Clustering Player Paths

Player traces can result in a vast amount of high-dimensional data. This information is crucial to game designers trying to understand player behaviour and for tailoring the game experience. In this work we investigate different approaches to clustering player traces in order to expose useful game information. We consider three trace similarity metrics, as well as two clustering algorithms. We evaluate and compare the different approaches using trivial and non-trivial game levels from different game genres, giving guidance on clustering and metric choices, and illustrating the potential value of cluster information.