Behavior Analysis of Executed and Attacked Players in Werewolf Game by ILP

Recently, the Werewolf game, one of multiplayer incomplete information games, is recognized as a promising research testbed for artificial intelligence. In this work-in-progress paper, in order to obtain useful knowledge on the Werewolf games, log analyses of human players in the Werewolf BBS are conducted. By applying inductive logic programming to the log data of six games in the BBS, we attempt to extract classification rules on being attacked, executed and voted, whose bodies compose predicates representing players’ past behaviors. In preliminary experiments, classification rules capturing characteristic behaviors among multiple agents were successfully obtained.