Rule-based role estimation in werewolf games using probabilistic logic programming

The werewolf game is one of incomplete information games by multiplayer, and it is recognized widely as a new promising standard problem in Artificial Intelligence recently. In this paper, we try to evaluate the validity and effectiveness of common knowledge, i.e. common tendencies felt by a large number of players, in werewolf games. Specifically, we extract such tendencies as rules from the werewolf BBS data manually, and verify their effects on the role and team estimation tasks by using a probabilistic logic programming model having game rules and extracted tendencies. As a result, we succeeded in extracting some common tendencies for team estimation, and it was confirmed that the extracted knowledges have positive effects in some purposes such as the prediction of werewolves.