Understanding players' identities and behavioral archetypes from avatar customization data

Virtual identities are an integral part of peoples' lives, from online shopping accounts to social networking profiles, from intelligent tutors to videogame avatars. In many videogames, players construct avatars to represent themselves within virtual environments and research has shown that players' sociocultural identities influence their avatar construction and can be a proxy for inferring their values in the non-virtual (real) world. In this paper, we present a computational approach to modeling players' real-world identities using behavioral data collected during the avatar customization process. We used archetypal analysis on player interaction data to develop “behavioral archetypes”, which are models of prototypical behavior patterns exhibited by players during the avatar customization process. We modeled patterns of (1) “avatar gender-preferring” behaviors (preferences for a particular avatar gender), (2) “styler” behaviors (preferences for different parts of their avatars, e.g., hair-styler, head-styler, etc.,) and (3) preferences for using avatars of a different gender (“gender-bending”) or the same gender (“gender-synchronizing”) as the players'. In a user-study with 190 participants, the behavioral archetype model trained via supervised learning had high accuracy (81%) in classifying players' real-world gender using only behavioral data. We show that behavioral archetypes are effective for understanding players in terms of their customization behaviors, real-world genders, and virtual avatar genders.