Quantifying attractiveness of incomplete-information multi-player game: case study using DouDiZhu

Abstract

This paper explores the nature of DouDiZhu which is an incomplete-information multi-player game and the most popular card game in China. While there are many DouDiZhu-like card games in the world, standard DouDiZhu card game is particularly characterized by its harmonic combination of several factors including the number of players, player’s roles such as landlord and peasant, the number of decks, and score system to determine the winner. However, a study on such factors in relation to DouDiZhu game sophistication and attractiveness is not yet established. As such, this paper conducts such study by building the artificial intelligence (AI) player of DouDiZhu game with different levels, conducting simulations of self-play among the AI players, and employing the game refinement measure to assess and identify the attractive settingsof this game. The results obtained show that standard DouDiZhu has provided the most sophisticated game setting for various AI players,which is analogous to actual human players of various levels.