Applying data mining to extract design patterns from Unreal Tournament levels

We present a study on the application of data mining to extract design patterns from Unreal Tournament III levels used in the online gaming scene for Duel and Team Deathmatch games. The maps' topological structure and their morphology was extracted using ad hoc bots we developed and several statistics have been computed using typical graph algorithms. The process resulted in datasets containing information about all the relevant positions in the maps (the nodes in the game navigation mesh used in the Unreal game engine) and their role in the game (i.e., whether they are navigation points, ammo pickups, weapon pickups, or powerup pickups). We have applied four data mining algorithms to the data to characterize both (i) the maps' type (Duel and Team Deathmatch) based on the feature of the nodes they contain and (ii) the node types (ammo, weapon, powerup, or navigation) based on their features. Our results suggest that the maps' type can be characterized in terms of the nodes they contain but it is difficult to characterize the role of nodes based solely on their features.