Multi-level Analysis on Structures and Dynamics of Online Social Networks

Abstract

The `electronic footprints` [1] such as e-mails [2], photos [3], or phone call logs [4] that reveal much of people`s thinking process and actions have become the new sources of human activity recording in the last decade [5]. The online social networking service is the largest store that captures human activities.

Understanding structures and dynamics of online social networks, thus, is important for two reasons at least. One is because we discover human nature in the online social network, as it is a miniature of our society at a variety scale. The other is because the online social network itself is a virtual world where various phenomena emerge, such as gossip propagation, money exchange, and reputation management. Nevertheless, structures and dynamics thereof are not so simple that we cannot clearly illustrate the whole process.

This dissertation tackles the complex structures and dynamics of online social networks by multilevel approach focusing on different entities from macroscopic view to microscopic view. We zoomed in on the online social networks from an entire network to a community, a dyad, and an individual. Each level has its own inherent, but complementary, resolution to capture elemental processes and guides us to understand the complex structures and dynamics as a combination of findings across multiple levels. We begin with the entire network structures and dynamics in Twitter. We focus on the networkwide structural properties and the information diffusion process therein. Through large-scale Twitter data, we analyze the topology of the social graph, identify influentials, characterize temporal behavior of trending topics, and present the impact of retweets with regard to the number of audience. We also propose a novel method to find influentials by considering both the link structure and the temporal order of information adoption in Twitter. Then, we move on the community structure as the mesoscale units of the network structure. We quantif...