System and method for creating dialog scripts of ai counselor

Abstract

A system and method for creating a conversation script for an artificial intelligence agent are disclosed. The method according to the present disclosure comprises the steps of creating or modifying a conversation script for an AI agent based on a Voice User Experience (VUX) guideline that includes rules for a plurality of conversation components and a plurality of conversation patterns, and the step of the AI ​​agent conducting a conversation with a customer based on the conversation script, wherein each conversation component is a minimum unit of conversation and each conversation pattern may be a combination of one or more conversation components according to the conversation context.

Description

Technology Field

[0001] This document relates to a system and device for writing conversation scripts for a voice-based AI (Artificial Intelligence) agent, and more specifically, to a system and method for systematically writing and verifying conversation scripts for a voice-based AI agent. Background Technology

[0002] With the recent advancement of AI technology, companies are gradually incorporating it into their existing services to enhance the efficiency and quality of customer service. In particular, customer centers utilizing voice-based artificial intelligence are garnering attention as an important means to improve customer satisfaction and reduce operating costs.

[0003] Customer centers utilizing AI are referred to as AI Customer Centers, AI Call Centers, AI Contact Centers, or AI-based Customer Centers, and are collectively called AICCs.

[0004] The AICC method is divided into a natural language-based voice AI system in which questions and answers are not standardized, and an IVR (Interactive Voice Response) system in which questions and answers for fixed questions are pre-built in the form of a callflow as shown in Fig. 1. Fig. 1 is a diagram showing an example of an IVR system case.

[0005] AICC, currently applied in the financial and telecommunications sectors, follows a predefined conversational flow structure for question and answer processing, but it employs a hybrid approach based on Natural Language Processing (NLP) that allows for a certain degree of flexibility in mapping customer questions and answers into the flow.

[0006] Therefore, in existing AICCs, conversation design for improving customer experience depends on how smoothly the conversation flow and the conversation script (a set of sentences that the system reads in one turn of utterance) are designed.

[0007] However, while the introduction of AICC has reduced the need for agent connections and improved operational efficiency, new customer complaints regarding AICC are emerging.

[0008] Furthermore, while guidelines exist for writing dialogue scripts for AI utterance in voice-based AICC systems, they are limited to sentences specifying the standards to be followed, making it difficult to apply them consistently. The problem to be solved

[0009] The present invention is proposed to solve the aforementioned problems and various related issues. The objective of the present invention is to provide a system and method for establishing voice customer experience guidelines specialized for voice-based AI agents to improve customer experience, and for optimizing conversation scripts included in the conversation flow of existing AICCs based on these guidelines.

[0010] Another objective of the present disclosure is to provide a system and method for establishing a system of conversation scripts for AI counselors and training a generative AI with it to streamline the process of writing and verifying conversation scripts.

[0011] However, the scope of rights of the embodiments is not limited to the technical problems described above, and may be extended to other technical problems that can be inferred by a person skilled in the art based on the entire content of this document. means of solving the problem

[0012] To achieve the aforementioned purpose and other advantages, a method for writing a conversation script for an AI agent comprises the steps of writing or modifying the conversation script of the AI ​​agent based on a Voice User Experience (VUX) guideline that includes rules for multiple conversation components and multiple conversation patterns, and the step of the AI ​​agent conducting a conversation with a customer based on the conversation script, wherein each conversation component is the minimum unit of conversation and each conversation pattern is a combination of one or more conversation components according to the conversation context.

[0013] One embodiment includes the plurality of dialogue components comprising one or more of a greeting dialogue component, an utterance example dialogue component, an understanding indication dialogue component, a key information dialogue component, an explanation dialogue component, a request dialogue component, an implicit acknowledgment dialogue component, a status notification dialogue component, a suggestion dialogue component, an apology dialogue component, an additional inquiry confirmation dialogue component, or a closing dialogue component.

[0014] One embodiment includes the following: the above plurality of conversation patterns include one or more of an information transmission conversation pattern, a processing result transmission conversation pattern, an additional instruction confirmation conversation pattern, an additional information request conversation pattern, an instruction execution failure conversation pattern, a no input conversation pattern, a no matching result conversation pattern, a welcome greeting conversation pattern, or a farewell greeting conversation pattern.

[0015] As an example, the above information transmission conversation pattern is composed of the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above key information conversation component, the above explanation conversation component, and the above additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the above key information conversation component and the above additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the above understanding indication conversation component, the above implicit acknowledgment conversation component, and the above explanation conversation component are optional.

[0016] As an example, the above processing result delivery conversation pattern is composed in the order of the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above status notification conversation component, the above explanation conversation component, and the above additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the above status notification conversation component and the above additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the above understanding indication conversation component, the above implicit acknowledgment conversation component, and the above explanation conversation component are optional.

[0017] As an example, the above additional instruction confirmation conversation pattern is composed of the above understanding display conversation component, the above status notification conversation component, the above explanation conversation component, and the above additional inquiry confirmation conversation component, wherein the above understanding display conversation component and the above additional inquiry confirmation conversation component are mandatory, and the above status notification conversation component and the above explanation conversation component are optional.

[0018] As an example, the above additional information request conversation pattern is composed of the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above explanation conversation component, the above request conversation component, and the above utterance example conversation component, wherein the above request conversation component is mandatory, and the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above explanation conversation component, and the above utterance example conversation component are optional.

[0019] The above instruction execution failure conversation pattern is composed of the above explanation conversation component, the above status notification conversation component / key information conversation component, the above explanation conversation component, and the above additional inquiry confirmation conversation component / suggestion conversation component in that order, and the above status notification conversation component / key information conversation component and the above additional inquiry confirmation conversation component / suggestion conversation component are mandatory, and the above explanation conversation component is optional as an example.

[0020] As an example, the above no-input conversation pattern is composed of the above-mentioned status notification conversation component and the above-mentioned utterance example conversation component, wherein the above-mentioned status notification conversation component is mandatory and the above-mentioned utterance example conversation component is optional.

[0021] The above-mentioned no-match-result conversation pattern is divided into a first no-match-result conversation pattern and a second no-match-result conversation pattern, the first no-match-result conversation pattern is a conversation pattern corresponding to incomprehensibility or incorrect input, and the second no-match-result conversation pattern is a conversation pattern of partial understanding or two or more intents, and the first no-match-result conversation pattern is composed in the order of the status notification conversation component, the previous script repetition, and the utterance example conversation component, wherein the status notification conversation component is mandatory and the utterance example conversation component is optional, as an example.

[0022] The above welcome conversation pattern is composed of the above greeting conversation component and the above utterance example conversation component in that order, and as an example, the above greeting conversation component and the above utterance example conversation component are mandatory.

[0023] As an example, the above farewell conversation pattern is composed of the above understanding indication conversation component and the above closing conversation component in that the above closing conversation component is mandatory and the above understanding indication conversation component is optional.

[0024] One embodiment comprises the step of modifying the above dialogue script by analyzing the above dialogue script to determine a dialogue pattern; when the dialogue pattern is determined, checking whether the above dialogue script is suitable for the configuration of dialogue components of the determined dialogue pattern; when it is confirmed that the above dialogue script is not suitable for the configuration of dialogue components of the dialogue pattern, checking whether there are unnecessary or missing dialogue components in the above dialogue script; based on the result of the check, adding necessary dialogue components to the above dialogue script or deleting unnecessary dialogue components and corresponding phrases from the above dialogue script; and after the above steps are performed, modifying the above dialogue script according to the dialogue pattern and the rules for configuring dialogue components of the above dialogue pattern.

[0025] As an example, the creation, modification, or verification of the above conversation script is performed through an AI-based agent.

[0026] A system for creating a conversation script for an AI agent according to the present invention comprises a memory storing a Voice Customer Experience (VUX) guideline including rules for a plurality of conversation components and a plurality of conversation patterns, and a processor that creates or modifies a conversation script for an AI agent based on the VUX guideline and controls the AI ​​agent to proceed with a conversation with a customer based on the conversation script, wherein each conversation component is a minimum unit of conversation and each conversation pattern is a combination of one or more conversation components according to a conversation context.

[0027] One embodiment includes the plurality of dialogue components comprising one or more of a greeting dialogue component, an utterance example dialogue component, an understanding indication dialogue component, a key information dialogue component, an explanation dialogue component, a request dialogue component, an implicit acknowledgment dialogue component, a status notification dialogue component, a suggestion dialogue component, an apology dialogue component, an additional inquiry confirmation dialogue component, or a closing dialogue component.

[0028] One embodiment includes the following: the above plurality of conversation patterns include one or more of an information transmission conversation pattern, a processing result transmission conversation pattern, an additional instruction confirmation conversation pattern, an additional information request conversation pattern, an instruction execution failure conversation pattern, a no input conversation pattern, a no matching result conversation pattern, a welcome greeting conversation pattern, or a farewell greeting conversation pattern.

[0029] As an example, the above information transmission conversation pattern is composed of the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above key information conversation component, the above explanation conversation component, and the above additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the above key information conversation component and the above additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the above understanding indication conversation component, the above implicit acknowledgment conversation component, and the above explanation conversation component are optional.

[0030] As an example, the above processing result delivery conversation pattern is composed in the order of the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above status notification conversation component, the above explanation conversation component, and the above additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the above status notification conversation component and the above additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the above understanding indication conversation component, the above implicit acknowledgment conversation component, and the above explanation conversation component are optional.

[0031] As an example, the above additional instruction confirmation conversation pattern is composed of the above understanding display conversation component, the above status notification conversation component, the above explanation conversation component, and the above additional inquiry confirmation conversation component, wherein the above understanding display conversation component and the above additional inquiry confirmation conversation component are mandatory, and the above status notification conversation component and the above explanation conversation component are optional.

[0032] As an example, the above additional information request conversation pattern is composed of the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above explanation conversation component, the above request conversation component, and the above utterance example conversation component, wherein the above request conversation component is mandatory, and the above understanding indication conversation component, the above implicit acknowledgment conversation component, the above explanation conversation component, and the above utterance example conversation component are optional.

[0033] The above instruction execution failure conversation pattern is composed of the above explanation conversation component, the above status notification conversation component / key information conversation component, the above explanation conversation component, and the above additional inquiry confirmation conversation component / suggestion conversation component in that order, and the above status notification conversation component / key information conversation component and the above additional inquiry confirmation conversation component / suggestion conversation component are mandatory, and the above explanation conversation component is optional as an example.

[0034] As an example, the above no-input conversation pattern is composed of the above-mentioned status notification conversation component and the above-mentioned utterance example conversation component, wherein the above-mentioned status notification conversation component is mandatory and the above-mentioned utterance example conversation component is optional.

[0035] The above-mentioned no-match-result conversation pattern is divided into a first no-match-result conversation pattern and a second no-match-result conversation pattern, wherein the first no-match-result conversation pattern is a conversation pattern corresponding to incomprehensibility or incorrect input, and the second no-match-result conversation pattern is a conversation pattern of partial understanding or two or more intents, and the first no-match-result conversation pattern is composed in the order of the status notification conversation component, the previous script repetition, and the utterance example conversation component, wherein the status notification conversation component is mandatory and the utterance example conversation component is optional, and the second no-match-result conversation pattern is composed of the request conversation component as an embodiment.

[0036] The above welcome conversation pattern is composed of the above greeting conversation component and the above utterance example conversation component in that order, and as an example, the above greeting conversation component and the above utterance example conversation component are mandatory.

[0037] As an example, the above farewell conversation pattern is composed of the above understanding indication conversation component and the above closing conversation component in that the above closing conversation component is mandatory and the above understanding indication conversation component is optional.

[0038] As an example, the processor analyzes the conversation script to determine a conversation pattern, and when the conversation pattern is determined, checks whether the conversation script is suitable for the configuration of conversation components of the determined conversation pattern, and if it is confirmed that the conversation script is not suitable for the configuration of conversation components of the conversation pattern, checks whether there are unnecessary or missing conversation components in the conversation script, adds necessary conversation components to the conversation script based on the result of the check, or deletes unnecessary conversation components and corresponding phrases from the conversation script, and after the operation, modifies the conversation script according to the conversation pattern and the rules for configuring conversation components of the conversation pattern.

[0039] As an example, the creation, modification, or verification of the above conversation script is performed through an AI-based agent operated by the processor.

[0040] According to the embodiments, a computer program stored on a computer-readable recording medium is combined with a computer, which is hardware, to perform the above method.

[0041] The aforementioned means of solution are merely some of the various embodiments of this specification, and variations reflecting the technical features of this specification can be inferred by those skilled in the art from the overall context of this specification. Effects of the invention

[0042] The system and method for writing conversation scripts for an AI counselor according to the embodiments have the effect of improving business processes by proposing conversation script guidelines applicable based on actual service data and proposing a method to support the writing and verification of conversation scripts.

[0043] In addition to the technical effects explicitly mentioned above, effects that are obvious to those skilled in the art can also be inferred from the entire description in this specification. Brief explanation of the drawing

[0044] Drawings are included to further understand the embodiments, and the drawings illustrate the embodiments along with descriptions related to the embodiments. Figure 1 is a diagram showing an example of an IVR system case. FIG. 2 is a diagram showing an embodiment of a method for deriving VUX principles by considering dialogue and interface perspectives according to the present invention. FIG. 3 is a diagram showing an example of a conversation component according to the present invention. FIG. 4 is a flowchart illustrating situations that a voice-based AI counselor may face during a consultation according to the present invention. FIGS. 5(a) and FIGS. 5(b) are drawings showing an example of the configuration of conversation components and a script of the information transmission conversation pattern of the present invention. Figure 6 is a diagram visually showing the relationship between the total number of characters in a script and the dropout rate according to the embodiments. FIGS. 7(a) and FIGS. 7(b) are drawings showing an example of the configuration of dialogue components and a script of the processing result transmission dialogue pattern of the present invention. FIGS. 8(a) and FIGS. 8(b) are drawings showing an example of the configuration of dialogue components and a script of the additional instruction confirmation dialogue pattern of the present invention. FIGS. 9(a) to 9(c) are drawings showing examples of the configuration of conversation components and scripts of the additional information request conversation pattern of the present invention. Figure 10 is a diagram showing the deviation rate according to the form of the query of the embodiments. FIG. 11 is a drawing showing an example of a break-up rate according to the number of options in the present invention. FIG. 12 is a drawing showing an example of a dropout rate according to the sum of the number of characters of the options in the present invention. FIGS. 13(a) to 13(c) are drawings showing examples of the configuration of dialogue components and scripts of the instruction-unexecutable dialogue pattern of the present invention. FIGS. 14(a) to 14(c) are drawings showing examples of the configuration of dialogue components and scripts of the no-input dialogue pattern of the present invention. FIG. 15 is a diagram showing examples of the configuration of dialogue components and scripts of the no-match-result dialogue pattern of the present invention. FIGS. 16(a) and FIGS. 16(b) are drawings showing an example of the configuration of conversation components and a script of the welcome greeting conversation pattern of the present invention. FIGS. 17(a) and FIGS. 17(b) are drawings showing an example of the configuration of conversation components and a script of the farewell conversation pattern of the present invention. FIG. 18 is a flowchart illustrating one embodiment of the method for modifying a dialogue script of the present invention. FIG. 19 is a diagram showing an example of modifying a conversation script according to the conversation pattern and conversation component rules of the present invention. FIG. 20 is a diagram showing another example of modifying a conversation script according to the conversation pattern and conversation component rules of the present invention. FIG. 21 is a diagram showing another example of modifying a conversation script according to the conversation pattern and conversation component rules of the present invention. FIG. 22 is a diagram showing an example of a business process utilizing an AI-based agent of the present invention. FIG. 23 is a diagram showing an example of a conversation script created using the AI-based agent of the present invention. FIG. 24 is a diagram showing another example of a conversation script created using the AI-based agent of the present invention. FIG. 25 is a diagram showing another example of a conversation script created using the AI-based agent of the present invention. FIG. 26 is a block diagram showing an example of a system for writing a conversation script for a voice-based AI counselor according to the present invention. Specific details for implementing the invention

[0045] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components regardless of drawing symbols are assigned the same reference numeral, and redundant descriptions thereof will be omitted. It should be understood that the following embodiments are merely for the purpose of embodying the present invention and do not limit or restrict the scope of the rights of the present invention. Anything that can be easily inferred by a person skilled in the art to which the present invention pertains from the detailed description and embodiments of the present invention is interpreted as falling within the scope of the rights of the present invention.

[0046] The detailed description of the invention should not be interpreted as restrictive in all respects but should be considered exemplary. The scope of the invention should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of this document are included within the scope of the invention.

[0047] Meanwhile, the blocks of the attached block diagram or the steps of the flowchart may be implemented directly in hardware, implemented as software modules executed by hardware, or implemented by a combination thereof. Additionally, the blocks of the attached block diagram or the steps of the flowchart may be interpreted as computer program instructions that perform designated functions by being loaded into the processor or memory of data processing equipment, such as a general-purpose computer, a specialized computer, a portable notebook computer, or a network computer. Since these computer program instructions may be stored in memory provided in a computer device or in memory readable by a computer, the functions described in the blocks of the block diagram or the steps of the flowchart may be produced as manufactured products containing means of instruction to perform them. Furthermore, each block or each step may represent a module, segment, or part of code containing one or more executable instructions for executing a specific logical function(s). Additionally, in some alternative embodiments, the functions mentioned in the blocks or steps may be executed out of a predetermined order. For example, two blocks or steps that are depicted consecutively may be performed substantially simultaneously or in reverse order, and in some cases, some blocks or steps may be omitted.

[0048] The present invention proposes a system and method for optimizing the conversational scripts of a voice-based AICC by establishing voice customer experience guidelines and utilizing a customized AI language model based thereon to streamline the process of writing and verifying conversational scripts. In the present invention, voice customer experience may be referred to as VUX (Voice User eXperience), voice user experience, or customer experience through a voice interface. In this disclosure, VUX applies the existing UX concept to a voice-based interface and encompasses all elements experienced by a user (i.e., a customer) when interacting with a device through voice commands or conversations.

[0049] This invention derives VUX principles that must be observed when writing dialogue scripts, taking into account the principles of dialogue cooperation and the characteristics of voice interfaces. Furthermore, based on these principles, it defines dialogue components, which are the smallest units of dialogue, and dialogue patterns, which are combinations of dialogue components according to the dialogue context, thereby providing a foundation for writing and verifying dialogue scripts logically and procedurally.

[0050] The present invention provides an AI agent that learns VUX guidelines and various conversation script analysis cases by utilizing the functions of a customized AI model.

[0051] As such, the present invention establishes a VUX (Voice User eXperience) guideline specialized for voice-based AI agents to improve the customer experience of AICC, and proposes a method to optimize conversation scripts included in the conversation flow of existing AICC based on this. In this disclosure, the voice-based AI agent may be referred to as a callbot.

[0052] In this invention, the VUX guidelines derive core principles of dialogue design through a literature review using a top-down approach, and establish dialogue components and patterns that enable practical implementation and improvement by analyzing existing dialogue scripts using a bottom-up approach. Furthermore, this invention utilizes a customized AI language model to streamline the process of writing dialogue scripts based on the VUX guidelines and proposes a method for efficiently reviewing the written dialogue scripts.

[0053] The present invention will be described in the following order. First, VUX principles for dialogue design are established using a top-down approach, and based on the established VUX principles, dialogue scripts of voice-based AI agents are analyzed to establish dialogue components and dialogue patterns. Then, an agent is built and verified to support script writing and verification based on the VUX guidelines.

[0054] The following is an explanation of establishing VUX principles.

[0055] In order to derive the core principles of VUX design, the present invention considers an approach from a conversational perspective and an approach from an interface perspective. That is, since VUX is communication between a voice-based AI agent and a person, it is considered from a conversational perspective, and subsequently, since it is a tool for voice-based communication, it is considered from an interface perspective.

[0056] Approach from a conversational perspective

[0057] This invention analyzes the characteristics of voice conversation based on the following four conversational cooperation principles, from the perspective that the VUX of a voice-based AI counselor is ultimately a voice conversation between a person and an AI system.

[0058] First, the principle of Maxim of Quantity dictates that only the amount of information required to meet the purpose should be provided in a conversation. In other words, since excessive or insufficient information can confuse the user, only the essential information relevant to the purpose should be provided. For example, in response to the question "Where do you live?", it is generally unnecessary to provide overly broad information such as "South Korea" or detailed information such as "32 Hangang-daero, Yongsan-gu." Instead, it is appropriate to respond clearly and concisely, such as "Seoul" or "Yongsan."

[0059] Second, the Maxim of Quality principle dictates that false or unfounded statements must be avoided. In other words, AI systems must deliver reliable information to users, and if there is uncertainty, they must clearly notify the user and suggest appropriate alternatives. In particular, given that voice-based AI agents engage in consultation, it is advisable to refrain from making uncertain or speculative remarks and to provide only substantiated information.

[0060] Third, the Maxim of Relation principle dictates that remarks must be appropriate to the context and situation of the conversation. In other words, only answers directly related to the user's question should be provided, and unnecessary information should be excluded. For instance, if a customer requests to check rate plans, providing additional information or recommendations for other services can disrupt the flow of the conversation.

[0061] Fourth, the principle of Maxim of Manner dictates that one should avoid ambiguous or vague expressions in conversation and respond concisely and clearly. In other words, one should quickly convey only the core information so that the user can easily understand it, and refrain from ambiguous explanations. For example, if the response ends with "It was not processed," the user cannot determine whether the cause lies with the system or their own mistake; therefore, an explanation should be provided with clear grounds, such as "It was not processed because the information you entered did not match."

[0062] Approach from an interface perspective

[0063] Voice-based AI systems are fundamentally different from existing GUIs (Graphic User Interfaces) capable of conveying visual information. Therefore, in the design of the VUX, the characteristics of the voice interface in this invention consider the following three elements.

[0064] First, voice-based AI systems provide all functions in the form of shortcuts. In other words, instead of navigating menus, users quickly request desired information via voice and expect an immediate response. Taking payment as an example, while a GUI-based system requires about 16 touches, a voice-based AI system requires only 2 to 3 utterances.

[0065] Second, voice is an interface where everyone is an expert. In other words, because people have used voice naturally since childhood, no separate learning is required regarding how to operate it. This characteristic raises expectations for AI systems, while simultaneously making users more critical of system errors and more easily disappointed. For example, if a system malfunctions, users in a GUI-based system can reflect on what they did wrong, whereas in a VUI-based system, it is easy to assume that the system simply did not understand them.

[0066] Third, voice is volatile. Unlike GUIs, information is not visually fixed and disappears after utterance, so users must remember it. From a conversational perspective, users must rely on memory for everything they have heard, said, and will say. To use a GUI analogy, this is similar to menus appearing and disappearing, or content vanishing as soon as it loads, in a voice interface. This characteristic reduces the user's sense of control over the system and places a cognitive load on them.

[0067] The present invention derives VUX principles by considering the aforementioned dialogue and interface perspectives as one embodiment.

[0068] FIG. 2 is a diagram showing an embodiment of a method for deriving VUX principles by considering dialogue and interface perspectives according to the present invention.

[0069] As an embodiment of the present invention, four principles for designing a VUX of a voice-based AI system are derived by comprehensively considering the conversational perspective and the interface perspective.

[0070] As previously mentioned, the conversational perspective includes the principle of quantity (i.e., purpose-driven information), the principle of query (i.e., reliable speech), the principle of relation (i.e., speaking only about relevant content), and the principle of attitude (i.e., concise speech), while the interface perspective includes all functions as shortcuts (i.e., no menu structure), everyone as an expert (i.e., no need to learn), and all information as volatile (i.e., relying on memory).

[0071] Furthermore, the VUX principles derived by comprehensively considering conversational and interface perspectives include the principles of concise and clear conversation, easy and natural conversation, context-aware flexible conversation, and reliable conversation. In this context, the principle of concise and clear conversation may include detailed principles such as efficiency, conciseness, and informational audibility, while the principle of easy and natural conversation may include detailed principles such as ease of operation / learnability, consistency, and naturalness. Additionally, the context-aware flexible conversation principle may include detailed principles such as scalability, flexibility, and intent conformity, and the reliable conversation principle may include detailed principles such as error prevention, security, and integrity.

[0072] The following is a detailed description of the four VUX principles of the present invention: the principle of concise and clear conversation, the principle of easy and natural conversation, the principle of flexible conversation considering context, and the principle of reliable conversation.

[0073] The first principle is concise and clear conversation. Speech appears and disappears over time. Users must wait for the desired information to appear and remember it before it vanishes. Therefore, necessary information must be conveyed unambiguously and expressed concisely, capturing the essence. Factors such as the volume of conversation, speaking speed, and sentence structure must be considered to ensure users can hear and understand clearly. The specific principles to be observed include 'efficiency,' which enables users to use the service more quickly and efficiently; 'conciseness,' which requires conveying necessary information concisely and unambiguously, capturing only the core; and 'information audibility,' which ensures users can hear and understand clearly by considering the volume of conversation, speaking speed, and sentence structure.

[0074] The second principle is easy and natural conversation. Users have been familiar with using voice since childhood. Therefore, they must be able to easily understand how to use the features without having to learn them separately, and interact naturally, just as if conversing with a person. The specific principles to be observed include 'ease of operation / learnability,' ensuring that features are not difficult and controls are not inconvenient; 'consistency,' providing a consistent user experience across the entire service; and 'naturalness,' enabling interaction as naturally as conversing with a real person.

[0075] The third principle is a flexible conversational approach that considers context. Conversation is created through cooperation that takes the other party into account. Therefore, the system requires the ability to understand and respond to various forms of utterance, and must provide answers that align with the user's intent by understanding and anticipating the context of the conversation. Furthermore, considering situations where interaction is difficult using voice alone, it must provide auxiliary or alternative means through other modalities. The specific principles to be observed in this regard include 'scalability,' which requires the ability to understand and respond to various forms of utterance (different expressions with the same meaning, slang, ungrammatical sentences, etc.); 'intent-correcting,' which requires providing answers that align with the user's intent by understanding and anticipating the context of the conversation; and 'flexibility,' which requires the ability to assist or change to modalities other than voice depending on the situation.

[0076] The fourth principle is the principle of reliable conversation. Since voice evaporates upon utterance, making it difficult to verify the content, and is emitted in all directions, there is a risk of personal information exposure in public spaces. Therefore, it is necessary to be able to prevent mistakes in advance or take quick action when errors occur, and protective devices to prevent the exposure of personal information, along with guidance regarding such measures, are required. The specific principles to be observed include 'error prevention,' which requires preventing mistakes in advance and taking quick action when errors occur; 'security,' which requires providing protective devices and guidance to prevent the exposure of personal information; and 'integrity,' which requires that there be no falsehoods in the information and that only reliable information is provided.

[0077] Establishing Dialogue Components and Dialogue Patterns

[0078] Based on the above VUX principles, the present invention analyzes conversation scenarios of multiple voice-based AI agents currently in commercial service and the numerous conversation scripts included therein, thereby establishing a system that can be practically utilized for writing conversation scripts. Furthermore, in GUI-based design systems, the smallest units of UI, such as buttons, menus, and labels, are referred to as components, and a combination of these components in a reusable form according to function and purpose is called a pattern. As an embodiment, the present invention adopts these concepts into VUI to define and concretize the concept of conversation components, which are the basic units of conversation, and conversation patterns, which are combinations of conversation components defined according to the purpose of the conversation.

[0079] Dialogue component

[0080] In this invention, a dialogue component refers to the smallest unit constituting a dialogue script, meaning a phrase, clause, or sentence having a single purpose. This is a unit designed to respond to user requests or perform specific tasks, and each component has a unique purpose and function. On the other hand, if a specific phrase, clause, or sentence does not match a defined dialogue component, it may be considered to have an unclear purpose or be incorrectly configured.

[0081] In the present invention, the conversation components may consist of a total of 12 components. The 12 conversation components in the present invention include a Greetings conversation component, an Examples conversation component, an Acknowledgements conversation component, a Core Information conversation component, an Explanations conversation component, a Requests conversation component, an Implicit Confirmations conversation component, an Alerts conversation component, a Recommendations conversation component, an Apologies conversation component, an Any Question? conversation component, and an Endings conversation component. These conversation components are an example of an embodiment, and in the present invention, conversation components may be added, deleted, or modified, and it is obvious that such added, deleted, or modified conversation components also belong to the present invention.

[0082] FIG. 3 is a diagram illustrating an example of a conversation component according to the present invention. In FIG. 3, the sentence includes three conversation components: an acknowledgments conversation component, an alert conversation component, and an any question? conversation component. According to the embodiments, 'Yes' is an 'acknowledgments' conversation component that functions to provide feedback that the user's words have been understood; 'I sent you the invoice via text message' is a 'alert' conversation component intended to inform the user of the result after performing a task; and 'Do you have any other questions?' is an 'any question?' conversation component that checks if there are any additional instructions for the user after the task is completed. As such, the conversation component serves as the minimum unit of sentence composition and has a single purpose.

[0083] The following is a detailed description of the 12 dialogue components of the present invention.

[0084] Greetings conversation component

[0085] The greeting conversation component is intended to initiate a conversation with the user as a welcoming greeting. Sentences such as "Hello, this is the AI ​​counselor" fall into this category. It is advisable to use the utterance example conversation component in conjunction with this to help users develop appropriate expectations regarding the system's intelligence and functionality.

[0086] Utterance Examples (Examples) Dialogue Component

[0087] The utterance example conversation component serves to inform the user how to speak by utilizing representative keywords or example commands. "Please simply state your inquiry, such as checking rates" is an example of this. It is advisable to appropriately utilize the utterance example conversation component when asking open-ended questions. However, excessive use can make the conversation unnatural or make the user feel like they are losing control, so it is best to use it restrictively only when necessary.

[0088] Acknowledgements dialog component

[0089] The understanding indicator dialogue component serves to confirm that the system understands the user's speech. This instills confidence in the user and creates a natural conversational feel. Examples include "Yes" and "Understood." However, excessive use and repetition of the understanding indicator dialogue component can make the conversation awkward and unnecessarily lengthen it; therefore, it is advisable to use it appropriately as needed within the flow of the conversation.

[0090] Core Information Dialogue Component

[0091] Key information conversation components refer to the most essential and important information that directly responds to a user's query. In voice-based AI systems, users must listen to the entire audio without skipping until the desired information is reached. Therefore, placing key information at the beginning of the sentence is considered as an example. In particular, unnecessary conversation components are not placed before the key information.

[0092] Explanations dialog component

[0093] The explanatory dialogue component serves to aid user understanding by providing the basis and detailed information regarding the system's response. Depending on the needs, the explanatory dialogue component may be placed before or after key information, status notifications, request, or suggestion dialogue components, but it cannot be placed independently. As an example, the explanatory component is not excessively placed, even if the information is necessary for the user. Furthermore, if there is a large amount of information to be provided, it is preferable to deliver it through other channels, such as text messages.

[0094] Requests dialog component

[0095] Request conversation components serve the role of requesting additional information necessary to execute user instructions or seeking user confirmation regarding the performance of uncancellable or critical tasks. They effectively convey information by presenting requirements concisely and proactively so that users can easily and clearly understand what they need to do, and by appropriately including explanations as needed. For example, if a user asks about a bill, additional information can be requested, such as "Which month's bill are you curious about?", or confirmation can be requested before changing a phone number, such as "Is the contact number you wish to change to 010-0000-0000?" In this case, since excessive use of confirmation request conversation components can make the conversation unnecessarily long, it is advisable to place them only for key points.

[0096] Implicit Confirmations dialog component

[0097] Requesting confirmation has the advantage of preventing user confusion by clarifying the system's understanding. While the request conversation component performs the role of explicitly asking the user a question to obtain confirmation, the implicit confirmation conversation component performs the role of indirectly expressing that the user's instruction has been understood. For example, when a user asks how much the billing amount is for January, in the response "The amount billed on January 26 is 98,000 won," "the amount billed on January 26" corresponds to the implicit confirmation conversation component. Although one could answer "98,000 won" in accordance with the principle of brevity, since "January" is likely to be confused with "February," the present invention can prevent errors and confusion by adding an implicit confirmation conversation component.

[0098] Status Alerts dialog component

[0099] The status notification dialog component serves to inform the user of the system's progress or results regarding their instructions. Sentences such as "We have sent your invoice via email" fall into this category. Additionally, the status notification dialog component must be used to inform the user of the situation even when the system is unable to execute their instructions due to a system issue. Since the status notification dialog component, along with the key information dialog component, represents the final outcome of the user's instructions, it is advisable to place it in a top-down structure whenever possible to quickly convey the essential information to the user.

[0100] Recommendations dialog component

[0101] Suggestion conversation components proactively present additional questions the user might ask or alternative solutions that could be helpful. Based on the principle of efficiency, tasks related to the answer can be facilitated by presenting them first. For example, this applies to users inquiring about selective contract discounts; after providing guidance on the contract via a core information conversation component, tasks likely needed for the next step, such as "Would you like me to guide you through the sign-up process?" In particular, if the user's instructions could not be followed, it is advisable to place a suggestion conversation component immediately afterward to present an alternative that can achieve the objective. However, even if a task is helpful, avoid making subsequent suggestions if the user has rejected one proposal.

[0102] Apologies dialog component

[0103] The apology dialogue component is intended to seek understanding when a user's instructions cannot be performed due to unavoidable circumstances. However, apologizing does not increase favorability toward the system; on the contrary, it can lower trust. In most situations, users prioritize achieving their objectives over hearing an apology. Therefore, whenever possible, it is more effective to provide a clear explanation of the situation or suggest alternatives instead of an apology.

[0104] Any Question? Dialogue Component

[0105] The additional inquiry confirmation dialog component serves to ask the user if they have any further requests or questions after completing an instruction. For example, a sentence like "Do you have any other inquiries?" falls into this category. When an instruction is completed, it is advisable not to immediately end the conversation, but to provide the user with an opportunity to present any remaining requirements through an additional inquiry confirmation process.

[0106] Endings Dialogue Component

[0107] The closing remark conversation component serves to end the conversation as a closing greeting. For example, it presents a closing remark to conclude the conversation when it is confirmed that the user has no further requests through confirmation of additional inquiries, or when further input can no longer be received from the user.

[0108] Conversation patterns

[0109] In this invention, the conversation pattern defines which combination of conversation components should be used depending on the context and purpose of the conversation.

[0110] FIG. 4 is a flowchart illustrating situations that a voice-based AI counselor may face during a consultation according to the present invention. In FIG. 4, the welcome greeting in step S111, no input in step S116, no matching result in step S117, request for additional information in step S120, inability to perform instructions in step S122, information transmission in step S123, transmission of processing results in step S124, and confirmation of additional instructions in step S125 represent situations in which the voice-based AI counselor must make a sound, i.e., conversation patterns. The conversation patterns illustrated in FIG. 4 are examples, and in the present invention, conversation patterns can be added, deleted, or modified, and it is obvious that such added, deleted, or modified conversation patterns also belong to the present invention.

[0111] In the present invention, a conversation pattern defines which combination of conversational components should be used for each utterance situation. In this case, one embodiment of the utterance situation is a voice-based AI counselor acting as a task-oriented agent to resolve service-related inquiries. In the present invention, one embodiment of the invention is that queries not directly related to the purpose of the service, such as weather inquiries or greetings, are processed flexibly without defining a separate conversation pattern.

[0112] Next, each conversation pattern will be described in detail with reference to Fig. 4.

[0113] Information transmission conversation pattern (S123)

[0114] In the present invention, a conversation pattern defines which conversation components should be used in what order in a specific utterance situation. In the present invention, the information transmission conversation pattern is designed to concisely and clearly convey content that meets the purpose by considering VUX principles in a situation where a user requests information. The situations in which this information transmission conversation pattern is used are as follows. For example, in a situation where a user requests information, the user's utterance is successfully received (Input = YES) (S113), the content is understood (Match Result = YES) (S114), there is no further information to be verified from the user (Verification / Information Needed = NO) (S119), and a response is possible (Response Possible = YES) (S121). The present invention can use the information transmission conversation pattern in such utterance situations.

[0115] In this case, the conversation components defined in the conversation pattern are used in a fixed order, and the conversation components indicated in parentheses are optional, and whether to use them is determined appropriately depending on the situation and context.

[0116] FIGS. 5(a) and FIGS. 5(b) are drawings showing the configuration of conversation components of the information transmission conversation pattern of the present invention and an example of a script. Specifically, FIG. 5(a) shows an example of conversation components constituting the information transmission conversation pattern, and FIG. 5(b) shows an example of a conversation script written based on the information transmission conversation pattern of FIG. 5(a).

[0117] The information transmission conversation pattern of Fig. 5(a) can be composed of an understanding indication conversation component, an implicit acknowledgment conversation component, a key information conversation component, an explanation conversation component, and an additional inquiry confirmation conversation component / suggestion conversation component. In this case, the key information conversation component and the additional inquiry confirmation conversation component / suggestion conversation component are mandatory conversation components that must be included in the information transmission conversation pattern, while the understanding indication conversation component, the implicit acknowledgment conversation component, and the explanation conversation component are optional conversation components whose use is determined by the situation and context.

[0118] That is, looking at the information transmission dialogue pattern of Fig. 5(a), the core information dialogue component, which is a user requirement, is mandatory, and based on the principle of naturalness, an understanding indication dialogue component and an implicit confirmation dialogue component can be added before the core information dialogue component depending on the context. In addition, supporting evidence or helpful information can be provided together as an explanation dialogue component after the core information dialogue component. At this time, the utterance time should not be prolonged.

[0119] According to the embodiments, even when adding useful explanations to the information transmission conversation pattern, it is desirable to keep the total length of the script within 150 characters. However, this is one embodiment, and the length of the script may be longer than 150 characters.

[0120] FIG. 6 is a diagram visually illustrating the relationship between the total number of characters in a script and the dropout rate according to embodiments. More specifically, FIG. 6 visually illustrates the relationship between the total number of characters in a script and the dropout rate by analyzing hundreds of conversation histories (e.g., 378) corresponding to information transmission conversation patterns (extracting only information transmission conversations from thousands of conversation histories (e.g., 1,867) over a specific period). As a result of the analysis, it was found that the dropout rate increases as the total number of characters in the script increases, and when it exceeds 150 characters, the dropout rate exceeds 10%. Based on this data and considering business targets, the present invention sets 150 characters as the maximum number of characters for information transmission conversation patterns; however, the length of the script may be longer than the maximum of 150 characters depending on the user and / or the language used.

[0121] According to the embodiments, after providing information, if there is useful information or additional work related to the provided information, suggestions are made proactively, and if not, the conversation can be concluded with an additional inquiry confirmation dialog component to determine whether to end the conversation.

[0122] FIG. 5(b) is an example of a script assuming a situation where a user inquires about the January billing charges. As an example, the script of FIG. 5(b), which is written based on the information transmission conversation pattern of FIG. 5(a), includes an implicit confirmation conversation component (e.g., the January charges you will pay are), a key information conversation component (e.g., 98,000 won), and a suggestion conversation component (e.g., would you like me to inform you of the February billing charges as well?).

[0123] According to the embodiments, the script example of FIG. 5(b) first places the core information conversation component 'It is 98,000 won' as a mandatory element, and adds an implicit confirmation conversation component 'The January fee you will pay is' in accordance with the principle of error prevention. Then, if it is determined that the user may also be curious about the February fee, a suggestion conversation component 'Would you like me to inform you of the February fee as well?' can be added. If it is determined that the current month is January and the user does not want any further information, an additional inquiry confirmation conversation component 'Do you have any further questions?' can be placed to naturally conclude the conversation.

[0124] Processing result delivery conversation pattern (S124)

[0125] In the present invention, the processing result transmission dialogue pattern is a dialogue pattern intended to notify that a task requested by a user has been successfully completed. This processing result transmission dialogue pattern is similar to the information transmission dialogue pattern, but differs in that the user's request is a command to perform a task rather than a request for information. Therefore, while a core information dialogue component is mandatory in the information transmission dialogue pattern, a status notification dialogue component is mandatory in the processing result transmission dialogue pattern of FIG. 7(a) instead of the core information dialogue component.

[0126] FIGS. 7(a) and FIGS. 7(b) are drawings showing the configuration of dialogue components of the processing result transmission dialogue pattern of the present invention and an example of a script. Specifically, FIG. 7(a) shows an example of dialogue components constituting the processing result transmission dialogue pattern, and FIG. 7(b) shows an example of a dialogue script written based on the processing result transmission dialogue pattern of FIG. 7(a).

[0127] The situation in which the processing result transmission dialogue pattern is used in the present invention is the same as the information transmission dialogue pattern as follows. For example, in a situation where a user requests information, the user's utterance is successfully received (Input = YES) (S113), the content is understood (Match Result (Match) = YES) (S114), there is no further information to be verified from the user (Verification / Information Needed = NO) (S119), and a response is possible (Response Possible = YES) (S121). The present invention can use the processing result transmission dialogue pattern in such utterance situations.

[0128] At this time, as an example, the conversation components defined in the processing result delivery conversation pattern are used in a fixed order, and the conversation components indicated in parentheses are optional, and whether to use them is determined appropriately depending on the situation and context.

[0129] The processing result delivery conversation pattern of Fig. 7(a) can be composed of an understanding indication conversation component, an implicit acknowledgment conversation component, a status notification conversation component, an explanation conversation component, and an additional inquiry confirmation conversation component / suggestion conversation component. In this case, the status notification conversation component and the additional inquiry confirmation conversation component / suggestion conversation component are mandatory conversation components that must be included in the processing result delivery conversation pattern, while the understanding indication conversation component, the implicit acknowledgment conversation component, and the explanation conversation component are optional conversation components whose use is determined by the situation and context.

[0130] That is, looking at the processing result delivery conversation pattern of Fig. 7(a), a status notification conversation component is mandatory, and based on the principle of naturalness, an understanding indicator conversation component and an implicit acknowledgment conversation component can be added before the status notification conversation component depending on the context. In addition, supporting grounds or helpful information can be provided together as an explanation conversation component after the status notification conversation component. At this time, the utterance time should not be prolonged.

[0131] According to the embodiments, even when adding useful explanations in the processing result delivery conversation pattern, it is desirable to keep the total length of the script within 150 characters. However, this is one embodiment, and the length of the script may be longer than 150 characters.

[0132] As described above, in the delivery of processing results as in the conversation pattern of Fig. 7(a), a status notification conversation component is configured to inform of the result of performing the instructions. At this time, similar to the information delivery conversation pattern, if there is additional information or useful work related to the result performed, suggestions can be added, and if not, it is decided whether to end the conversation by confirming the additional inquiry.

[0133] FIG. 7(b) is an example of a script that assumes a situation where a user requests a payment confirmation via email. The script of FIG. 7(b), written based on the processing result delivery conversation pattern of FIG. 7(a), includes, as an example, an implicit confirmation conversation component (e.g., to your email), a status notification information conversation component (e.g., We have sent you the payment confirmation), an explanation conversation component (e.g., Emails are sent sequentially, so it may take some time), and an additional inquiry confirmation conversation component (e.g., Do you have any other questions?).

[0134] That is, in the script example of Fig. 7(b), a status notification dialog component indicating that the payment confirmation has been successfully sent is first placed as a mandatory step. Next, an implicit confirmation dialog component, 'To your email,' is added to implicitly confirm that it has been sent via email. Additionally, considering that emails may not be received immediately, an explanation dialog component is added stating that it may take some time. Finally, even if the instructions are completed, the conversation is not immediately terminated, and a follow-up inquiry confirmation dialog component is used to check if there are any remaining requirements.

[0135] Additional instruction confirmation conversation pattern (S125)

[0136] In the present invention, the additional instruction confirmation conversation pattern is intended to confirm whether there are any requirements before ending the conversation in a situation where the user has rejected a proposal. The situation in which the additional instruction confirmation conversation pattern is used in the present invention is the same as the information transmission conversation pattern and the processing result transmission conversation pattern as follows. For example, in a situation where the user requests information, the user's utterance is successfully received (Input = YES) (S113), the content is understood (Match Result = YES) (S114), there is no further information to be confirmed from the user (Confirmation / Information Needed = NO) (S119), and a response is possible (Response Possible = YES) (S121). The present invention can use the additional instruction confirmation conversation pattern in such utterance situations.

[0137] In this case, the conversation components defined in the conversation pattern are used in a fixed order, and the conversation components indicated in parentheses are optional, and whether to use them is determined appropriately depending on the situation and context.

[0138] FIGS. 8(a) and FIGS. 8(b) are drawings showing the configuration of dialogue components of the additional instruction confirmation dialogue pattern of the present invention and an example of a script. Specifically, FIG. 8(a) shows an example of dialogue components constituting the additional instruction confirmation dialogue pattern, and FIG. 8(b) shows an example of a dialogue script written based on the additional instruction confirmation dialogue pattern of FIG. 8(a).

[0139] The additional instruction confirmation dialog pattern of Fig. 8(a) can be composed of an understanding indicator dialog component, a status notification dialog component, an explanation dialog component, and an additional inquiry confirmation dialog component in that order. In this case, the understanding indicator dialog component and the additional inquiry confirmation dialog component are mandatory dialog components that must be included in the additional instruction confirmation dialog pattern, while the status notification dialog component and the explanation dialog component are optional dialog components whose use is determined by the situation and context.

[0140] That is, looking at the additional instruction confirmation conversation pattern of Fig. 8(a), first, acceptance of the user's refusal is guided through the understanding indicator conversation component, and then the requirements before termination are confirmed through the additional inquiry confirmation conversation component. Then, depending on the context and situation of the conversation, the status can be announced through the status notification conversation component between the understanding indicator conversation component and the additional inquiry confirmation conversation component, or an explanation of the situation can be added through the explanation conversation component.

[0141] FIG. 8(b) is an example of a script assuming a situation where the user rejects a proposal. As an example, the script of FIG. 8(b), which is written based on the additional instruction confirmation conversation pattern of FIG. 8(a), includes an understanding indication conversation component (e.g., Understood) and an additional inquiry conversation component (e.g., Do you have any further questions?).

[0142] That is, the script example in Fig. 8(b) guides the acceptance of the refusal with the understanding indicator dialog component 'Understood', and checks whether there are any further requirements before closing with the additional inquiry confirmation dialog component 'Do you have any more questions?'.

[0143] Additional information request conversation pattern (S120)

[0144] In the present invention, the additional information request conversation pattern is a conversation pattern used when the user's requirements have been identified, but additional information or user confirmation is required to process them.

[0145] FIGS. 9(a) to 9(c) illustrate examples of the configuration of conversation components and scripts of the additional information request conversation pattern of the present invention. Specifically, FIG. 9(a) shows an example of conversation components constituting the additional information request conversation pattern, and FIGS. 9(b) and 9(c) show examples of conversation scripts written based on the additional information request conversation pattern of FIG. 9(a).

[0146] The situation in which the additional information request conversation pattern is used in the present invention is as follows. For example, it is the case where a user's utterance is successfully received (Input = YES) (S113), the content is understood (Match Result (Match) = YES) (S114), and information to be verified from the user is needed (Verification / Information Needed = Yes) (S119). The present invention can use the additional information request conversation pattern in such utterance situations.

[0147] In this case, the conversation components defined in the additional information request conversation pattern are used in a fixed order, and the conversation components indicated in parentheses are optional, and whether to use them is determined appropriately depending on the situation and context.

[0148] The additional information request conversation pattern of Fig. 9(a) can be composed of an understanding indication conversation component, an implicit acknowledgment conversation component, an explanation conversation component, a request conversation component, and an utterance example conversation component. In this case, the request conversation component is a required conversation component that must be included in the additional information request conversation pattern, and the understanding indication conversation component, the implicit acknowledgment conversation component, the explanation conversation component, and the utterance example conversation component are optional conversation components whose use is determined by the situation and context.

[0149] That is, looking at the additional information request conversation patterns and scripts of FIGS. 9(a) to 9(c), the request conversation component is placed at the end of the script to make it easier for the user to remember what task needs to be performed. Additionally, depending on the context and situation of the conversation, an understanding indication conversation component, an implicit confirmation conversation component, and an explanation conversation component may be placed before the request conversation component. As an embodiment of the present invention, the content of the request is limited to one to reduce the user's cognitive load. If more than one piece of information or confirmation is required, the request can be made by dividing it into steps.

[0150] In this invention, when making a request, the question must be in a form that allows the user to answer accurately and easily. Therefore, in this invention, numerical input, which is susceptible to errors in input and misrecognition, can be received via keypad input instead of spoken words. Furthermore, it is preferable to ask questions that can be answered with "yes" or "no," rather than questions that require memorizing specific words.

[0151] In the present invention, as a result of analyzing tens of thousands of scripts (e.g., 31,857) related to conversation patterns for requesting additional information, it was observed that the dropout rate was more than four times higher in cases where a specific word had to be answered than in cases where a 'yes / no' answer had to be answered.

[0152] FIG. 10 is a table showing the dropout rate according to the form of the query of the embodiments. FIG. 10 shows that the dropout rate is 2% when the answer is 'Yes / No', whereas it increases to 9% when asking if a specific word is correct, and to 10% when asking to select one of two. Therefore, in the present invention, rather than a form of answering 'Yes / No', when an answer is required by selecting from multiple options, the number of options and the number of characters per option are limited as an embodiment.

[0153] In particular, an analysis of hundreds of thousands of scripts (e.g., 293,943) related to additional information request conversation patterns revealed that, as shown in Figures 11 and 12, the sum of the total characters of the options had a greater impact on the dropout rate than the number of options.

[0154] FIG. 11 is a drawing showing an example of a dropout rate according to the number of options in the present invention. FIG. 12 is a drawing showing an example of a dropout rate according to the sum of the number of characters of the options in the present invention.

[0155] In other words, Figure 11 shows a weak correlation (r=-0.19) between the number of options and the dropout rate. On the other hand, Figure 12 shows a strong positive correlation (r=0.742) between the sum of the characters of the options and the dropout rate (calculated by substituting the range variable on the horizontal axis with representative values ​​of 5, 15, 25, 35, and 45 characters, respectively). These results indicate that, regardless of the number of options, the total number of characters of the options has a more significant impact on the user's cognitive burden. Since users must listen to and understand all the options, it can be concluded that information processing and selection are easier when the total length of the options is shorter.

[0156] In the present invention, in order to maintain a dropout rate of 10% or less, the total number of characters of the options can be limited to 10 characters or less, and the number of options can be limited to 2 by default, with a maximum of 5 characters per option.

[0157] FIG. 9(b) is an example script assuming a situation where a user requests a pause. As an example, the script of FIG. 9(b), written based on the additional information request conversation pattern of FIG. 9(a), includes an implicit acknowledgment conversation component (e.g., "To release the pause") and a request conversation component (e.g., "Please press the six-digit verification code sent via text message and the hash symbol"). That is, the script example of FIG. 9(b) first implicitly acknowledges the implicit acknowledgment conversation component "To release the pause" to indirectly indicate that the user's request has been understood, and then requests the verification code, which is information necessary to perform the pause, through the request conversation component.

[0158] FIG. 9(c) is an example script assuming a situation where a user has entered a pause period. As an example, the script of FIG. 9(c), written based on the additional information request dialogue pattern of FIG. 9(a), includes an explanation dialogue component (e.g., the period you entered is from January 2024 to December 2024) and a request dialogue component (e.g., is that correct?). That is, in the script example of FIG. 9(c), since there is a possibility of mis-input, mis-uttering, or mis-recognition of the period, a request is made to clarify the recognized period and to receive confirmation.

[0159] Conversation pattern unable to execute instructions (S122)

[0160] In this invention, the conversation pattern indicating that instructions cannot be executed serves to signal that further progress is impossible when the user's intent has been understood but a system failure has occurred or when the function cannot be provided by a voice-based AI counselor. Through this, the present invention can reduce user confusion by clearly and quickly informing them of the system's situation and limitations.

[0161] FIGS. 13(a) to 13(c) are drawings showing examples of the configuration of dialogue components and scripts of the instruction-unexecutable dialogue pattern of the present invention. Specifically, FIG. 13(a) shows an example of dialogue components constituting the instruction-unexecutable dialogue pattern, and FIGS. 13(b) and FIG. 13(c) show examples of dialogue scripts written based on the instruction-unexecutable dialogue pattern of FIG. 13(a).

[0162] The situation in which the instruction execution failure dialogue pattern is used in the present invention is as follows. For example, it is a case where a user's utterance is successfully received (Input = YES) (S113), the content is understood (Match Result (Match) = YES) (S114), there is no information to be verified from the user (Verification / Information Needed = Yes) (S119), and a response is not possible (Response Possible = No). The present invention can use the instruction execution failure dialogue pattern in such utterance situations.

[0163] At this time, as an example, the dialogue components defined in the dialogue pattern that cannot be executed are used in a fixed order, and the dialogue components indicated in parentheses are optional, and whether to use them is determined appropriately according to the situation and context.

[0164] The instruction execution failure conversation pattern of Fig. 13(a) may be composed of an explanation conversation component, a status notification conversation component / key information conversation component, an explanation conversation component, and an additional inquiry confirmation conversation component / suggestion conversation component in that order. In this case, the status notification conversation component / key information conversation component and the additional inquiry confirmation conversation component / suggestion conversation component are mandatory conversation components that must be included in the instruction execution failure conversation pattern, and the explanation conversation component is an optional conversation component whose use is determined by the situation and context.

[0165] That is, looking at the instruction execution failure dialogue pattern of Fig. 13(a), the system's status or limitations regarding the situation where instruction execution is impossible are conveyed through a status notification dialogue component or a key information dialogue component. At this time, depending on the context, an explanation dialogue component containing specific explanations can be placed before or after the status notification dialogue component / key information dialogue component. Subsequently, even if instruction execution is impossible, if there is an alternative, a solution is presented through a proposal dialogue component, and if an alternative cannot be presented, an inquiry can be made through an additional inquiry confirmation dialogue component to ask whether to continue the Q&A.

[0166] FIG. 13(b) is an example script assuming a situation where a user requests a rate plan change, but the voice-based AI agent cannot proceed with it and only an actual human agent can, and it is a time when connecting with a human agent is difficult. The script of FIG. 13(b), written based on the instruction execution failure conversation pattern of FIG. 13(a), includes, as an example, a core information conversation component (e.g., Rate plan changes are possible through an agent), an explanation conversation component (e.g., All consultation services have ended now. The hours when you can connect with an agent are weekdays from 9:00 AM to 6:00 PM, and consultation is available from 1:00 PM as the customer center lunch break is from 1:00 PM to 1:00 PM), and an additional inquiry confirmation conversation component (e.g., Do you have any other inquiries?). That is, the script example of FIG. 13(b) first informs the user that it is possible only through an agent via the core information conversation component, and additionally places an explanation conversation component stating that it is a time when connecting with an agent is difficult. Then, since it is a situation where an alternative cannot be presented, it checks whether to continue the Q&A via the additional inquiry confirmation conversation component.

[0167] FIG. 13(b) is an example script assuming a situation where a user requests the issuance of a tax invoice, but issuance is impossible due to an unknown error in the system. The script of FIG. 13(c), written based on the instruction execution failure dialogue pattern of FIG. 13(a), is an example of including a status notification dialogue component (e.g., "We cannot issue the tax invoice you requested") and a suggestion dialogue component (e.g., "Would you like us to connect you with a counselor for detailed guidance?"). That is, the script example of FIG. 13(c) first informs the system of its status through the status notification dialogue component, and then helps the user resolve the problem through the suggestion dialogue component, which is the suggestion to connect with a counselor.

[0168] No input conversation pattern (S116)

[0169] In the present invention, the no input dialogue pattern is used to inform the user that nothing was heard and to request speech when there is no voice signal input during the user's speech turn. Together with the instruction execution failure dialogue pattern and the no match result dialogue pattern described later, this can help the user proactively adjust their interaction with the system and prevent errors by notifying the error situation through a status notification dialogue component.

[0170] In the present invention, the no-input dialogue pattern is approached stepwise by considering situations where the user cannot input. That is, the no-input dialogue pattern can be used when there was a user utterance (S112) but no input was detected (S113), or when there is no user utterance input during the user's utterance turn.

[0171] FIGS. 14(a) to 14(c) illustrate examples of the configuration of dialogue components and scripts of the no-input dialogue pattern of the present invention. Specifically, FIG. 14(a) shows an example of dialogue components constituting the no-input dialogue pattern, and FIGS. 14(b) and FIGS. 14(c) show examples of dialogue scripts written based on the no-input dialogue pattern of FIG. 14(a).

[0172] In this case, the dialogue components defined in the no-input dialogue pattern are used in a fixed order, and the dialogue components indicated by parentheses are optional, and whether to use them is determined appropriately depending on the situation and context.

[0173] The no-input conversation pattern of Fig. 14(a) can be composed of a status notification conversation component and an utterance example conversation component. In this case, the status notification conversation component is a required conversation component that must be included in the no-input conversation pattern, and the utterance example conversation component is an optional conversation component whose use is determined by the situation and context.

[0174] That is, looking at the no-input conversation pattern in Fig. 14(a), the first no-input situation may be one where the user did not hear or understand the system's request, in which case the previous script containing the status notification conversation component is repeated to request the user to speak again. The second no-input situation may be one where the user does not know how to respond, in which case an utterance example conversation component is added to assist the user in speaking. Nevertheless, if the third no-input situation occurs, it is assumed that the user is in a situation where they cannot speak, and a farewell is said and the conversation is ended through the farewell conversation component.

[0175] FIGS. 14(b) and FIGS. 14(c) are examples of scripts assuming cases where no input occurs once and twice. The script in FIG. 14(b), written based on the no input dialogue pattern in FIG. 14(a), represents the first case of no input and notifies that the user's speech was not heard through a status notification dialogue component (e.g., "Customer, I didn't hear you well."). Then, the previous script is repeated. The script in FIG. 14(c) represents the second case of no input and helps the user speak by adding an utterance example dialogue component.

[0176] No match conversation pattern (S117)

[0177] In this invention, the "no match result" conversation pattern aims to notify the user that it has not understood and to request a re-utterance when the user's voice has been input but the requirements cannot be clearly understood or when information outside the system's processing range has been input. This is used in two situations where understanding has not been achieved. One is when the user's utterance is not understood at all, or when the input information is outside the system's processing range, resulting in no possible outcome. The second is when the user's utterance is partially understood but the entire sentence is not, or when the user asks multiple questions simultaneously, requiring clarification of ambiguous requests for the system to process them.

[0178] FIG. 15 is a diagram showing examples of the configuration of dialogue components and scripts of the no-match-result dialogue pattern of the present invention.

[0179] In the present invention, the no match result conversation pattern can be used in a situation where the user's utterance is successfully received (input = YES) (S113), but the content is not understood (match result (Match) = No) (S114).

[0180] In the present invention, the no-match-result conversation pattern can be divided into incomprehensible and incorrect input conversation patterns and partially understood or two or more intent conversation patterns.

[0181] Looking at the incomprehensible and incorrect input conversation patterns in Fig. 15, first, a status notification conversation component indicates that the user did not understand and requests that the previous script be repeated. The script example assumes a situation where the user's utterance is not understood at all. In this case, the user first indicates that they did not understand through a status notification conversation component (e.g., "I didn't understand well") and then requests that the previous script be repeated.

[0182] In FIG. 15, when looking at a conversation pattern involving partial understanding or two or more intents (intentions contained in a single script), the system seeks confirmation from the user through a request conversation component regarding the part that is more clearly understood. According to the embodiments, in the case of partial understanding, the system asks for the function or service most relevant to the understood part, and in the case of two or more intents, it seeks confirmation regarding the more clearly recognized intent or the first mentioned intent. Furthermore, the script example assumes a situation where the IPTV inquiry is recognized, but other utterances are not understood. In this case, a request to confirm whether the content of the query is correct is performed through a request conversation component using the recognized IPTV inquiry keyword.

[0183] Welcome greeting conversation pattern (S111)

[0184] In the present invention, the welcome conversation pattern initiates a conversation through a greeting conversation component and an utterance example conversation component, and aims to help the user form appropriate expectations regarding the intelligence and functions of the system.

[0185] FIGS. 16(a) and FIGS. 16(b) are drawings showing an example of the configuration of conversation components and a script of the welcome greeting conversation pattern of the present invention. Specifically, FIG. 16(a) shows an example of conversation components constituting the welcome greeting conversation pattern, and FIG. 16(b) shows an example of a conversation script written based on the welcome greeting conversation pattern of FIG. 16(a).

[0186] In other words, in the welcome conversation pattern of the present invention, both the greeting conversation component and the utterance example conversation component are used as essential conversation components, and the greeting conversation component is placed only once at the start of the conversation. Even if the conversation is restarted from the beginning in a no-input situation or a situation where additional inquiries are confirmed, the greeting conversation component is not repeated.

[0187] The script of Fig. 16(b), written based on the welcome conversation pattern of Fig. 16(a), includes a greeting conversation component (e.g., Hello. I am an LG Uplus AI counselor here to quickly assist you with various inquiries 24 hours a day) and a speech conversation component (e.g., Please simply tell me what task you want, such as billing charges or roaming).

[0188] Farewell conversation pattern (S115)

[0189] In the present invention, the farewell conversation pattern is used to indicate that the interaction with the voice-based AI agent has been completely terminated according to the user's request or situation. If the user no longer has any requests for the voice-based AI agent, the conversation is terminated with a closing conversation component such as "Have a nice day." However, the farewell conversation pattern can be used not only when there are no further requests, but also when it is difficult to proceed due to repeated errors, or in situations where the connection with the voice-based AI agent must be terminated and connected to a human agent. Here, errors are indicated by no input (S116) and no match result (S117), and if these errors accumulate three or more times consecutively (S118), it is determined that the conversation cannot proceed, and the farewell conversation pattern is applied.

[0190] FIGS. 17(a) and FIGS. 17(b) are drawings showing an example of the configuration of conversation components and a script of the farewell conversation pattern of the present invention. Specifically, FIG. 17(a) shows an example of conversation components constituting the farewell conversation pattern, and FIG. 17(b) shows an example of a conversation script written based on the farewell conversation pattern of FIG. 17(a).

[0191] At this time, as an example, the conversation components defined in the farewell conversation pattern are used in a fixed order, and the conversation components indicated in parentheses are optional, and whether to use them is determined appropriately depending on the situation and context.

[0192] The farewell conversation pattern of Fig. 17(a) can be composed of an understanding expression conversation component and a closing conversation component. In this case, the closing conversation component is a mandatory conversation component that must be included in the farewell conversation pattern, and the understanding expression conversation component is an optional conversation component whose use is determined by the situation and context.

[0193] That is, looking at the farewell conversation pattern of Fig. 17(a), an understanding indicator conversation component indicating that the user's intention to end is understood is included as an option, and a closing conversation component is placed as a requirement.

[0194] FIG. 17(b) is an example script assuming a situation where a user requests a connection to a counselor. The script in FIG. 17(b), written based on the farewell conversation pattern of FIG. 17(a), first indicates that the request has been understood through the understanding indicator conversation component 'Yes', and then places the closing conversation component 'I will connect you to a counselor'. In this situation, since the user wants a quick resolution through an actual human counselor, it is desirable not to include unnecessary farewell phrases such as 'Have a nice day'.

[0195] VUX Guideline Application Cases

[0196] Next, we will explain the application examples of the VUX guidelines.

[0197] Based on the aforementioned VUX guidelines, the present invention can modify the conversation scripts of existing voice-based AI counselors. In the present invention, the modification process follows the following procedure.

[0198] First, check whether the flow of the conversation conforms to VUX principles. For example, this applies to cases where the flow itself needs modification, such as asking the user multiple questions at once or ending the conversation without confirming additional instructions. Once the flow is properly determined, proceed with the process of modifying the conversation script as shown in Fig. 18.

[0199] FIG. 18 is a flowchart illustrating one embodiment of the method for modifying a dialogue script of the present invention.

[0200] That is, as the first step in modifying the dialogue script, the script is analyzed according to the flow of the conversation to determine which conversation pattern it corresponds to (S211).

[0201] When a conversation pattern is determined as a result of the analysis in step S211, the current script is reviewed to see if it is suitable for configuring the conversation components of the corresponding conversation pattern (S212). If it is confirmed in step S212 that the current script is not suitable for configuring the conversation components of the corresponding conversation pattern, it is checked to see if there are any unnecessary conversation components or missing conversation components (S213). Based on the results of the check in step S213, necessary conversation components are added, or unnecessary conversation components and their corresponding phrases are deleted (S214). Afterwards, the conversation script is modified according to the conversation pattern and conversation component rules (S215).

[0202] FIGS. 19 to 21 are drawings showing examples of conversation scripts modified according to conversation patterns and conversation component rules after going through the steps of FIG. 18. That is, FIGS. 19 to 20 show examples of conversation scripts modified based on the VUX guidelines of the present invention.

[0203] In FIG. 19, the existing conversation script corresponds to a conversation pattern where instructions cannot be executed. However, an unnecessary apology is included before the core information conversation component indicating that a connection to a counselor is necessary (e.g., "Set-top box reinstallation can be guided through a counselor"), and excessive content that could be explained during the reinstallation process via a counselor is added, causing the user (i.e., customer) to hear an excessive amount of information before connecting to a counselor. Therefore, as shown in FIG. 19, the present invention removes unnecessary explanations and concisely modifies the conversation script to consist only of the core information conversation component and the additional inquiry confirmation conversation component. This modification allows the user to quickly recognize necessary information through the top-down placement of the core information conversation component and reduces unnecessary cognitive burden.

[0204] In FIG. 20, the conversation script is an example of a situation where a user has rejected a proposal from a voice-based AI agent once, and corresponds to a conversation pattern confirming additional instructions. In existing conversation scripts, another proposal is made consecutively even after the user's proposal has already been rejected. However, making another proposal consecutively after the user's proposal has already been rejected can impair the user experience. Therefore, the present invention modifies the conversation script by removing the proposal conversation component from the existing conversation script and adding an additional inquiry confirmation conversation component, as shown in FIG. 20.

[0205] In FIG. 21, the conversation script represents a situation where the user's utterance is not understood, corresponding to a no-match conversation pattern. In existing conversation scripts, the repetition of the previous script is omitted, which is insufficient to help the user attempt to utter again. Therefore, according to the no-match conversation pattern rule as shown in FIG. 21, if it is the first no-match conversation pattern situation, the repetition of the previous script is added, and if it is the second no-match conversation pattern situation, an utterance example conversation component is added along with the repetition of the previous script to make it easier for the user to utter.

[0206] Streamlining script writing and review using customized AI language models

[0207] The following describes an efficient method for writing and reviewing scripts using a customized AI language model.

[0208] Generally, the conversation scripts used by voice-based AI agents are so voluminous that no single person can manage them all; consequently, the task of writing and refining them must be shared among multiple people. This can lead to situations where maintaining consistency in the scripts becomes difficult. In particular, while VUX guidelines are designed to enhance the voice user experience, their sheer volume and complexity limit the ability of multiple workers to understand and apply them beyond a certain level. Due to these differences in understanding among workers, there is a risk that the quality and consistency of the conversation scripts may deteriorate, even with the existence of VUX guidelines.

[0209] The present invention proposes an AI-based agent to overcome the limitations of applying VUX guidelines. In the present invention, the AI-based agent may be referred to as an AI-based agent or Colin. In the present invention, the AI-based agent can improve the efficiency and effectiveness of tasks by supporting the work of script writers and script reviewers through the use of a customized AI language model. In the present invention, the customized AI language model is, as an example, a GPT, which is a custom GPT (Generative Pre-trained Transformer) function based on ChatGPT.

[0210] In the present invention, while writing a script based on VUX guidelines, the script writer can check whether the written script complies well with VUX guidelines or discuss any questions or concerns with an AI-based agent.

[0211] In the present invention, a script reviewer can quickly receive a preliminary review of a large volume of scripts by utilizing an AI-based agent.

[0212] In addition, in order to minimize the hallucination problem that may occur due to the characteristics of a Large Language Model (LLM), the present invention is designed so that an AI-based agent infers only within limited data and standardizes the output format as one embodiment.

[0213] In other words, GPTs operate based on two concepts: knowledge and guidelines. When a file is uploaded to the knowledge section, the information in that file can be utilized during a conversation with the GPT. In the guidelines section, principles, attitudes, and tone of voice that the GPT must observe during the conversation can be explicitly defined and specified.

[0214] Accordingly, to minimize hallucination problems in the responses of an AI-based agent, the present invention limits the basis of inference through knowledge and standardizes the form of the response through guidelines when the AI-based agent operates based on knowledge and guidelines. At this time, the knowledge contains a document describing VUX guidelines, a Do / Don't checklist based on VUX guidelines, and numerous (e.g., about 1,000) examples of scripts before and after the application of VUX guidelines, and the guidelines standardize the form of the AI-based agent's response by causing the AI-based agent to output the following five items regarding the script. At this time, the 12 conversation components, 9 conversation patterns, and the combination rules of conversation components for each conversation pattern are to be referred to in the description of FIGS. 3 to 16 described above, and are omitted here to avoid redundant explanation.

[0215] First, when a script is input, it infers which conversation pattern it corresponds to and outputs it.

[0216] Second, explain the basis inferred from the conversation pattern based on knowledge.

[0217] Third, based on the corresponding conversation pattern, it is analyzed whether the conversation script complies with the rules for combining conversation components. This includes whether it includes necessary conversation components and whether it includes unnecessary conversation components.

[0218] Fourth, determine whether modifications to the dialogue script are necessary based on the third analysis and the principles of the VUX guidelines; if modifications are required, analyze and explain the issues. At this time, the basis for the analysis must be the file uploaded to the knowledge database, and the response is structured to mention the relevant content.

[0219] Fifth, we propose a revised phrase.

[0220] Improvement of script writing work processes through AI-based agents

[0221] By utilizing AI-based agents in their work, script writers and reviewers have significantly improved the process of applying VUX guidelines and developing conversational scripts.

[0222] Figure 22 is a diagram showing an example of a business process utilizing the AI-based agent of the present invention. Specifically, Figure 22 visually illustrates a conventional business process (left diagram) and a business process utilizing the AI-based agent (right diagram). Previously, a planner first planned the functions of a voice-based AI counselor and the corresponding conversation flow structure. Subsequently, multiple script writers wrote conversation scripts for their respective functions based on VUX guidelines. Next, a script reviewer reviewed the written scripts based on VUX guidelines. As previously mentioned, even with the existence of VUX guidelines, differences in understanding levels between script writers and reviewers could occur; therefore, a step was required for a VUX expert to conduct a full review of the written scripts to ensure consistency. Finally, the script writers and reviewers went through the process of writing and reviewing the final scripts. As such, a lengthy process was required in the past to consistently apply a large amount of guidelines.

[0223] However, the present invention achieves the effect of significantly shortening business processes by utilizing an AI-based agent.

[0224] First, a planner plans the functions and conversation flow structure of the voice-based AI agent, after which a script writer writes the script in collaboration with an AI-based agent. While the script writer makes the final decision on the script, they can work by consulting with the AI-based agent regarding any confusing or concerning parts during the writing process. Subsequently, the reviewer can also utilize the AI-based agent during the final review of the written script, enabling a more efficient review process. Through the assistance of the AI-based agent, the scripts written and reviewed in this manner can ensure a certain degree of consistency between the script writers and the reviewers.

[0225] FIG. 23 is a diagram showing an example of creating a conversation script using the AI-based agent of the present invention. FIG. 23 is an example in which the task of analyzing and modifying (or correcting) a conversation script based on the VUX guidelines described in FIG. 18 is performed through the AI-based agent. That is, the AI-based agent analyzes which conversation pattern the conversation script of a voice-based AI counselor corresponds to to determine the conversation pattern, and reviews whether the conversation script is suitable for the composition of conversation components of the determined conversation pattern. If it is not suitable, it checks whether there are unnecessary conversation components or missing conversation components, and based on the results of the check, adds necessary conversation components or deletes unnecessary conversation components and corresponding phrases. Afterward, the script is modified according to the conversation pattern and conversation component rules.

[0226] More specifically, the script author can identify which conversation pattern the script corresponds to based on the feedback from the AI-based agent. Once the conversation pattern is determined, the author can check whether the current script is properly composed of the conversation components of that pattern. If modifications are necessary—that is, if the current script is not properly composed of the conversation components of the pattern—the author can create a revised version of the script by referring to the AI-based agent's suggestions. In other words, if there are unnecessary or missing conversation components, the author can add the necessary components, delete unnecessary components and corresponding phrases, and modify the script according to the rules of the conversation pattern and its components.

[0227] In this case, the analysis of conversational components and proposed modifications based on whether modifications are necessary can be utilized only when the conversational pattern has been accurately analyzed, and if there are disagreements regarding the AI-based agent's analysis of the conversational pattern, opinions can be coordinated and discussed.

[0228] FIG. 24 is a diagram showing another example of writing a conversation script using the AI-based agent of the present invention. Specifically, FIG. 24 shows a case where a script writer expresses a dissenting opinion regarding the results of the conversation pattern analysis by the AI-based agent. The AI-based agent provides answers based on VUX guidelines and conversation script examples uploaded to the knowledge, and the script writer can write or modify the conversation script by referring to these opinions.

[0229] Furthermore, script reviewers can pre-review scripts with the assistance of an AI-based agent. For example, if dialogue script sentences are entered into the AI-based agent's dialog box, the agent examines which dialogue pattern each script corresponds to, whether dialogue components are properly configured, and whether modifications are necessary, then suggests revisions. If necessary, it is also possible to configure the system to output the dialogue pattern of each script or to analyze only whether modifications are required.

[0230] FIG. 25 is a diagram showing another example of creating a conversation script using the AI-based agent of the present invention. Specifically, FIG. 25 shows a part of an example in which multiple scripts are input into the AI-based agent in batches to perform analysis of conversation patterns, basis for pattern inference, and conversation components.

[0231] FIG. 26 is a block diagram showing an example of a system for writing a conversation script for a voice-based AI counselor according to the present invention. The system of FIG. 26 may be implemented in hardware, software, a processor connected to memory, and / or a combination thereof.

[0232] In FIG. 26, the memory can be connected to the processor during operation and can store programs and / or instructions for processing and controlling the processor, and can store data and information used in the present invention, control information necessary for data and information processing according to the present invention, temporary data generated during the data and information processing process, etc. The memory can be implemented as a storage device such as ROM (Read Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, SRAM (Static RAM), HDD (Hard Disk Drive), SSD (Solid State Drive), etc.

[0233] In addition, the processor may be operatively connected to memory and / or network interfaces and may perform various control functions to carry out the proposed method of the present invention. The processor may also be referred to as a controller, microcontroller, microprocessor, microcomputer, etc. The proposed method of the present invention may be implemented by hardware, firmware, software, or a combination thereof. When implementing the present invention using hardware, an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), etc. configured to carry out the present invention may be provided in the processor (21). Meanwhile, when implementing the proposed method of the present invention using firmware or software, the firmware or software may include instructions related to modules, procedures, or functions that perform functions or operations necessary to implement the proposed method of the present invention, and the instructions may be stored in memory or on a computer-readable recording medium (not shown) separately from memory, and configured so that the system implements the proposed method of the present invention when executed by a processor.

[0234] Additionally, the system may include a network interface device. The network interface device is connected to the processor during operation, and the processor controls the network interface device to transmit or receive wireless / wired signals carrying information and / or data, signals, messages, etc., through a wireless / wired network.

[0235] As an example, the memory stores a VUX guideline that includes rules for the aforementioned multiple conversation components and multiple conversation patterns, and the processor creates or modifies a conversation script of a voice-based AI agent based on the VUX guideline, and controls the voice-based AI agent to provide a response to a customer based on the conversation script.

[0236] In addition, the processor may operate an AI-based agent to analyze which conversation pattern the conversation script of a voice-based AI agent corresponds to, determine the conversation pattern, and review whether the conversation script is suitable for the configuration of conversation components of the determined conversation pattern. If it is not suitable, it may check whether there are unnecessary or missing conversation components in the conversation script, and based on the results of the check, add necessary conversation components to the conversation script or delete unnecessary conversation components and corresponding phrases from the conversation script. Furthermore, the conversation script may be modified according to the conversation pattern and conversation component rules.

[0237] To this end, as an example, the memory stores knowledge and guidelines including a document describing the VUX guidelines of the present invention derived through the aforementioned process, a Do / Don't checklist based on the VUX guidelines, and multiple script examples before and after the application of the VUX guidelines. That is, knowledge is information uploaded as a file, etc., so that an AI-based agent can utilize it. At this time, the VUX guidelines consist of 12 conversation components and 9 conversation patterns described in FIGS. 3 to 16, and combination rules for conversation components for each conversation pattern.

[0238] That is, when the processor generates an answer through an AI-based agent, it limits the range of information that the AI-based agent can reference through knowledge stored in the memory and standardizes the format of the answer through guidelines.

[0239] As described above, the present invention establishes VUX principles to provide a better user experience in the design of conversation scripts for a voice-based AI system (or voice-based AI agent) and establishes a system of conversation components and conversation patterns based on these principles. Furthermore, the present invention generates an AI-based agent by utilizing a custom GPT function based on ChatGPT. In the present invention, the AI-based agent supports the script writing and verification process based on VUX guidelines, and enables business processes to be streamlined by maintaining consistency while processing a large volume of scripts. As such, the present invention presents a guide that can be practically reflected based on data currently in service, thereby enhancing business efficiency and consistency, and the specific effects are as follows.

[0240] First, we established VUX guideline principles that voice-based AI agents (or callbots) can refer to when writing scripts. In other words, through VUX guideline principles that consider the characteristics of conversation and voice interfaces, it has become possible to work based on principles and the corresponding value priorities when writing scripts.

[0241] Second, we established a system of conversation components and patterns. Specifically, we structured the scripts of voice-based AI agents currently in service to lay the foundation for logically designing sentences.

[0242] Third, we developed an AI-based agent based on VUX guidelines. This enables us to overcome differences in the level of understanding of VUX guidelines to some extent and improve business processes by streamlining script writing and verification.

[0243] Each of the aforementioned parts, modules, or units may be software, processors, or hardware parts that execute successive processes stored in memory (or storage units). Each step described in the aforementioned embodiments may be performed by processors, software, or hardware parts. Each module / block / unit described in the aforementioned embodiments may operate as a processor, software, or hardware. Additionally, the methods presented in the embodiments may be executed as code. This code may be written to a storage medium that is readable by a processor and thus read by a processor provided by the apparatus.

[0244] Although the drawings have been described separately for convenience of explanation, it is also possible to design a new embodiment by combining the embodiments described in each drawing. Furthermore, it is within the scope of the present disclosure to design a computer-readable recording medium on which a program for executing the previously described embodiments is recorded, as required by a person skilled in the art.

[0245] The apparatus and method according to the present disclosure are not limited to the configurations and methods of the embodiments described above; rather, the embodiments may be configured by selectively combining all or part of each embodiment so that various modifications can be made.

[0246] In this disclosure, terms such as first, second, etc., may be used to describe various components of this disclosure. However, the interpretation of the various components according to this disclosure should not be limited by these terms. These terms are merely used to distinguish one component from another. For example, a first user input signal may be referred to as a second user input signal. Similarly, a second user input signal may be referred to as a first user input signal. The use of these terms should be interpreted as not departing from the scope of the various embodiments. Although the first user input signal and the second user input signal are both user input signals, they do not mean the same user input signals unless clearly indicated in the context.

[0247] The terms used to describe the present disclosure are intended for the purpose of describing specific embodiments and are not intended to limit the embodiments. As used in the description of the embodiments and in the claims, the singular is intended to include the plural unless explicitly indicated in the context. Also, the expression "and / or" is used to mean including all possible combinations between the terms. The expression "includes" describes the presence of features, numbers, steps, elements, and / or components and does not mean that additional features, numbers, steps, elements, and / or components are not included. Conditional expressions such as "if" or "when" used to describe the present disclosure are not to be interpreted as being limited to optional cases. It is intended that when a specific condition is satisfied, a related action is performed in response to a specific condition, or a related definition is interpreted.

Claims

Claim 1 A method comprising: a step of writing or modifying a conversation script of an artificial intelligence (AI) agent based on a voice customer experience (VUX) guideline that includes rules for multiple conversation components and multiple conversation patterns; and a step of the AI ​​agent conducting a conversation with a customer based on the conversation script, wherein each conversation component is a minimum unit of conversation and each conversation pattern is a combination of one or more conversation components according to the conversation context. Claim 2 A method according to claim 1, wherein the plurality of conversation components comprises one or more of a greeting conversation component, an utterance example conversation component, an understanding indication conversation component, a key information conversation component, an explanation conversation component, a request conversation component, an implicit acknowledgment conversation component, a status notification conversation component, a suggestion conversation component, an apology conversation component, an additional inquiry acknowledgment conversation component, or a closing conversation component. Claim 3 A method according to claim 2, wherein the plurality of conversation patterns comprises one or more of an information transmission conversation pattern, a processing result transmission conversation pattern, an additional instruction confirmation conversation pattern, an additional information request conversation pattern, an instruction execution failure conversation pattern, a no input conversation pattern, a no matching result conversation pattern, a welcome conversation pattern, or a farewell conversation pattern. Claim 4 In claim 3, the information transmission conversation pattern is composed in the order of the understanding indication conversation component, the implicit acknowledgment conversation component, the key information conversation component, the explanation conversation component, and the additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the key information conversation component and the additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the understanding indication conversation component, the implicit acknowledgment conversation component, and the explanation conversation component are optional. Claim 5 In claim 3, the processing result delivery conversation pattern is composed in the order of the understanding indication conversation component, the implicit acknowledgment conversation component, the status notification conversation component, the explanation conversation component, and the additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the status notification conversation component and the additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the understanding indication conversation component, the implicit acknowledgment conversation component, and the explanation conversation component are optional. Claim 6 In claim 3, the additional instruction confirmation dialog pattern is composed in the order of the understanding display dialog component, the status notification dialog component, the explanation dialog component, and the additional inquiry confirmation dialog component, wherein the understanding display dialog component and the additional inquiry confirmation dialog component are mandatory, and the status notification dialog component and the explanation dialog component are optional. Claim 7 In claim 3, the additional information request conversation pattern is composed of the understanding indication conversation component, the implicit acknowledgment conversation component, the explanation conversation component, the request conversation component, and the utterance example conversation component in that order, wherein the request conversation component is mandatory, and the understanding indication conversation component, the implicit acknowledgment conversation component, the explanation conversation component, and the utterance example conversation component are optional. Claim 8 In claim 3, the above instruction execution impossible conversation pattern is composed in the order of the above explanation conversation component, the above status notification conversation component / key information conversation component, the above explanation conversation component, and the above additional inquiry confirmation conversation component / suggestion conversation component, wherein the above status notification conversation component / key information conversation component and the above additional inquiry confirmation conversation component / suggestion conversation component are mandatory, and the above explanation conversation component is optional. Claim 9 In claim 3, the above-mentioned no-input conversation pattern is composed of the above-mentioned status notification conversation component and the above-mentioned utterance example conversation component in that order, wherein the above-mentioned status notification conversation component is mandatory and the above-mentioned utterance example conversation component is optional. Claim 10 In claim 3, the above-mentioned no-match-result conversation pattern is divided into a first no-match-result conversation pattern and a second no-match-result conversation pattern, wherein the first no-match-result conversation pattern is a conversation pattern corresponding to incomprehensibility or incorrect input, and the second no-match-result conversation pattern is a conversation pattern of partial understanding or two or more intents, and the first no-match-result conversation pattern is composed in the order of the status notification conversation component, the previous script repetition, and the utterance example conversation component, wherein the status notification conversation component is mandatory and the utterance example conversation component is optional, and the second no-match-result conversation pattern is composed of the request conversation component. Claim 11 In claim 3, the welcome greeting conversation pattern is composed of the greeting conversation component and the utterance example conversation component in that order, and the greeting conversation component and the utterance example conversation component are mandatory, method. Claim 12 In claim 3, the farewell conversation pattern is composed of the above-mentioned understanding expression conversation component and the above-mentioned closing conversation component in that order, wherein the above-mentioned closing conversation component is mandatory and the above-mentioned understanding expression conversation component is optional. Claim 13 A method according to claim 1, wherein the step of modifying the conversation script comprises: a step of analyzing the conversation script to determine a conversation pattern; a step of, when the conversation pattern is determined, checking whether the conversation script is suitable for the configuration of conversation components of the determined conversation pattern; a step of, when it is confirmed that the conversation script is not suitable for the configuration of conversation components of the conversation pattern, checking whether there are unnecessary or missing conversation components in the conversation script; a step of adding necessary conversation components to the conversation script or deleting unnecessary conversation components and corresponding phrases from the conversation script based on the result of the check; and a step of modifying the conversation script according to the conversation pattern and the rules for configuring conversation components of the conversation pattern after the steps are performed. Claim 14 A method according to claim 1, wherein the creation, modification, or verification of the conversation script is performed through an AI-based agent. Claim 15 A system comprising: a memory storing voice customer experience (VUX) guidelines including rules for multiple conversation components and multiple conversation patterns; and a processor for creating or modifying a conversation script of an AI agent based on the VUX guidelines and controlling the AI ​​agent to proceed with a conversation with a customer based on the conversation script, wherein each conversation component is a minimum unit of conversation and each conversation pattern is a combination of one or more conversation components according to the conversation context. Claim 16 In claim 15, the plurality of dialogue components comprises one or more of a greeting dialogue component, an utterance example dialogue component, an indication of understanding dialogue component, a key information dialogue component, an explanation dialogue component, a request dialogue component, an implicit acknowledgment dialogue component, a status notification dialogue component, a suggestion dialogue component, an apology dialogue component, an additional inquiry acknowledgment dialogue component, or a closing dialogue component. Claim 17 In claim 16, the plurality of conversation patterns comprises one or more of an information transmission conversation pattern, a processing result transmission conversation pattern, an additional instruction confirmation conversation pattern, an additional information request conversation pattern, an instruction execution failure conversation pattern, a no input conversation pattern, a no matching result conversation pattern, a welcome conversation pattern, or a farewell conversation pattern. Claim 18 In claim 17, the information transmission conversation pattern is composed in the order of the understanding indication conversation component, the implicit acknowledgment conversation component, the key information conversation component, the explanation conversation component, and the additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the key information conversation component and the additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the understanding indication conversation component, the implicit acknowledgment conversation component, and the explanation conversation component are optional, a system. Claim 19 In claim 17, the processing result delivery conversation pattern is composed in the order of the understanding indication conversation component, the implicit acknowledgment conversation component, the status notification conversation component, the explanation conversation component, and the additional inquiry acknowledgment conversation component / suggestion conversation component, wherein the status notification conversation component and the additional inquiry acknowledgment conversation component / suggestion conversation component are mandatory, and the understanding indication conversation component, the implicit acknowledgment conversation component, and the explanation conversation component are optional, a system. Claim 20 In claim 17, the additional instruction confirmation dialog pattern is composed of the understanding display dialog component, the status notification dialog component, the explanation dialog component, and the additional inquiry confirmation dialog component in that order, wherein the understanding display dialog component and the additional inquiry confirmation dialog component are mandatory, and the status notification dialog component and the explanation dialog component are optional, a system. Claim 21 In claim 17, the additional information request conversation pattern is composed of the above-mentioned understanding indication conversation component, the above-mentioned implicit acknowledgment conversation component, the above-mentioned explanation conversation component, the above-mentioned request conversation component, and the above-mentioned utterance example conversation component in that order, wherein the above-mentioned request conversation component is mandatory, and the above-mentioned understanding indication conversation component, the above-mentioned implicit acknowledgment conversation component, the above-mentioned explanation conversation component, and the above-mentioned utterance example conversation component are optional, a system. Claim 22 In claim 17, the above instruction execution failure conversation pattern is composed of the above explanation conversation component, the above status notification conversation component / key information conversation component, the above explanation conversation component, and the above additional inquiry confirmation conversation component / suggestion conversation component in that order, wherein the above status notification conversation component / key information conversation component and the above additional inquiry confirmation conversation component / suggestion conversation component are mandatory, and the above explanation conversation component is optional, a system. Claim 23 In claim 17, the above-mentioned no-input conversation pattern is composed of the above-mentioned status notification conversation component and the above-mentioned utterance example conversation component in that order, wherein the above-mentioned status notification conversation component is mandatory and the above-mentioned utterance example conversation component is optional, a system. Claim 24 In claim 17, the above-mentioned no-match-result conversation pattern is divided into a first no-match-result conversation pattern and a second no-match-result conversation pattern, wherein the first no-match-result conversation pattern is a conversation pattern corresponding to incomprehensibility or incorrect input, and the second no-match-result conversation pattern is a conversation pattern of partial understanding or two or more intents, and the first no-match-result conversation pattern is composed in the order of the status notification conversation component, the previous script repetition, and the utterance example conversation component, wherein the status notification conversation component is mandatory and the utterance example conversation component is optional, and the second no-match-result conversation pattern is composed of the request conversation component, a system. Claim 25 In claim 17, the welcome greeting conversation pattern is composed of the greeting conversation component and the utterance example conversation component in that order, and the greeting conversation component and the utterance example conversation component are mandatory, in a system. Claim 26 In claim 17, the farewell conversation pattern is composed of the above-mentioned understanding expression conversation component and the above-mentioned closing conversation component in that order, wherein the above-mentioned closing conversation component is mandatory and the above-mentioned understanding expression conversation component is optional, a system. Claim 27 In claim 15, the processor analyzes the conversation script to determine a conversation pattern, and when the conversation pattern is determined, checks whether the conversation script is suitable for the configuration of conversation components of the determined conversation pattern, and when it is confirmed that the conversation script is not suitable for the configuration of conversation components of the conversation pattern, checks whether there are unnecessary or missing conversation components in the conversation script, and based on the result of the check, adds necessary conversation components to the conversation script or deletes unnecessary conversation components and corresponding phrases from the conversation script, and after the operation, modifies the conversation script according to the conversation pattern and the configuration rules for the conversation components of the conversation pattern. Claim 28 In claim 15, the creation, modification, or verification of the above conversation script is performed through an AI-based agent operated by the processor, in a system. Claim 29 A computer program stored on a computer-readable recording medium that is combined with a computer, which is hardware, to perform the method of claim 1.

Images