Suffix Tree Index Structure on Go Game Records

抄録

Computer Go is important in the computer game field. The state-of-the-art Go programs contain tactical look-ahead, pattern-matching, evaluation function and highly selective global search. One issue in computer Go is to analyze a database constructed from game records collection and to discover patterns contained in it. To acquire sequence patterns from game records, the methods used in natural language processing and statistics were applied by regarding each game record as a text string and suitably encoding each move to a character. Furthermore, the way to look for a specific pattern from the huge database efficiently is critical to computer Go programs. Both search depth and branching factor of computer Go are much larger than those of other chess-like games, due to the bigger board size and more available moves. Therefore, pattern matching in computer Go requires more computation resource. In this paper, we propose a mechanism which applies the suffix tree algorithm to build an efficient data structure with indexing function for numerous game records. Besides, we also show the position and occurrence times in the game records of Go. The experimental results show that the search time for patterns of Go is speeded up.