Analysts studying relationships between people often rely on information that is inadequate or inaccurate. Collecting information to map out a social network may not be immediately possible, can overlook certain connections between individuals, and may even record ties incorrectly. These problems are particularly problematic when studying networks representing the veiled activities of covert groups like troll farms and terrorist organizations, since they intentionally try to hide their behavior. This thesis develops methods for understanding underlying network activity from inadequate or inaccurate information by detecting possible anomalies in online account creations, identifying missing or spurious connections in observed social networks, and measuring the uncertainty in metrics calculated on imperfect network data. Evaluations using real and simulated networks demonstrate that these techniques can provide early warnings of online influence campaigns, emend complex network data that vary over time, and put bounds on the relative importance of covert actors. This thesis then applies these approaches to a sample of social media messages coupled with more complete records of the authors’ activities in order to explore how these techniques can improve the reliability of subsequent analysis. Ultimately, this research provides guidance on how network scientists can adjust their analysis to cope with missing or spurious data, such as when social media analysts study concerning online activity. Finally, this thesis highlights how law enforcement and government intelligence agencies can improve their analysis of social networks representing clandestine behavior.

274 pages

Thesis Committee:
Kathleen M. Carley (Chair)
Hirokazu Shirado
Nynke M. D. Niezink
Gian Maria Campedelli (Foundazione Bruno Kessler)

Nicolas Christin, Head, Software and Societal Systems Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu