Detection of Landmarks for Clustering of Online-Game Players

Understanding of player behaviors is an important issue to keep online games interesting to their players. Focusing on player movement, in our previous work, we proposed a method for clustering online-game players based on the transition probabilities between manually determined landmarks in a game map of interest. In this paper, we first propose a method for automatically detecting landmarks from player trails based on weighted entropy of the distribution of visiting players. Next, we describe how to visualize player clusters using multidimensional scaling and prefuse, respectively. Their inputs are the Euclidean distances between players based on their weighted transition probabilities between the derived landmarks. The effectiveness of our approach is confirmed when it is evaluated using trail logs from both an online game developed at the first two authors’ laboratory and a commercial massively multiplayer online role-playing game (MMORPG).