KR-102964041-B1 - Method for providing answer that minimizing error response case about user's input and apparatus thereof

Abstract

An embodiment of the present invention discloses a method for providing an answer that minimizes answer error cases for a user input, comprising: a step of performing natural language processing on a received user input to calculate domain and intent information for the user input; a step of determining whether there is intent information among previously stored intent information that matches the calculated intent information; a step of determining, if there is no matching intent information, whether there is a domain among a plurality of domains that corresponds to the domain for the user input; and a step of generating and outputting an answer for the user input based on prompt conditions set in the domain and the corresponding domain.

Inventors

• 이단비
• 이주영
• 김정도
• 김혜영
• 이종호
• 조윤아

Assignees

• 포티투닷 주식회사

Dates

Publication Date

20260513

Application Date

20231121

Claims (10)

1. A step of performing natural language processing on a received user input to produce domain and intent information for said user input; a step of determining whether there is intent information among the stored intent information that matches said intent information; If there is intent information that matches the intent information calculated above, a step of generating and outputting an answer to the user's input according to a script set in a scenario corresponding to the matched intent information; If there is no intent information matching the intent information calculated above, a step of determining whether there is a domain among a plurality of domains that corresponds to the domain for the user's input; If there is a corresponding domain, a step of generating and outputting an answer to the user's input through a generative AI model based on the corresponding domain and prompt conditions set in the corresponding domain; and A method for providing an answer that minimizes cases of answer errors for user input, comprising the step of, when there is no corresponding domain, inputting the result of the natural language processing into the generative AI model to generate and output an answer corresponding to the user input.
2. In paragraph 1, The above user input is, A method for providing answers that minimizes instances of answer errors in response to user input, which is the voice of an authenticated user.
3. delete
4. delete
5. In paragraph 1, A method for providing answers that minimizes instances of answer errors in response to user input, further comprising the step of providing realtime guidance through an agent when it is detected that the user input contains a preset forbidden word.
6. In paragraph 1, The above prompt conditions are, A method for providing an answer that minimizes answer error cases for user input, including a limit value on the number of characters included in the above answer.
7. In paragraph 1, The above prompt conditions are, A method for providing answers that minimizes instances of answer errors in response to user input by restricting the writing style constituting the above answer to an honorific style.
8. In paragraph 1, The step of generating and outputting the above answer is, A method for providing answers that minimizes cases of answer errors for user input, by further considering reference materials linked to the above domain and generating and outputting an answer to the user input.
9. A computer-readable recording medium storing a program for executing the method according to paragraph 1.
10. Memory in which at least one program is stored; and By executing at least one of the above programs, the processor performs operations, and The above processor is, Natural language processing is performed on the received user input to produce domain and intent information regarding the user input, and Determining whether there is intent information among the stored intent information that matches the intent information calculated above, and If there is intent information that matches the intent information calculated above, an answer to the user's input is generated and output according to the script set in the scenario corresponding to the matched intent information, and If there is no intent information matching the intent information calculated above, determine whether there is a domain among multiple domains that corresponds to the domain for the user's input, and If there is a corresponding domain, based on the corresponding domain and the prompt conditions set in the corresponding domain, an answer to the user's input is generated and output through a generative AI model, and A device for implementing a method of providing answers that minimizes cases of answer errors for user input, wherein, in the case where there is no corresponding domain, the result of the natural language processing is input into the generative AI model to generate and output an answer corresponding to the user input.

Description

Method for providing answer that minimizes error response case about user's input and apparatus thereof The present invention relates to a method for providing an answer to input such as a user's question, and more specifically, to a device equipped with an artificial intelligence module that accepts user input and provides an answer, a method for minimizing the number of occurrences of answer errors, and a device for implementing the method. A chatbot refers to a bot trained to understand human questions and respond automatically by simulating human conversation using AI (Artificial Intelligence) and NLP (Natural Language Processing). More specifically, user input, whether through voice or typing on a keyboard, is inevitably human language. A Natural Language Processing Unit (NLU) performs natural language processing so that the computer can understand and process this human language. As a result of another AI model learning from the processed natural language, chatting with the user becomes possible. Recently, technology that uses such chatbots to automatically respond to customers without the need for human intervention has become commonplace. However, unlike automated response devices that employ simple automated response algorithms, chatbots can systematically learn human language and output responses accordingly, but they cannot output perfect responses to various user inputs, and this limitation is pointed out as an element that needs to be addressed. Figure 1 is a diagram illustrating an exemplary example of a user input processing process of a conventional chatbot. First, the chatbot receives the user's voice (S110). In step S110, what the chatbot receives may not be the user's voice, but rather the user's input entered via a keyboard. Next, the chatbot performs natural language processing on the received user's voice (S120). The chatbot determines whether there is a value corresponding to the natural language processing result (S130), and if there is a corresponding value and there is also a pre-set keyword in the natural language processing result (S140), it can generate and output options based on that keyword (S150). Additionally, if the chatbot has a value corresponding to the natural language processing result but there is no pre-set keyword in the natural language processing result, it can output an intent response resulting from the natural language processing result (S160). In steps S150 and S160, the chatbot can generate a response to the user's input using a built-in generative AI module. The generative AI used at this time may be any one of ChatGPT, Bard, DALL-E 2, Stable Diffusion, and Midjourney, but is not limited to the aforementioned types. In step S130, if there is no value corresponding to the natural language processing result of the user's input, the chatbot can provide a message to the user indicating that it cannot provide an appropriate answer by outputting a fallback phrase as a pre-stored phrase. Here, the fallback is classified as an error case, and if the fallback phrase is repeatedly output, the user may give up on obtaining an answer through the chatbot, which has the limitation of seriously hindering the effectiveness of automated customer service. Therefore, in an automated customer service process using a chatbot, an algorithm is required to minimize cases where a fallback occurs. Figure 1 is a diagram illustrating an exemplary example of a user input processing process of a conventional chatbot. FIG. 2 is a schematic diagram showing the entire system implementing the answer-providing method according to the present invention. FIG. 3 is a flowchart illustrating an example of a method according to the present invention. Figure 4 is a diagram illustrating the prompt conditions set for each domain in the present invention. Figure 5 is a diagram schematically illustrating the process in which a response to a user's input is generated through a domain in the present invention. FIG. 6 is a diagram schematically illustrating an example of the result of applying the present invention. FIG. 7 is a diagram schematically illustrating another example of the result of applying the present invention. FIG. 8 is a block diagram showing an example of a processor of a user answer output device implementing the method according to the present invention. The present invention is capable of various modifications and may have various embodiments; specific embodiments are illustrated in the drawings and described in detail in the detailed description. The effects and features of the present invention, and the methods for achieving them, will become clear by referring to the embodiments described below in detail together with the drawings. However, the present invention is not limited to the embodiments disclosed below but can be implemented in various forms. Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. When describi