Extracting Stylistic Features and Intentions that Create Characterization for Chatbots that Speak Based on Characterization

Abstract

In order to add characterization to chatbots, we conducted intent classification and morphological analysis of in-game dialogue from "Umamusume Pretty Derby" a game with high sales in the character business. The analysis methods used were residual analysis of the intentions used, the bigram of intentions, and a comparison of the first person and second person pronouns used for each intention. The residual analysis of intention to use was also compared with that of the general women. The results revealed that there was a significant difference in the intentions used by the characters and the general women. Characteristics such as spontaneous and positive intentions, calm speech by thinking and agreeing with others, and sweetness to others by agreeing and agreeing with others, quantitatively revealed character traits of confident, ladylike, and timid attributes. We believe that the obtained character traits can be used as prompts for chatbots to evoke the characters.