Understanding and Reshaping Social Networks with Advanced Computational Techniques

Abstract

Social networks are powerful in modeling interdependence among individuals. Recently, the availability of large-scale social network data and advances in computational tools have facilitated the rapid development in social network research. However, a few important aspects of social networks have been understudied, and advanced computational tools may not directly help social scientists draw scientific knowledge. My thesis thus aims to move towards applying and developing computational tools that help investigate important questions on social networks.

The first component of my thesis focuses on understanding social interactions and networks, which offers implications for reshaping social networks to improve social cohesion. Specifically, I examine the formation and dynamics of social networks, with a focus on social exchange and "long ties." Utilizing large-scale social network data and computational tools, I first discuss benefits of the social exchange with dissimilar people in social networks; and then I proceed to study dynamic social networks and focus on long ties, or the social ties that bridge different communities in dynamic networks. Methodologically, I develop a novel interdisciplinary approach that combines game theory and machine learning techniques.

Second, I study what features on online platforms may improve social interactions and reshape social networks. To do so, I utilize large-scale data of online social media and provide two examples in the field. The first example is the identification of social contagion of online gift giving. This study examines how receiving a gift would promote the person to pay forward the gift, and also discusses how this social contagion can promote social interactions and tight social bonds. The other example is to examine how the designs of peer effects and prosociality on online social platforms encourage users' offline fitness behavior. Methodologically, both studies involve advanced causal inference and machine learning techniques to test the main hypotheses.

Moreover, I develop computational tools that analyze social network data. In the final component of my thesis, I introduce an algorithm for controlled experiments in social networks. This algorithm detects heterogeneous spillover effects -- how the treatment assignments received by one's network neighbors affect a person's behavior -- in the data of networked experiments. This interdisciplinary algorithm combines approaches in causal inference, machine learning, and network science.