Mining Balanced Patterns in Real-Time Strategy Games
Guillaume Bosc, Mehdi Kaytoue, Chedy Ra, Philip Tan · 2014
Given a set of class-labeled objects taking descriptions from a partially ordered set (e.g. itemsets, graphs, intervals, etc.), the task of finding description general-izations that strongly cover a class and weakly the others, has attracted a lot of attention in artificial intelligence (machine learning and data-mining) under vari-ous names (hypothesis, contrast sets, subgroups, emerging/jumping patterns, etc.). We propose in this paper to discover sequential patterns that discriminate victory from zero-sum games. We present efficient algorithms and show how emerging pattern mining may be used to balance games (an important issue in the game in-dustry) by using it to analyze strategies with a novel measure, the balance measure. We experiment with the designed algorithms on a real-world strategy game played professionally as an electronic sport. 1