Anonymous Yet Alike: A Privacy-Preserving DeepProfile Clustering for Mobile Usage Patterns

The ubiquity of mobile devices and unprecedented use of mobile apps have catalyzed the need for an intelligent understanding of user’s digital and physical footprints. The complexity of their inter-connected relationship has contributed to a sparsity of works on multi-contextual clustering of mobile users based on their digital and physical patterns. Moreover, with personalization the norm in users’ lives and corporations collecting a multitude of sensitive data, it is increasingly important to profile users effectively while preserving their privacy. In this paper, we propose DeepProfile: a Multi-context Mobile Usage Patterns Framework for predicting contextually-aware clusters of mobile users and transition of clusters throughout time, based on their behaviors in three contexts - app usage, temporal and geo-spatial. Our DeepProfile framework preserves users’ privacy as it intelligently clusters their mobile usage patterns and their transition behaviors while maintaining users’ anonymity (i.e., without their gender, GPS location and high-level granularity application usage data). Our experimental results on a mobile app usage dataset show that the predicted user clusters have distinct characteristics in app usage, visited locations and behavioral characteristics over time. We found that on average, 18.6% to 23.6% of a cluster moves together to the next time segment, and other interesting insights such as over 90% of cluster transitions where users moved together, moved from a period of activity to inactivity at the same time.