Skip to Main Content
Knowledge modeling, acquisition, and proper use are crucial for successful computer playing of the game of GO. This paper presents a systematic approach for knowledge representations of GO-playing in which the meaning of a move is defined as a contextual pattern with respect to local contexts of surroundings of the move. This representation allows large amounts of contextual patterns, along with their usage statistics, to be acquired efficiently from game records. Contextual patterns that appear jointly in local contexts may form a pattern collocation, with its significance measured by hypothesis testing. Our experiments show that GO-playing knowledge acquired in the form of contextual patterns and pattern collocations is effective in move generation at various stages of GO-playing.