A Comparison of Methods for Player Clustering via Behavioral Telemetry

The analysis of user behavior in digital games has been aided by theintroduction of user telemetry in game development, which providesunprecedented access to quantitative data on user behavior from theinstalled game clients of the entire population of players. Playerbehavior telemetry datasets can be exceptionally complex, withfeatures recorded for a varying population of users over a temporalsegment that can reach years in duration. Categorization of behaviors,whether through descriptive methods (e.g. segmentation) orunsupervised/supervised learning techniques, is valuable for findingpatterns in the behavioral data, and developing profiles that areactionable to game developers. There are numerous methods forunsupervised clustering of user behavior, e.g. k-means/c-means, NonnegativeMatrix Factorization, or Principal Component Analysis.Although all yield behavior categorizations, interpretation of theresulting categories in terms of actual play behavior can be difficult ifnot impossible. In this paper, a range of unsupervised techniques areapplied together with Archetypal Analysis to develop behavioralclusters from playtime data of 70,014 World of Warcraft players,covering a five year interval. The techniques are evaluated withrespect to their ability to develop actionable behavioral profiles fromthe dataset