Macrocosm: Social Media Persona Linking for Open Source Intelligence Applications

Abstract

Online Social Networks (OSNs) provide a wealth of intelligence to analysts in assisting tasks such as tracking cyber-attacks, human trafficking activities, and misinformation campaigns. Open Source Intelligence (OSINT) analysts monitoring social media typically track users of interest manually, spending hours or days linking personas of interest within and across OSNs. This paper presents a multi-modal analysis of cross-contextual online social media (Macrocosm), a data-driven approach to detect similarities among user personas over six modalities: usernames, patterns-of-life, stylometry, semantic content, image content, and social network associations. It fuses component modalities into an ensemble similarity judgment. To the best of our knowledge, Macrocosm is the first research effort to apply Siamese neural networks to the persona linking problem. An important lesson is that SNNs---deep learning models that infer a distance function from high-dimensional data---consistently provide improvements over traditional models in testing.

Tasks

Reproductions