Opponent modeling by analysing play

Abstra t. Opponent modeling is useful for a player both to have a better han e to win and for tea hing a novi e player. In this paper we dis uss opponent modeling by analysing the play of the opponent. We view this as two dual tasks: hara terising the play of a parti ular opponent and re ognising an unknown opponent. We learn understandable opponent models with a logi al de ision tree learning system. We illustrate our approa h by an experiment in the game of go and dis uss the appli ation of similar ideas in other games.