Host: The Japanese Society for Artificial Intelligence
Name : The 37th Annual Conference of the Japanese Society for Artificial Intelligence
Number : 37
Location : [in Japanese]
Date : June 06, 2023 - June 09, 2023
Characters in a story may have a background history that differs from the timeline of the story, and this history may help readers to understand and sympathize with the actions of the characters. In this study, we call such a character's past history "background history." Although the recent rapid development of automatic story generation technology has made it possible to generate plots suited to each genre, few attempts have been made to automatically generate settings such as character attributes and background histories that are deeply related to the story. Therefore, in this study, we collected data on character characteristics and background histories of several highly acclaimed existing works limited to a specific genre, and extracted four types of background history patterns through factor analysis. Prompts were created for computer game scenarios according to the extracted patterns, and a large-scale language model was used to automatically generate sentences of character background histories. The automatically generated background histories were then proofread by humans and implemented in a game system.