[2F5-OS-20b] OS-20 (2)

[2F5-OS-20b-05] Role Estimation in Werewolf Game Based on Rules About Players’ Utterances

〇Daisuke Shimizu¹, Koji Hasebe¹ (1. College of Information Science, School of Informatics, University of Tsukuba)

Keywords:werewolf game, incomplete information game, AIWolf

In recent years, there have been many studies for playing imperfect information games. Among them, werewolf game has been actively studied through competitions of programs. For playing werewolf game, it is important to predict the other players' hidden roles and choose suitable actions. This study proposes a role estimation method based on rules about players' utterances in werewolf game of 5 players. Our proposed set of rules consists of those created by paying attention to the utterances of the players who claim to be a seer, and some additional statistical rules. We evaluated our method by simulations, and as a result, our method could narrow down the combination of roles correctly about 87 % of games. It also correctly answered about 77 % of the questions whether the given player is a werewolf or not.