Social networks are growing in reach and impact but little is known about their structure, dynamics, or users’ behaviors. New techniques and approaches are needed to study and understand why these networks attract users’ persistent attention, and how the networks evolve. This thesis investigates questions that arise when modeling human behavior in social networks, and its main contributions are: • an infrastructure and methodology for understanding communication on graphs; • identification and exploration of sub-communities; • metrics for identifying effective communicators in dynamic graphs; • a new definition of dynamic, reciprocal social capital and its iterative computation • a methodology to study influence in social networks in detail, using • a class hierarchy established by social capital • simulations mixed with reality across time and capital classes • various attachment strategies, e.g. via friends-of-friends or full utility optimization • a framework for answering questions such as “are these influentials accidental” • discovery of the “middle class” of social networks, which as shown with our new metrics and simulations is the real influential in many processes Our methods have already lead to the discovery of “mind economies” within Twitter, where interactions are designed to increase ratings as well as promoting topics of interest and whole subgroups. Reciprocal social capital metrics identify the “middle class” of Twitter which does most of the “long-term” talking, carrying the bulk of the system-sustaining conversations. We show that this middle class wields the most of the actual influence we should care about — these are not “accidental influentials.” Our approach is of interest to computer scientists, social scientists, economists, marketers, recruiters, and social media builders who want to find and present new ways of exploring, browsing, analyzing, and sustaining online social networks.