a first input step of receiving an input of a user utterance;

a step of generating a conversation content by using a conversation model based on an inputted user utterance content;

a step of converting the generated conversation content into a voice and outputting the voice;

a first recognition step of recognizing emotion of the user by using an emotion analysis model based on the user utterance content while performing the first input step, the generation step, and the output step; and

a first monitoring step of monitoring the recognized user emotion,

wherein the conversation model is a neural network that generates a conversation content for responding to an utterance content when a user utterance content is inputted, and outputs the conversation content,

wherein the emotion analysis model is a neural network that recognizes emotion of a user from an utterance content when a user utterance content is inputted, and outputs the emotion of the user,

wherein the step of generating comprises: a step of determining one of a first conversation model and a second conversation model based on an inputted user utterance content; and a step of generating a conversation content by using the determined conversation model,

wherein the first conversation model is a neural network that receives a user utterance content and knowledge related to the utterance content, and generates a response from the knowledge related to the user utterance content and outputs the response,

wherein the second conversation model is a neural network that receives only a user utterance content and generates a response from the inputted user utterance content and outputs the response,

wherein the first recognition step is performed when the second conversation model is determined at the determination step, but is not performed when the first conversation model is determined at the determination step.