Research Of Chinese Checkers Game Intelligence Based On Sequential Pattern Mining

In recent years, with the rapid development of the game industry in the international scope,the research on game intelligence gets people’s attention gradually. In this paper, according tothe simplified model of the Chinese Checkers, a game engine with the off-line study ability isdesigned and a method based on Sequential Pattern Mining is researhed, which is used toimprove the game’s intelligence. The game engine consists of game control system and dataanalysis system. The game control system is responsible for controlling game process, andgets the effective game logs by human-computer play and human-human play. The dataanalysis system is responsible for analysing and mining aforementioned game logs, and thenobtains patterns such as empirical rules, key states, falling point usage frequency, etc. Thesepatterns will be applied to the moving algorithm in the game control system for improvingthe computer-player’s moving skill.In the game control system, a moving algorithm, which is based on empirical rulesmatching, key states approximating and evaluation function computing, of computer-player isproposed. On the one hand, an unfixed coordinate system and a pieces-state-set based on theviewpoint of each player are designed respectively for locating the checkers on thechessboard and being the data structure that describes the status of the chessboard. The abovedesigning makes the moving algorithm simple and makes the game engine be able to provideeffective game logs. On the other hand, the data mining results that created by the dataanalysis system are applied to the moving algorithm so that the computer-player’s movingability can be improved.In the data analysis system, a sequential pattern mining algorithm based on pattern growthis proposed. This algorithm mine the frequent contiguous two-items subsequences, which areconverted to empirical rules soon afterwards, and count for gaining key states and fallingpoint usage frequency in the meantime. At the beginning of the mining procedure, the gamelogs are transformed into sequential data of a new format, which is easy to be mined. Next, asequential pattern tree with weighted branches is created from these sequential data.Furthermore, the sequential patterns are mined from the tree. And then, the trusted sifting isdone on the sequential patterns that have be mined. Finally, the trusted sequential patterns areconverted to empirical rules that can be recognised by computer-player.At the end, the validity of the method researched in this paper is verified throughexperiments and tests.