatic response server 20. This allows an operator to intervene in a chat where a bot is automatically responding.
[0027] For example, the chat screen in the upper left of pane 301 shows that the operator has clicked the "Stop Generation" button 303 to intervene in the chat and is manually entering a response.
[0028] When an operator intervenes in a chat, the operator may create a response themselves, or they may have the automated response server 20 create a response for them. In other words, the web server 30 can receive the response from the automated response server 20 from the chat server 10, display it on the operator terminal 300, and, according to the operator's instructions, accept the response from the automated response server 20 or a response modified or newly created by the operator and forward it to the chat server 10.
[0029] For example, in the chat screen in the upper right corner of pane 301, a suggested response from the automated response server 20 is displayed. The operator can modify this suggested response by clicking the "Edit" button 304.
[0030] When the operator finishes providing a human response and wants to return to the automated response server 20, the operator should click the "Resume Generation" button 305 displayed on the chat screen.
[0031] The right pane 302 of the operator terminal 300 screen displays a queue of chats awaiting operator assistance. For example, chats awaiting operator assistance are displayed in a stacked banner format. Each banner displays the chat recipient, the time the automated response stopped, and the first two lines of the message (or a summary of the message) that caused the chat monitoring server 40 to require human assistance.
[0032] In the queue displayed on the screen, banners are shown in ascending order of the time when the automated chat response stopped, but they can also be shown in descending order. For chats in the queue that have been left unattended for a long time, you can make them stand out by, for example, changing the background color of the banner or making it blink.
[0033] An operator can intervene in a chat waiting for operator assistance by clicking on any banner. Figure 4 shows an example of the operator terminal screen while an operator is intervening in a chat. The right pane 302 changes from the chat queue to the screen of the intervened chat, and the operator can edit and send the message by operating the operator terminal 300.
[0034] When an operator selects a banner from the queue, the chat's status changes from (b) waiting for operator assistance to (c) in progress, and the banner disappears from the queue. This is reflected in other operator terminals 300, allowing operators to always check the latest queue.
[0035] To view the past messages of that chat, the user can use the up and down scroll bars on the chat screen to see the chat's message history. If the message history has not been retrieved to the operator terminal 300, the web server 30 retrieves the message history of the selected chat from the chat server 10 and displays it on the operator terminal 300, in response to the operator's request.
[0036] When intervening in a chat waiting for an operator response, the operator may create a draft response themselves, or they may have the automated response server 20 create a draft response as an aid. In other words, the web server 30 can receive the draft response from the automated response server 20 from the chat server 10 and display it on the operator terminal 300, and, according to the operator's instructions, accept the draft response from the automated response server 20 or a draft response that the operator has modified or newly created and forward to the chat server 10.
[0037] For example, in the chat screen of pane 302, a suggested response from the automated response server 20 is presented. If the operator accepts the suggested response, they click the "Send as is" button 306, which sends the suggested response to the user terminal 200 via the web server 30 and the chat server 10. On the other hand, if the operator wants to modify the suggested response from the automated response server 20, they can click the "Edit and send" button 307. In this case, they can modify the suggested response or create a new suggested response themselves and send it to the user terminal 200.
[0038] <Effects> As described above, according to this embodiment, if a situation arises where the chatbot is unable to provide a satisfactory answer, an operator can intervene in the chat and provide human support. This enables both the efficiency of the chatbot and the high quality of service provided by the operator.
[0039] ≪Variations≫ The chat monitoring server 40 may also determine the importance of a chat when deciding whether or not it requires human intervention. For example, if a neural network is used to determine whether or not a chat requires human intervention, the importance can be determined according to the output value of the neural network. If importance is also considered, the queue of chats waiting for operator assistance, displayed on the operator terminal 300 screen, may be sorted in order of importance.
[0040] As described above, embodiments have been explained as examples of the technology in the present invention. For this purpose, accompanying drawings and a detailed description have been provided. Therefore, among the components described in the accompanying drawings and detailed description, there may be not only components that are essential for solving the problem, but also components that are not essential for solving the problem, in order to illustrate the above technology. For this reason, the mere fact that these non-essential components are described in the accompanying drawings and detailed description should not be immediately assumed to be essential. Furthermore, since the above embodiments are for the purpose of illustrating the technology in the present invention, various changes, substitutions, additions, omissions, etc., can be made within the scope of the claims or equivalents. [Explanation of symbols]
[0041] 100 Chatbot Systems 10 Chat Servers 20 Automated response servers 21 LLM engine 22 RAG Component 30 Web Servers 40 Chat monitoring servers 200 user terminals 300 Operator Terminals 400 External Databases
Claims
1. A chat server having the function of receiving chat messages of user queries in text data from a user terminal and the function of sending chat messages of answers in text data to the user terminal, An automated response server receives the user query from the chat server, automatically generates a response in text data format, and returns it to the chat server. A web server having the function of displaying each ongoing chat on the chat server on the operator terminal and the function of receiving text data responses from the operator terminal and forwarding them to the chat server, The system includes a chat monitoring server that monitors each ongoing chat in real time on the aforementioned chat server and determines whether human intervention is required for each chat based on the content of the chat messages in text data, When the chat monitoring server determines that a human intervention is necessary for a chat, it instructs the chat server to temporarily suspend the automated response by the automated response server and send the response as text data from the operator terminal, which is transferred from the web server, to the user terminal. The Web server displays the chat screen of each ongoing chat on the chat server in a list format on the operator terminal, clearly indicating the chat response status as either automated response, human response, or awaiting human response. The aforementioned chat screen during automated response includes a button to switch the chat response from automated to human response. The aforementioned chat screen during a human-assisted response includes a button to switch the chat response from human-assisted to automated. A chatbot system characterized by the following features.
2. The chatbot system according to claim 1, wherein the Web server displays a queue of chats on the operator terminal that are awaiting human intervention, and the operator selects a chat from the queue and receives a prepared response by operating the operator terminal and forwards it to the chat server.
3. The chatbot system according to claim 2, wherein the web server displays chats that have been left unattended for a long period of time in the queue in a prominent manner within the queue.
4. The chatbot system according to claim 2 or 3, wherein the Web server retrieves the message history of the selected chat from the chat server in response to a request from the operator and displays it on the operator's terminal.
5. The aforementioned automated response server generates a proposed response for a chat that is being handled by a human. The chatbot system according to any one of claims 1 to 3, wherein the Web server receives a proposed answer from the automated response server from the chat server and displays it on the operator terminal, and accepts the proposed answer from the automated response server or a proposed answer created by the operator in accordance with the operator's instructions and forwards it to the chat server.
6. The aforementioned automated response server, The Large Language Model (LLM) engine that processes the user queries, A chatbot system according to any one of claims 1 to 3, comprising a RAG (Retrieval Augmented Generation) component that works in conjunction with the LLM engine to search an external database and collect information related to the user query.