Modeling of social media behaviors using only account metadata

Abstract

Applications in Augmented Cognition can be hampered by obstacles to the effective instrumentation of the data space, making the collection of informative feature data difficult. These obstacles usually arise from technical limitations, but can also be present due to methodological and legal considerations. We address a specific instance of the difficulty of characterizing a complex behavior space under legally constrained data collection: The instrumentation of social media platforms, where privacy, policy, and marketing considerations can severely hamper 3rd-party data collection activities. This paper documents our constrained empirical analysis and characterization of the behaviors of Twitter account-holders from their account metadata alone. The characterization is performed by coding user account data as feature vectors in a low-dimensional Euclidean space, then applying parametric and non-parametric methods to the resulting empirical distribution. Suggestions for future work are offered.