Automatic Generation of Baseball Game Digest by Using Win Probability Added

Abstract

　Digest sentences of baseball game are written manually nowadays. These sentences are written by many writers with different perspectives and contained professional expression about play. There are some automatic generation researches which conveying about digest of baseball game, but there are insufficient to pick up a plate appearance where the situation changed most during the game. There are the methods which measure a player’s contribution to a win using change of win expectancy among that for statistically analyzing baseball game. In this paper, automatic generation of digest sentences are used this change of winning probability. Following the analysis of existing digest documents, each sentence is into template and words are replaced to variables. A program that applies input play-by play and box scores to variables is created. Texts on outline part of victory which are made automatically are including same contents in comparison with sentences which are written manually. Sentences which are written automatically about a batter who made an active part are based on objective viewpoint by using winning probability. In the future, the program needs to make improvement about description for effective batters. Furthermore, it is necessary to describe about pitchers or why the team is lost.