Extracting Branch Factors of Scenarios from a Gaming Simulation Using Log-Cluster Analysis

Abstract

This study proposes a method for analyzing gaming simulation results. The gaming simulation we focus on intends to be played by both human and computer agent players. To extract the knowledge of what and how they have played, we must determine what type of decision-making process leads to specific scenarios. Such simulation results, however, tend to have so many branch factors of scenarios that it is hard to understand by manual operations. To deal with the issues, we have developed a method for obtaining the branch factors of scenarios from gaming simulation results. We have demonstrated the effectiveness of the proposed method by identifying the branching factors of scenarios as follows. First, software agents were asked to play a gaming simulation for career education. Next, logs acquired through gaming were classified into multiple scenarios using machine learning techniques. Finally, decision-making factors separating the scenarios were identified using a decision tree.