Method for providing ai-based customer support service and contact center system
Patent Information
- Authority / Receiving Office
- KR · KR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-02-05
- Publication Date
- 2026-08-12
Smart Images
Figure P1020250014510_ABST
Abstract
Description
Technology Field
[0001] This disclosure relates to a customer service center. Background Technology
[0002] Companies or organizations providing services to customers establish customer service centers to communicate with them. The network is configured so that calls received on the main number are forwarded to agents, enabling them to make calls to customers. Customer service centers utilize a Private Branch Exchange (PBX) to distribute incoming inbound calls to agents.
[0003] Recently, technology capable of responding to customer questions in real time through bots is being introduced. However, bots replace simple tasks that can be processed according to established scenarios, such as frequently asked questions (FAQs) or simple requests, and agents still have to handle complex requests that are not included in the scenarios. The problem to be solved
[0004] The present disclosure provides a method for providing an artificial intelligence-based customer consultation service using a virtual counselor and a customer consultation center system. means of solving the problem
[0005] A customer service center system according to some embodiments comprises: a dialogue processor that generates a response from a virtual agent based on customer utterance text; and an automatic response server that converts customer utterance voice transmitted from a customer terminal into text to generate said customer utterance text, transmits a response request including said customer utterance text to said dialogue processor, generates a synthesized voice for said response received from said dialogue processor using a voice model of said virtual agent, and transmits said synthesized voice to said customer terminal. The dialogue processor extracts interaction data in which a human agent converses with an arbitrary customer and causes an artificial intelligence model to generate said response for said customer utterance text by referencing said interaction data.
[0006] The above automatic response server can determine the virtual agent to respond to the customer among a plurality of virtual agents and transmit the response request, which includes the customer utterance text and virtual agent information, to the conversation processor.
[0007] The above dialogue processor can extract interaction data of the human counselor mapped to the virtual counselor through the virtual counselor information.
[0008] The above automatic response server may determine whether the virtual agent responds to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer.
[0009] The above automatic response server can connect inbound calls received from the switchboard and the customer's terminal when it is determined that a human agent will respond to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer.
[0010] The above automatic response server includes a voice model for each virtual agent, and generates the synthesized voice using the voice model of the virtual agent among the voice models for each virtual agent, and the voice model for each virtual agent can be generated based on the voice data of a human agent mapped to the corresponding virtual agent.
[0011] The above dialogue processor can extract information related to the customer utterance text from a source database storing source data necessary for consultation, and enable the artificial intelligence model to generate the response to the customer utterance text by referencing the interaction data and the information related to the customer utterance text.
[0012] The above dialogue processor may call a specific function among the functions defined for interoperability with the legacy system to process a customer request analyzed from the customer utterance text, obtain a processing result from the legacy system, and enable the artificial intelligence model to generate the response to the customer utterance text by referring to the interaction data and the processing result.
[0013] The above dialogue processor can call the specific function after obtaining the arguments of the specific function from the customer utterance data or the subsequent utterance data of the customer utterance data.
[0014] The above subsequent utterance data can be obtained by inquiring of the customer for at least one argument required for calling the specific function. A sentence inquiring of the customer for the at least one argument can be generated by the artificial intelligence model. Synthesized speech for the sentence can be generated via the automatic response server and transmitted to the customer's terminal.
[0015] A method of operation of an automatic response server according to some embodiments comprises: a step of converting a customer utterance voice transmitted from a customer terminal into text to generate a customer utterance text; a step of transmitting a response request including the customer utterance text to a dialogue processor; a step of receiving a response for the customer utterance text from the dialogue processor; a step of generating a synthesized voice for the response using a voice model of a virtual agent; and a step of transmitting the synthesized voice to the customer terminal.
[0016] The step of transmitting the above response request to a conversation processor may involve determining the virtual agent to respond to the customer among a plurality of virtual agents, and transmitting the above response request, which includes the customer utterance text and virtual agent information, to the conversation processor.
[0017] The step of transmitting the above response request to a conversation handler may determine whether the virtual agent responds to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer.
[0018] The above method of operation may further include the step of connecting an inbound call received from the switchboard and the customer's terminal when it is determined that a human agent will respond to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer.
[0019] The step of generating the above-mentioned synthesized voice may generate the synthesized voice using the voice model of the virtual counselor among the voice models for each virtual counselor. The voice model for each virtual counselor may be generated based on the voice data of a human counselor mapped to the corresponding virtual counselor.
[0020] A method of operation of a dialogue processor according to some embodiments may include: receiving customer utterance text from an automatic response server; extracting interaction data from an interaction database in which a human counselor converses with an arbitrary customer; analyzing the customer utterance text to determine the type of customer request; if the customer request is a first type of request, extracting information related to the customer utterance text from a source database storing source data necessary for counseling; obtaining a first response generated for the customer utterance text by referencing the interaction data and the information related to the customer utterance text through linkage with an artificial intelligence model; and transmitting the first response to the automatic response server.
[0021] The above method of operation may further include, when the customer request is a second type of request requiring integration with a legacy system, a step of calling a specific function for processing the customer request among functions defined for integration with the legacy system, a step of obtaining a processing result from the legacy system, a step of obtaining a second response generated for the customer utterance text by referring to the interaction data and the processing result through integration with the artificial intelligence model, and a step of transmitting the second response to the automatic response server.
[0022] The step of calling the specific function may involve obtaining the arguments of the specific function from the customer utterance data or the subsequent utterance data of the customer utterance data, and then calling the specific function.
[0023] The above subsequent utterance data can be obtained by inquiring of the customer for at least one argument required for calling the above specific function through integration with an artificial intelligence model.
[0024] The above first response can be output as a synthesized voice using a voice model of a virtual counselor. Effects of the invention
[0025] According to the embodiment, by dynamically generating a response that follows the tone and conversation patterns of a human counselor by referencing the interaction data of a human counselor, the consistency and accuracy of the response can be guaranteed, natural language sentences suitable for counseling work and natural can be generated, and synthetic speech suitable for counseling work can be generated through a voice model of a human counselor.
[0026] According to an embodiment, by having a virtual counselor respond to customers instead of a human counselor, customers can be connected with a counselor without waiting time and receive counseling services through conversation rather than chat.
[0027] According to the embodiment, the customer service center system can increase work processing efficiency through virtual agents and alleviate customer complaints caused by waiting times, thereby increasing customer satisfaction while reducing operating costs such as labor and training costs, and can provide high-quality consultation services by assigning human agents to difficult-to-process requests or personalized requests.
[0028] According to the embodiment, by having virtual counselors with various voice features, various conversation patterns, and various work processing histories respond to customers, diversity in counseling services can be provided. Brief explanation of the drawing
[0029] FIG. 1 is a configuration diagram of a customer consultation center system according to one embodiment. FIG. 2 is a diagram illustrating a procedure for connecting a human counselor according to one embodiment. FIG. 3 is a diagram illustrating a method of providing customer consultation services by a virtual counselor according to one embodiment. FIG. 4 is a diagram illustrating an STT module and a TTS module according to one embodiment. FIGS. 5 and FIGS. 6 are drawings illustrating, respectively, the generation of a response to a customer utterance text according to one embodiment. FIG. 7 is a method of operation of a customer consultation center system according to one embodiment. Specific details for implementing the invention
[0030] Embodiments of the present disclosure are described below with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present disclosure may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present disclosure in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.
[0031] Throughout the specification, when a part is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.
[0032] In the description, "transmission," "delivery," or "provision" may include not only direct transmission, delivery, or provision, but also indirect transmission, delivery, or provision through another device or by using an alternative route.
[0033] In the description, expressions written in the singular may be interpreted as singular or plural unless explicit expressions such as "one" or "singular" are used.
[0034] In the description, the order of operations listed in the flowchart may be changed, multiple operations may be merged, some operations may be split, and specific operations may not be performed.
[0035] In the description, various devices are composed of hardware including at least one processor, memory, communication device, etc., and a computer program that is executed in combination with the hardware is stored in a designated location. The hardware has a configuration and performance capable of executing the method of the present disclosure. The computer program includes instructions that implement the method of operation of the present disclosure described with reference to the drawings, and executes the present disclosure in combination with hardware such as a processor and memory.
[0036] FIG. 1 is a configuration diagram of a customer consultation center system according to one embodiment, FIG. 2 is a diagram explaining a procedure for connecting a human counselor according to one embodiment, FIG. 3 is a diagram explaining a method for providing customer consultation services by a virtual counselor according to one embodiment, and FIG. 4 is a diagram explaining an STT module and a TTS module according to one embodiment.
[0037] Referring to FIG. 1, a customer service center system (100) provides a customer service service in which a virtual agent (VAgent) responds to a customer. In the present disclosure, the virtual agent can be implemented by synthesizing a response to a customer utterance using designated voice features and then transmitting the synthesized voice to a customer terminal (10). The response is generated based on customer utterance data and virtual agent information, and a conversational response can be generated through an artificial intelligence model such as a Large Language Model (LLM). That is, the customer service center system (100) can be implemented to provide a series of procedures that provide an interaction environment between a customer and an agent by generating a response and synthesizing the response using designated voice features and transmitting it. In the present disclosure, the response may include not only a sentence answering a customer request, but also various sentences containing content sent to the customer, such as a sentence inquiring of the customer to process the customer request, or a sentence for proceeding with a conversation with the customer.
[0038] The customer consultation center system (100) may assign virtual counselors with various voice characteristics, various conversation patterns, and various work processing histories to provide diversity in customer consultation services. Like a digital twin or an avatar, the virtual counselor may be implemented to converse with customers by mimicking the characteristics of a human counselor. Here, the characteristics of a human counselor may include voice characteristics such as tone of voice and intonation, conversation patterns, and work processing history. The virtual counselor may be implemented to converse with customers by mimicking the voice characteristics, conversation patterns, and work processing history of multiple human counselors; however, in this disclosure, it is assumed that the virtual counselor is implemented to converse with customers by mimicking the voice characteristics, conversation patterns, and work processing history of a designated human counselor. The voice characteristics, conversation patterns, and work processing history of a human counselor may be transferred to the virtual counselor through interaction data between the human counselor and the customer. In this disclosure, the interaction data may include conversation content exchanged between the human counselor and the customer, and the conversation content may be stored as text, may consist only of the counselor's speech data, or may consist of speech data of both the counselor and the customer. Additionally, interaction data may include voice data recording the conversation exchanged between a human agent and a customer. Interaction data in text form can be referenced when generating responses. Voice data can be used for training when creating a voice model.
[0039] The customer service center system (100) can be implemented to select excellent counselors with work capabilities above a standard among human counselors (e.g., counselors whose customer satisfaction is evaluated above a standard, counselors with work experience above a standard) and to have virtual counselors who mimic their characteristics provide customer service. The customer service center system (100) grants the excellent counselors the authority to use the virtual counselor function, and the excellent counselors can activate / deactivate the virtual counselor function in the counselor application. In this case, the activation of the virtual counselor is controlled by the human counselor, and a work transfer to the human counselor may be performed during work processing by the virtual counselor, or a work transfer to the virtual counselor may be performed during work processing by the human counselor. For example, if a customer requests a transfer to a human counselor, if the request is outside the scope of the virtual counselor's work, or if it is difficult to generate an accurate response, a call assigned to the virtual counselor may be routed to the human counselor. Human counselors receive the content of the conversation between the virtual counselor and the customer through a counselor terminal and can continue the conversation with the customer based on this, and customers can continue the conversation without a sense of disconnect by talking to a human counselor with the same voice while talking to the virtual counselor.
[0040] The customer service center system (100) can be implemented to assign an agent to an inbound call through intelligent call routing, and the agent can be a virtual agent or a human agent. Additionally, the customer service center system (100) can be implemented to support switching between a virtual agent and a human agent through intelligent call routing.
[0041] The customer service center system (100) can generate at least one virtual counselor to perform the corresponding work for each business field. In this case, a virtual counselor can be assigned according to the business field corresponding to the customer request.
[0042] The customer consultation center system (100) is described as providing customer consultation services through human counselors or virtual counselors, but it may also be implemented to provide customer consultation services through virtual counselors only. The customer consultation center system (100) can be constructed in various ways depending on the services provided, but the present disclosure focuses on devices that provide AI-based customer consultation services using virtual counselors.
[0043] The customer service center system (100) includes an automatic response server (110) that receives inbound calls sent from a customer terminal (10), a counselor connection manager (120), a dialogue processor (130), and an artificial intelligence model (140), and may include an interaction database (150) that stores counselor interaction data and a source database (160) that stores source data required for counseling. The customer service center system (100) may further include a voice model generator (170) that trains a voice model to generate a synthetic voice of a virtual counselor. In the description, the automatic response server (110) is referred to as an IVR (Interactive Voice Response) server.
[0044] The counselor connection manager (120) can be implemented to basically manage the status of human counselors and sequentially assign counselors according to the order of calls registered in the counseling request queue, and can be implemented to perform tasks such as assigning virtual counselors, managing the virtual counselor function status (enabled / disabled), and switching between virtual counselors and human counselors.
[0045] The artificial intelligence model (140) can be implemented as a Large Language Model (LLM). In the description, it is explained that the IVR server (110) performs speech synthesis, and the artificial intelligence model (140) can be implemented to generate response text and synthesize speech for it to output. In the description, it is assumed that in order to efficiently process multiple inbound calls simultaneously, a dialogue processor (130) interacts with the artificial intelligence model (140) to generate response text for customer requests, and the media session with the customer terminal is initiated by the IVR server (110) acting as the endpoint to generate synthesized speech for the response data.
[0046] The IVR server (110) is an interface system implemented to receive inbound calls and respond to customer requests, and can perform routing with the determined agent until a human agent or virtual agent is connected, or can be implemented so that a virtual agent starts the initial response to the inbound call.
[0047] When it is determined that a human agent will handle an inbound call, the IVR server (110) can connect the call to a Private Branch Exchange (PBX) (180) that creates a call path with the agent terminal (not shown), and subsequently, the customer and the human agent can converse through the call path connecting the customer terminal and the agent terminal. Since the human agent is assigned according to the inbound call queue, the customer must wait for a certain amount of time until connected to an agent.
[0048] Referring to FIG. 2, in the case where it is decided that a human agent will handle an inbound call, a call sent by a customer through a customer terminal is received by the IVR server (110) (①), and the IVR server (110) connects the call to a switchboard (180) to connect with a human agent (②). The switchboard (180) registers the call for consultation in the consultation request queue of the agent connection manager (120) and waits (③), and when it receives a request for connection with an assigned agent terminal from the agent connection manager (120) (④), it signals so that the waiting call can be connected to the agent terminal (⑤).
[0049] Meanwhile, when it is determined that a virtual agent will respond to an inbound call, the IVR server (110) can transmit a response request containing customer speech text to a conversation handler (130), synthesize the response received from the conversation handler (130) with the voice characteristics of the virtual agent, and transmit the synthesized voice to the customer terminal (10). The response request may further include virtual agent information, thereby enabling the differentiation of virtual agents and providing diversity of virtual agents. That is, even with the same answer, different responses with different tones or conversation patterns can be provided depending on the characteristics of each virtual agent identified through the virtual agent information. The virtual agent information may include identification information used to extract interaction data of a human agent to be imitated by the virtual agent from the agent interaction database (150). The identification information may be a virtual agent identifier or a human agent identifier matched to the virtual agent. The virtual agent information may further include information indicating the characteristics of the virtual agent (gender, age group, tone of voice, conversation pattern, work processing history, etc.).
[0050] As human counselor interaction data accumulates and new interaction data is added, an artificial intelligence model (140) based on a generative language model can generate a response from a virtual counselor by referencing the human counselor's interaction data. Due to the nature of the generative language model, which generates response sentences in various forms, there is a possibility of hallucination occurring. Therefore, the customer service center system (100) can provide guidelines for the artificial intelligence model (140) to generate a response sentence for a customer's utterance by referencing the human counselor's interaction data.
[0051] Referring to FIG. 3, in the case where it is decided that a virtual agent will respond to an inbound call, a call made by a customer through a customer terminal is received by the IVR server (110) (①), and the IVR server (110) can obtain a response by sending a response request containing customer speech text and virtual agent information to a conversation processor (130) (②) (③). The IVR server (110) can synthesize the response with the voice characteristics of the virtual agent and send the synthesized voice to the customer terminal (④). Here, if a virtual agent can be selected from among multiple virtual agents, the virtual agent assignment is handled by the IVR server (110), or the IVR server (110) can request virtual agent assignment from the agent connection manager (120) and obtain the assigned virtual agent information (①-1, ①-2). Unlike the consultation procedure through a human agent, if there are sufficient computing resources for the virtual agent function, the customer can proceed with the consultation procedure without waiting time.
[0052] The timing and method of determining which agent to handle the customer between a human agent and a virtual agent may vary. For example, a human agent or a virtual agent may be determined by the customer's choice. When an inbound call is received, the IVR server (110) may request a selection of a human agent or a virtual agent, and in the case of consultation with a human agent, it may provide information on the waiting time.
[0053] When an inbound call is received, the IVR server (110) initiates an initial response by a virtual agent. If the customer agrees to the virtual agent's response, the virtual agent proceeds with the consultation; otherwise, the IVR server proceeds with the procedure to connect to a human agent, and may provide a waiting time for the connection to the human agent. For example, the IVR server (110) can synthesize a welcome message from the virtual agent (e.g., "Hello? This is Avatar Agent OOO. May I help you instead?") and send it to the customer. If the customer agrees to the routing with the virtual agent, the IVR server can synthesize a follow-up message from the virtual agent (e.g., "Please tell me what task you would like") and send it to the customer. Subsequently, regarding the customer's voice transmitted, the IVR server (110) can generate customer speech text, obtain a response to the customer speech text by linking with a conversation processor (130), synthesize the response into speech, and send it to the customer.
[0054] A human or virtual agent may be customized based on the characteristics of the task requested by the customer (e.g., type of task, difficulty of task, etc.) or customer characteristics. For example, virtual agents may be preferentially assigned for basic tasks such as balance inquiries, transfer limit inquiries, account transfers, account opening, and product information, or for tasks with a history of processing, while human agents may be preferentially assigned for handling customer complaints or new tasks with no history of processing. For example, human agents may be preferentially assigned to customers who frequently file complaints or who prefer human agents.
[0055] According to one embodiment, the IVR server (110) can determine the type of counselor by analyzing the business characteristics or customer characteristics requested by the customer based on customer utterance text and customer information, and can assign a virtual counselor from among a plurality of virtual counselors. The IVR server (110) can connect the call to the switchboard or transmit a response request to the conversation processor (130) depending on the type of counselor. According to another embodiment, the IVR server (110) can request counselor assignment by transmitting the customer utterance text to the counselor connection manager (120), and the counselor connection manager (120) can determine the type of counselor by analyzing the business characteristics or customer characteristics requested by the customer based on customer utterance text and customer information, and can assign a virtual counselor. That is, since the counselor connection manager (120) is implemented to manage the status of human counselors and assign counselors sequentially according to the order of calls registered in the counselor request queue, the entire counselor resource can be managed by having the counselor connection manager (120) also assign virtual counselors.
[0056] There may be various methods for assigning virtual agents. An IVR server (110) or an agent connection manager (120) may randomly assign a virtual agent from among multiple virtual agents, assign one according to priority, or assign a virtual agent whose voice model is idle. The priority of a virtual agent may be determined through evaluations such as customer satisfaction with the virtual agent, or may be determined based on evaluations of human agents mapped to the virtual agent. An IVR server (110) or an agent connection manager (120) may assign a virtual agent based on the relationship history between the human / virtual agent and the customer. For example, if there is a human agent who has previously conversed with a customer, the virtual agent mapped to that human agent may be assigned preferentially. Similarly, if there is a virtual agent who has previously conversed with a customer, that virtual agent may be assigned preferentially. Even if there are human / virtual agents who have previously conversed with a customer, the virtual agent may be assigned by excluding the history with human / virtual agents whose evaluation is below the standard. If there are multiple human or virtual agents who have previously spoken with the customer, a virtual agent may be assigned according to a predetermined priority based on designated criteria. The optimal virtual agent for the customer can be determined through the analysis of customer preferences, requests, and relationship history.
[0057] Referring to FIG. 4, the IVR server (110) may include a STT (Speech To Text) module (111) that recognizes speech and converts it into text, and a TTS (Text To Speech) module (112) that converts text into speech. The STT module (111) can convert the recognized customer speech into text and output the customer speech text. The TTS module (112) can synthesize the response received from the dialogue processor (130) into the voice features of a virtual agent and output a synthesized voice, and can generate the synthesized voice through a voice model of a virtual agent.
[0058] The TTS module (112) can generate synthetic voice using the voice model of the assigned virtual agent among the voice models for each virtual agent. For example, the IVR server (110) can generate synthetic voice corresponding to the response of inbound call #1 using the voice model of virtual agent #1 (VAgent #1) assigned to inbound call #1, and generate synthetic voice corresponding to the response of inbound call #2 using the voice model of virtual agent #2 (VAgent #2) assigned to inbound call #2. The voice models for each virtual agent can be generated by a voice model generator (170).
[0059] The voice model generator (170) can train a basic voice model using voice data of a human counselor and, through this, generate a voice model that learns voice characteristics such as the voice, tone, pronunciation, manner of speech, and speed of the human counselor.
[0060] Referring again to FIG. 1, the conversation processor (130) receives a response request containing customer utterance text and virtual agent information from the IVR server (110), and can obtain a response to the customer request included in the customer utterance text by linking with the legacy system (20), the language model-based artificial intelligence model (140), the interaction database (150) that stores agent interaction data, and the source database (160) that stores source data. The artificial intelligence model (140) can generate and output a response corresponding to the customer request in conversational natural language based on the input data received from the conversation processor (130), and can dynamically generate a response according to the conversation context, but can generate a response that does not go beyond a defined category by referencing existing interaction data.
[0061] The dialogue processor (130) may be a framework server that interacts with a language model-based artificial intelligence model (140). The dialogue processor (130) can retrieve reference data that supports the generation of responses by the artificial intelligence model (140) from the interaction database (150) and the source database (160) through a Retrieval-Augmented Generation (RAG) function. Here, the reference data is configured in a data form that the artificial intelligence model (140) can understand; for example, the reference data may be converted into vector data and input into the artificial intelligence model (140), or the interaction database (150) or the source database (160) may be constructed as a vector database.
[0062] A conversation processor (130) that receives a response request can analyze the customer utterance text and determine the necessity of linking with a legacy system (20). The legacy system (20) to be linked may vary depending on the services provided by the customer service center system (100). Meanwhile, based on the results of analyzing the customer utterance text, if the conversation processor (130) determines that the customer request is unclear, that the customer request exceeds the scope of work of the virtual agent, or that source data or interaction data related to the customer request cannot be retrieved, it may request the IVR server (110) to switch to a human agent.
[0063] A conversation processor (130) that is not linked with the period system (20) can operate as follows.
[0064] The conversation processor (130) analyzes the customer utterance text (e.g., 'Please tell me the interest rate and preferential conditions of the Rainbow deposit product') and determines that the customer request is a simple request for which a response can be generated using source data (e.g., Rainbow Smart Fixed Deposit product information) stored in the source database (160). Then, the conversation processor (130) extracts information related to the customer utterance text from the source database (160) as reference data and, in conjunction with the artificial intelligence model (140), obtains a response to the customer request from the reference data.
[0065] The conversation processor (130) can identify a human counselor mapped to a virtual counselor in the interaction database (150) based on virtual counselor information, and can extract the interaction data of the identified human counselor as reference data. Here, the conversation processor (130) can extract the interaction data containing information related to customer utterance text from the interaction data of the human counselor as reference data, or can extract interaction data of a certain length that can refer to the tone of voice and conversation patterns of the human counselor as reference data.
[0066] The dialogue processor (130) can be implemented so that the artificial intelligence model (140) generates a response to a customer utterance text by referencing reference data (information related to the customer utterance text, interaction data). For example, the dialogue processor (130) can input a command (referred to as a prompt) to the artificial intelligence model (140) to generate a response to the customer utterance text in the tone and conversation pattern of a designated counselor by referencing the reference data (information related to the customer utterance text, interaction data).
[0067] The artificial intelligence model (140) can generate a response to customer utterance text according to a command, in the tone and conversational pattern of a designated human counselor. The conversation processor (130) can transmit the response output from the artificial intelligence model (140) to the IVR server (110).
[0068] A conversation processor (130) that interacts with a period system (20) can operate as follows.
[0069] The conversation processor (130) analyzes customer utterance text (e.g., 'I want to transfer 20,000 won from account OOOO to account OOOO') and determines that the customer request is an integration request that requires integration with the legacy system (20). The conversation processor (130) includes functions (features) defined for integration with the legacy system (20) and can obtain a processing result from the legacy system (20) by calling a function related to the customer utterance text. The conversation processor (130) can implement call functions for integration with the legacy system (20) using, for example, a Langchain Tool. For example, the call functions may include a balance inquiry function, a transfer limit inquiry function, an account transfer function, an account opening function, etc. Each call function defines arguments required for integration with the legacy system. For example, the required arguments of the account transfer function may be the account number to be withdrawn, the account number and bank name to be deposited, and the amount to be transferred, and the account transfer function can be defined using the LangChain tool as shown in Table 1.
[0070]
[0071] The conversation processor (130) can select a calling function corresponding to the customer's utterance text, determine whether there are arguments for the function to be called, and if there are missing arguments or arguments that need to be verified, generate a sentence asking the customer for the necessary arguments through the artificial intelligence model (140) and transmit it to the IVR server (110). Alternatively, the conversation processor (130) can input a command requesting the selection of a calling function corresponding to the customer's utterance text and verification of the arguments of the corresponding function into the artificial intelligence model (140), thereby causing the artificial intelligence model (140) to select a calling function corresponding to the customer's utterance text, determine whether there are arguments for the function to be called, and generate a sentence asking the customer for the necessary arguments. In this case, the artificial intelligence model (140) can analyze the customer's utterance text and the calling functions based on the command, determine the function to be called (e.g., account transfer function) to generate a response, and send it back to the conversation processor (130). At this time, the artificial intelligence model (140) can determine whether there are arguments required for a function call from the customer's utterance text, generate a sentence in natural language asking the customer for missing arguments (e.g., bank name) or arguments that need to be verified (e.g., account number, transfer amount, etc.), and reply to the conversation processor (130).
[0072] The dialogue processor (130) transmits the sentence output from the artificial intelligence model (140) to the IVR server (110) and can obtain the arguments required for a function call from the customer utterance text transmitted from the IVR server (110). The dialogue processor (130) and the artificial intelligence model (140) can repeat the process of generating a sentence requesting information from the customer and analyzing the customer utterance text received in response to it until all the arguments required for a function call are obtained.
[0073] When the conversation processor (130) obtains all the arguments required for the function call, it calls the function to interact with the period system (20), and thereby obtains a processing result (e.g., transfer success) related to the customer utterance text from the period system (20).
[0074] The conversation processor (130) can input a command to the artificial intelligence model (140) to generate a response to the customer utterance text using the processing result of the period system (20).
[0075] The conversation processor (130) can identify a human counselor mapped to a virtual counselor in the interaction database (150) based on virtual counselor information, and extract the interaction data of the identified human counselor as reference data. The conversation processor (130) can input a command to the artificial intelligence model (140) to generate a response to a customer utterance text using the tone and conversation pattern of the designated human counselor by referring to the interaction data.
[0076] The artificial intelligence model (140) can generate a response to customer utterance text based on a command, in the tone and conversational pattern of a designated human counselor. The conversation processor (130) can transmit the response output from the artificial intelligence model (140) to the IVR server (110).
[0077] Meanwhile, the conversation processor (130) may include human counselor interaction data as reference data for each command input to the artificial intelligence model (140), but may include human counselor interaction data as reference data only for the first command input to the artificial intelligence model (140). That is, the artificial intelligence model (140) can generate the current response sentence using previously provided reference data while the conversation with the customer is in progress.
[0078] In this way, the conversation processor (130) can interact with the artificial intelligence model (140) to generate natural language sentences to respond to customer utterances according to the conversation context, natural language sentences to inquire about information necessary to process customer requests, and repeat the process of transmitting the generated sentences to the IVR server (110) that performs speech synthesis, thereby enabling the conversation with the customer to take the place of a human counselor. In particular, the conversation processor (130) can select the interaction data of a human counselor and have the artificial intelligence model (140) refer to it, thereby enabling the artificial intelligence model (140) to generate response sentences by mimicking the tone and conversation patterns of a skilled human counselor.
[0079] FIGS. 5 and FIGS. 6 are drawings illustrating, respectively, the generation of a response to a customer utterance text according to one embodiment.
[0080] Referring to FIG. 5, the conversation handler (130) receives the customer utterance text "Please tell me the interest rate and preferential conditions of the Rainbow Deposit product" from the IVR server (110), and a virtual agent named Agent #1 is assigned to the customer's inbound call. It is assumed that the interaction data of Agent #1 is stored in the interaction database (150). Agent #1 may be an identifier for the virtual agent, or an identifier for a human agent mapped to the virtual agent, and may be used as information to extract the interaction data.
[0081] The conversation processor (130) analyzes the customer's utterance text and can determine that the customer's request is a simple request that does not require integration with the legacy system. The conversation processor (130) can extract information related to the customer's utterance text, specifically 'Rainbow Smart Fixed Deposit', as reference data from the source database (160). Additionally, the conversation processor (130) can extract Agent #1's interaction data as reference data from the interaction database (150). The conversation processor (130) can extract interaction data containing a conversation related to 'Rainbow Smart Fixed Deposit' from Agent #1's interaction data as reference data.
[0082] The conversation processor (130) can input a command statement (e.g., 'Please tell me the interest rate and preferential conditions of the Rainbow Deposit product. Please generate a detailed answer regarding the interest rate and preferential conditions of the Rainbow Smart Deposit in accordance with Agent #1's tone and conversation pattern by referring to information about the Rainbow Smart Deposit and the interaction data of Agent #1') into the artificial intelligence model (140) to generate a response to a customer utterance text using reference data.
[0083] The artificial intelligence model (140) can generate a response to a customer's utterance text according to a command, using the tone and conversational pattern of a designated human counselor. For example, the artificial intelligence model (140) can generate a natural language sentence such as, "Dear customer, the basic interest rate for the Rainbow Smart Fixed Deposit is 3.5% per annum, and it is applied at 3.8% per annum for a subscription period of 3 years or more. If you sign up via internet / mobile, an additional preferential interest rate of 0.1% is provided. Do you have any further questions?"
[0084] The conversation processor (130) can transmit the response output from the artificial intelligence model (140) to the IVR server (110). The IVR server (110) can synthesize the received response into the voice features of a virtual counselor and output a synthesized voice.
[0085] Referring to FIG. 6, the conversation handler (130) receives the customer utterance text "I would like to transfer 20,000 won from account OOOO to account OOOO" from the IVR server (110), and a virtual agent named Agent #2 is assigned to the customer's inbound call. It is assumed that the interaction data of Agent #2 is stored in the interaction database (150). Agent #2 may be an identifier for the virtual agent, or an identifier for a human agent mapped to the virtual agent, and may be used as information to extract interaction data.
[0086] The conversation processor (130) analyzes the customer's utterance text and can determine that the customer's request is a request for integration that requires integration with the legacy system (20).
[0087] The conversation processor (130) can input into the artificial intelligence model (140) a command requesting the selection of a call function corresponding to a customer's utterance text and the verification of arguments of the corresponding function (e.g., 'The user said, "I want to transfer 20,000 won from account OOOO to account OOOO." Please select a call function from the attached call functions, determine whether there are missing arguments or arguments that need to be verified for the call of the selected function, and generate a sentence asking for the missing arguments or arguments that need to be verified in accordance with Agent #2's tone and conversation pattern. Please refer to Agent #2's interaction data for Agent #2's tone and conversation pattern.')
[0088] The artificial intelligence model (140) can generate a response to a customer's utterance text according to a command, in the tone and conversational pattern of a designated human counselor. For example, the artificial intelligence model (140) can generate a natural language sentence such as, "Customer, please repeat the account number to withdraw from, the account number to deposit, the bank name, and the amount to be transferred for the account transfer." The conversation processor (130) can wait for a function call until the account number to withdraw from, the account number to deposit, the bank name, and the amount to be transferred are obtained.
[0089] The conversation processor (130) transmits the response output from the artificial intelligence model (140) to the IVR server (110) and can receive customer speech data, 'Please transfer 20,000 won from the OOOO account to the OO Bank OOOO account' from the IVR server (110).
[0090] When the arguments required for a function call are obtained through the customer utterance text, the conversation processor (130) calls the account transfer function to interact with the period system (20), and thereby obtains a processing result (e.g., transfer success) related to the customer utterance text from the period system (20).
[0091] The conversation processor (130) can input a command to the artificial intelligence model (140) to generate a response to the customer's utterance text using the processing result of the legacy system (20) (e.g., 'Please generate the processing result to match Agent #2's tone and conversation pattern. Processing result: success').
[0092] The artificial intelligence model (140) can generate a response to a customer's utterance text according to a command, using the tone and conversational pattern of a designated human counselor. For example, the artificial intelligence model (140) can generate a natural language sentence such as, "Customer, we have completed the transfer of 20,000 won from account OOOO to account OOOO. Do you have any further questions?"
[0093] The conversation processor (130) can transmit the response output from the artificial intelligence model (140) to the IVR server (110). The IVR server (110) can synthesize the received response into the voice features of a virtual counselor and output a synthesized voice.
[0094] FIGS. 5 and 6 illustrate at least part of a conversation with a customer, and the conversation processor (130) can repeat the process of generating a contextually appropriate response to a customer utterance in conjunction with an artificial intelligence model (140) until the session for an inbound call ends.
[0095] FIG. 7 is a method of operation of a customer consultation center system according to one embodiment.
[0096] Referring to FIG. 7, the IVR server (110) converts customer speech voice received through an inbound call into customer speech data (S110). The IVR server (110) can convert the customer speech voice into text through an STT module (111) and output customer speech text.
[0097] The IVR server (110) transmits a response request containing customer utterance text to a conversation handler (130) (S120). The response request may further include virtual agent information. Response to inbound calls may be performed exclusively by a virtual agent, or a choice may be made between a human agent and a virtual agent. The timing and method of determining which agent to respond to the customer between a human agent and a virtual agent may vary; a human agent or a virtual agent may be determined by the customer's choice, or a human agent or a virtual agent may be customized according to the job characteristics requested by the customer (e.g., type of job, difficulty of job, etc.) or customer characteristics. If the customer selects a human agent or a virtual agent, the IVR server (110) may provide a waiting time for connection with a human agent. It may be implemented so that a single virtual agent (a virtual agent with the same voice, tone of voice, conversation patterns, etc.) handles all customer responses, or so that virtual agents with various voice characteristics, various conversation patterns, and various work processing histories handle customer responses. Various virtual agents can be implemented by devices constituting the customer service center system (100) generating a response by referencing the interaction data of human agents and generating a synthetic voice for the response through a voice model trained based on the voice data of human agents. The method of assigning a virtual agent among multiple virtual agents can be varied, such as random assignment, priority assignment, assignment based on the idle state of the voice model, or assignment based on the relationship history between the human / virtual agent and the customer. The assignment of virtual agents can be handled by an IVR server (110) or by an agent connection manager (120).
[0098] The conversation processor (130) extracts interaction data from the interaction database (150) in which a human counselor converses with a random customer (S130).
[0099] The conversation processor (130) analyzes the customer utterance text to determine the type of customer request (S140). The types of customer requests can vary; for example, the conversation processor (130) can determine whether the customer request is a simple request that can generate a response using source data stored in the source database (160), or whether the customer request requires integration with the legacy system (20). If, as a result of analyzing the customer utterance text, the conversation processor (130) determines that the customer request is unclear, that the customer request exceeds the scope of work of the virtual agent, or that source data or interaction data related to the customer request cannot be retrieved, it may request the IVR server (110) to switch to a human agent.
[0100] If the customer request is a simple request, the conversation handler (130) extracts information related to the customer utterance text from the source database (160) (S150).
[0101] The conversation processor (130) obtains a response generated for a customer utterance text by referencing information related to interaction data and customer utterance text through linkage with the artificial intelligence model (140) (S152). For example, the conversation processor (130) can obtain a natural language response by referencing information related to the customer utterance text and interaction data to generate a command statement to generate a response to the customer utterance text in the tone and conversation pattern of a designated counselor, and inputting this into the artificial intelligence model (140).
[0102] Meanwhile, if the customer request is a request that requires linkage with the legacy system (20), the dialogue processor (130) obtains arguments for a specific function to process the customer request from the initial customer utterance data or subsequent utterance data through linkage with the artificial intelligence model (140) (S160). The dialogue processor (130) and the artificial intelligence model (140) can repeat the process of generating a request sentence to the customer and analyzing the customer utterance text until all arguments required for the function call are obtained. For example, if the customer request analyzed from the customer utterance data is an account transfer, the dialogue processor (130) can repeat the process of generating a request sentence to the customer and analyzing the customer utterance data for it until the arguments of the account transfer function—namely, the account number to be withdrawn, the account number to be deposited, the bank name, and the transfer amount—are obtained.
[0103] The dialogue processor (130) obtains a processing result from the period system (20) through a function call using the obtained arguments (S162).
[0104] The conversation processor (130) obtains a response generated for a customer utterance text by referencing interaction data and processing results from the legacy system (20) through linkage with the artificial intelligence model (140) (S164). For example, the conversation processor (130) can obtain a natural language response by referencing processing results and interaction data related to the customer utterance text to generate a command statement to generate a response to the customer utterance text in the tone and conversation pattern of a designated counselor, and inputting this into the artificial intelligence model (140).
[0105] The conversation handler (130) sends the response to the IVR server (110) (S170).
[0106] The IVR server (110) generates a synthesized voice for the response and transmits it to the customer terminal (S180). The IVR server (110) can generate the synthesized voice through the TTS module (112). The TTS module (112) can generate the synthesized voice using the voice model of the assigned virtual agent among the voice models for each virtual agent.
[0107] As such, according to the embodiment, by dynamically generating a response that follows the tone and conversation patterns of a human counselor by referencing the interaction data of a human counselor, the consistency and accuracy of the response can be guaranteed, natural language sentences suitable for counseling work and natural can be generated, and synthetic speech suitable for counseling work can be generated through a voice model of a human counselor.
[0108] According to an embodiment, by having a virtual counselor respond to customers instead of a human counselor, customers can be connected with a counselor without waiting time and receive consistent and accurate counseling services.
[0109] According to the embodiment, the customer service center system can increase work processing efficiency through virtual agents and alleviate customer dissatisfaction caused by waiting times, thereby increasing customer satisfaction while reducing operating costs such as labor and training costs, and can provide high-quality consultation services by focusing human agents on requests that are difficult to handle or personalized requests.
[0110] According to the embodiment, by having virtual counselors with various voice characteristics, various conversation patterns, and various work processing histories respond to customers, diversity in counseling services can be provided.
[0111] The embodiments of the present disclosure described above are not implemented only through devices and methods, but may also be implemented through a program that realizes a function corresponding to the configuration of the embodiments of the present disclosure or a recording medium on which such program is recorded.
[0112] Although embodiments of the present disclosure have been described in detail above, the scope of the present disclosure is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concepts of the present disclosure as defined in the following claims also fall within the scope of the present disclosure.
Claims
Claim 1 A customer service center system comprising: a dialogue processor that generates a response from a virtual agent based on customer utterance text; and an automatic response server that converts customer utterance voice transmitted from a customer terminal into text to generate said customer utterance text, transmits a response request including said customer utterance text to said dialogue processor, generates a synthesized voice for said response received from said dialogue processor using a voice model of said virtual agent, and transmits said synthesized voice to said customer terminal; wherein the dialogue processor extracts interaction data in which a human agent converses with an arbitrary customer, and causes an artificial intelligence model to generate said response for said customer utterance text by referencing said interaction data. Claim 2 A customer service center system according to claim 1, wherein the automatic response server determines the virtual agent to respond to the customer among a plurality of virtual agents, and transmits the response request including the customer utterance text and virtual agent information to the conversation processor. Claim 3 In paragraph 2, the conversation processor extracts interaction data of the human counselor mapped to the virtual counselor through the virtual counselor information, in a customer service center system. Claim 4 A customer service center system according to claim 1, wherein the automatic response server determines whether the virtual agent responds to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer. Claim 5 In paragraph 4, the customer service center system, wherein the automatic response server connects an inbound call received from a switchboard and a customer's terminal when it is determined that a human agent will respond to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer. Claim 6 A customer service center system according to claim 1, wherein the automatic response server includes a voice model for each virtual agent, generates the synthesized voice using the voice model of the virtual agent among the voice models for each virtual agent, and the voice model for each virtual agent is generated based on voice data of a human agent mapped to the corresponding virtual agent. Claim 7 A customer consultation center system according to claim 1, wherein the conversation processor extracts information related to the customer utterance text from a source database storing source data necessary for consultation, and the artificial intelligence model generates the response to the customer utterance text by referring to the interaction data and the information related to the customer utterance text. Claim 8 A customer service center system according to claim 1, wherein the dialogue processor calls a specific function for processing a customer request analyzed from the customer utterance text among functions defined for interoperability with a legacy system, obtains a processing result from the legacy system, and causes the artificial intelligence model to generate the response to the customer utterance text by referring to the interaction data and the processing result. Claim 9 In claim 8, the conversation processor obtains the arguments of the specific function from the customer utterance data or subsequent utterance data of the customer utterance data, and then calls the specific function, a customer consultation center system. Claim 10 A customer service center system according to claim 9, wherein the subsequent utterance data is obtained by inquiring of at least one argument required for calling the specific function from the customer, the sentence inquiring of the at least one argument from the customer is generated by the artificial intelligence model, and the synthesized speech for the sentence is generated through the automatic response server and transmitted to the customer's terminal. Claim 11 A method of operation for an automatic response server comprising: a step of converting a customer utterance voice transmitted from a customer terminal into text to generate a customer utterance text; a step of transmitting a response request including said customer utterance text to a dialogue processor; a step of receiving a response to said customer utterance text from said dialogue processor; a step of generating a synthesized voice for said response using a voice model of a virtual agent; and a step of transmitting said synthesized voice to said customer terminal. Claim 12 In claim 11, the step of transmitting the response request to a dialogue processor comprises determining the virtual agent to respond to the customer among a plurality of virtual agents, and transmitting the response request including the customer utterance text and virtual agent information to the dialogue processor. Claim 13 In paragraph 12, the step of transmitting the response request to a conversation handler determines whether the virtual agent responds to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer. Claim 14 A method of operation according to claim 12, further comprising the step of connecting an inbound call received from a switchboard and a terminal of the customer when it is determined that a human agent will respond to the customer based on the customer's selection, the business characteristics requested by the customer, or the characteristics of the customer. Claim 15 In claim 11, the step of generating the synthetic voice is to generate the synthetic voice using the voice model of the virtual counselor among voice models for each virtual counselor, and the voice model for each virtual counselor is generated based on voice data of a human counselor mapped to the virtual counselor. Claim 16 A method of operation for a conversation processor comprising: receiving customer utterance text from an automatic response server; extracting interaction data from an interaction database in which a human counselor converses with a random customer; analyzing the customer utterance text to determine the type of customer request; if the customer request is a first type of request, extracting information related to the customer utterance text from a source database storing source data necessary for counseling; obtaining a first response generated for the customer utterance text by referencing the interaction data and the information related to the customer utterance text through linkage with an artificial intelligence model; and transmitting the first response to the automatic response server. Claim 17 In claim 16, if the customer request is a second type of request requiring integration with a legacy system, the method of operation further comprises the steps of: calling a specific function for processing the customer request among functions defined for integration with the legacy system; obtaining a processing result from the legacy system; obtaining a second response generated for the customer utterance text by referring to the interaction data and the processing result through integration with the artificial intelligence model; and transmitting the second response to the automatic response server. Claim 18 In claim 17, the step of calling the specific function is a method of operation in which arguments of the specific function are obtained from the customer utterance data or subsequent utterance data of the customer utterance data, and then the specific function is called. Claim 19 In claim 18, the method of operation in which the subsequent utterance data is obtained by inquiring of the customer for at least one argument required for calling the specific function through linkage with an artificial intelligence model. Claim 20 In paragraph 16, the above-mentioned first response is output as a synthesized voice using a voice model of a virtual counselor, a method of operation.