Social gaze fingerprints: identifying social virtual reality users by their eye gaze patterns

Abstract Previous research demonstrated the possibility for implicit user identification in virtual reality (VR) based on motion data, but little is known about user identifiability during interactions in Social VR where users have no behavioral task and may show comparatively little head and body movements. Here we used Siamese Convolutional Neural Networks (CNNs) to test user identification based on eye gaze patterns during conversations in 128 participants. Combining separately trained models for gaze while speaking and gaze while listening to a partner allowed for a 95.50% identification accuracy. Gaze patterns were only weakly associated with self-reported personal characteristics and likely represent idiosyncratic behavior. The possibility to identify users in Social VR merely along eye gaze behavior in dyadic interactions substantiates concerns regarding data privacy and security in such environments. The technique can be applied to gaze data as they are commonly received from other users within Social VR environments where sample rate may be lower and less stable compared to data obtained in research laboratories. Future research is needed to assess how identification procedures generalize across a wider range of settings. Graphical abstract