Network Structure and its Influence on User Behavior

Social networks are now ubiquitous across many online services in diverse areas such as communication, music, education, health, and news. This thesis provides an understanding of the interplay between network structure and user behavior in the Twitter network and on the Duolingo language learning platform. We develop models and algorithms that can explain observed user behavior and distinguish between different patterns of behavior. In the context of Twitter we present a model for diffusion that captures differences across topics such as news, politics, and popular culture and examine the initial conditions required to produce large information cascades. Additionally, we devise an algorithm for accurately inferring the times at which edges are created in a large subgraph of the social network. Using these inferred edge creation times we evaluate several models of network formation, see the impact of real-world events such as news stories on network evolution, and show how user interface design choices such as ‘Suggested Followers’ lists impact users’ following behavior. Duolingo is the world’s largest language learning service and provides a unique opportunity to understand the impact of social features in an online education platform. Features such as an activity stream, leaderboard, and community sections exist to promote student-student interactions and keep students engaged in the learning process. We study the impact of these features on student behaviors and engagement over long periods of time (greater than six months). The techniques developed in this thesis can be used by creators of social systems to measure, model, and understand the impact of their design decisions.