[4Yin2] Interactive session 2

[4Yin2-20] Simultanious MLE of skills and advantages from asymmetric plays

〇Satoru Fujii¹ (1.Graduate School of Human and Environmental Studies, Kyoto University)

Keywords:rating, maximum likelihood estimation, Bradley-Terry model

Rating systems based on the Bradley-Terry model has been widely used in many areas. These methods assume symmetry of players. However, in some environment, there are advantages on one side of the player: it can be first-move player of the game, or offense side of the sports. In this paper, we propose a method to estimate players' skills and advantage values by applying MLE to an adjusted Bradley-Terry model. We present experimental results which show this method is better than naive estimation on biased dataset.