Game difficulty prediction algorithm based on improved Monte Carlo tree

In the unreleased stage of the game, how to correctly set the game difficulty is an important task to user’s feeling. Whether the difficulty of the game can be accurately measured will directly affect whether the difficulty setting is reasonable. As a commonly used stochastic simulation algorithm, Monte Carlo tree Search(MCTS) has been widely used in the field of game simulation. However, the traditional MCTS does not consider the specific situation of the game during the simulation, resulting in a low simulation success rate.In this paper, an improved MCTS is proposed to address above problems, and the UCB formula used for node selection is improved. The distance factor is included in the consideration of the process of Monte Carlo tree select node in the maze game. After compared the simulation success rate with the unimproved UCB, and tested the influence of the number of obstacles and the distance of path-finding on the performance of the algorithm, it is proved that in the case of less and shorter path-finding distance, the simulation success rate is greatly improved compared to the previous algorithm.