Irregularities in the Dynamics of Large Networks: Cases on Social Networks and the Internet

Dynamic networks capture evolving relationships among entities, and irregularities in their behavior often signal significant events. This dissertation investigates such irregularities across social and infrastructural networks. First, we analyze Twitter (prior to its rebranding to X) to characterize influencers, introducing a follower-overlap similarity measure that identifies micro-influencers beyond content-based methods. We then examine election-related irregularities by comparing suspended and non-suspended accounts during the 2020 U.S. presidential election, revealing behavioral patterns linked to misinformation and coordinated manipulation. Next, we develop an algorithm to infer post-specific diffusion networks, reconstructing who-saw-from-whom pathways for tweets at scale. This method uncovers structural differences in diffusion and highlights concentrated bot activity in certain cascades. Finally, using large-scale Russian network measurements, we expose censorship-related irregularities, including elevated TCP connection delays during geopolitical events. Together, these studies demonstrate how irregularities in dynamic networks illuminate influence, manipulation, diffusion, and state-imposed interference.