Computational Lens on Big Social and Information Networks

The connections between individuals form the structural backbone of human societies, which manifest as networks. In a network sense, individuals matter in the ways in which their unique demographic attributes and diverse interactions activate the emergence of new phenomena at larger, societal levels. Accordingly, this thesis develops computational models to investigating the ways that individuals are embedded in and interact within a wide range of over one hundred big networks---the biggest with over 60 million nodes and 1.8 billion edges---with an emphasis on two fundamental and interconnected directions: user demographics and network diversity.

Work in this thesis in the direction of demographics unveils the social strategies that are used to satisfy human social needs evolve across the lifespan, examines how males and females build and maintain similar or dissimilar social circles, and reveals how classical social theories---such as weak/strong ties, social balance, and small worlds---are influenced in the context of digitally recorded big networks coupled with socio-demographics. Our work on demographics also develops scalable graphical models that are capable of incorporating structured discoveries (features), facilitating conventional data mining tasks in networks. Work in this part demonstrates the predictability of user demographic attributes from networked systems, enabling the potential for precision marketing and business intelligence in social networking services. Work in this thesis in the direction of diversity examines how the diverse structures of common neighborhood influence link formation locally and network organization globally, how this influence varies across different types of social and information networks, and how it concords or conflicts with the principle of homophily. Work in this direction reveals how topic diversity---in contrast to authority and popularity---drives the growth of impact in academic collaboration and citation networks as well. Finally, our work on diversity presents neural network based representation learning models for embedding heterogeneous networks in which there exist diverse types of nodes and edges, giving rise to important implications for traditional mining and learning tasks in heterogeneous network data, including similarity search, clustering, and classification.