Understanding Emerging Social Structures — A Group Profiling Approach Lei Tang Computer Science & Engineering Arizona State University Tempe, AZ 85287, USA L.Tang@asu.edu Xufei Wang Computer Science & Engineering Arizona State University Tempe, AZ 85287, USA Xufei.Wang@asu.edu Huan Liu Computer Science & Engineering Arizona State University Tempe, AZ 85287, USA Huan.Liu@asu.edu ABSTRACT The prolific use of participating web and social networking sites is reshaping the way in which people interact with each other, and it has become an increasingly vital part of social life of human beings. People sharing certain similarities tend to form communities in social media. At the same time, they participate in various online activities: content sharing, tag- ging, twittering, etc. These diverse activities leave traces of their social life, providing clues to understand emerging so- cial structures. Plenty of existing work focus on extracting cohesive groups based on network topology. In this work, we advance further to explore different group-profiling strate- gies to construct descriptions of a group, helping explain the group formation. This research and results can help network navigation, visualization and analysis, as well as monitoring and tracking the ebbs and tides of different groups in evolv- ing networks. By exploiting the information collected from real-world social media sites, we conduct extensive experi- ments to evaluate group-profiling results. The pros and cons of different group-profiling strategies are analyzed with con- crete examples. We then use LiveJournal as a testbed to show some potential applications based on group profiling. Interesting findings with discussions are reported. Keywords Group Profiling, Social Structures, Group Characterization, Group Formation, Social Media 1. INTRODUCTION Recently, a surge of work has reported the statistical pat- terns presented in complex networks across many domains [23, 7]. The majority work study the global patterns presented in a static or an evolving network [15, 17]. Microscopic pat- terns such as the individual interaction patterns are also at- tracting increasing attentions [16]. This work, alternatively, focuses on meso-level analysis of a network as in Figure 1. In particular, we study groups (communities) in social me- dia. Group-level analysis plays a key role in social science. Actually, “the founders of sociology claimed that the causes of social phenomena were to be found by studying groups rather than individuals”([13] Chapter 2, Page 15). A group (or community) can be considered as a set of ac- tors where each actor interacts with the other actors within the community more frequently than with those actors out- side the community [31]. Finding out groups from network interactions has a broad range of applications, including net-    Figure 1: Network Analysis at Different Levels work visualization, intelligence analysis [4], network com- pression [26], behavioral study [13], and collaborative fil- tering [9]. A variety of community detection (a.k.a. find- ing cohesive subgroups [31]) methods have been proposed to capture such social structures in a network. With the ex- panded use of Web and availability of large-scale social net- works, community evolution in dynamic networks is gaining increasing attentions [14, 24, 26, 29]. While a large body of work has been devoted to discover groups based on network topology, few systematically delve into the extracted groups to understand the formation of a group. some fundamental questions remain unaddressed: What is the particular reason that binds the group members together? How to interpret and under- stand a social structure emanated from a net- work? Some pioneering work attempt to understand the group formation based on statistical structural analysis. Back- strom et al. [3] studied prominent online groups in the digi- tal domain, aiming at answering some basic questions about the evolution of groups, one of which is what are the struc- tural features that influence whether individuals will join communities. They found that the number of friends in a group is the most important factor to determine whether a new actor would join the group. This result is interesting, though not surprising. It provides a global level of structural analysis to help understand how communities attract new actors. Leskovec [18] observed that spectral clustering (a popular method used for community detection) always finds tight and small-scale but almost trivial communities, i.e., the community is connecting to the remaining network via one single edge. Both work above present a global (statistical) picture of communities. However, more efforts are required to understand the formation of a particular group. In social media, people tend to interact with each other if they share certain similarity (also known as homophily [20]), resulting in assorted communities. There can be various rea- sons leading to the formation of a community. some users interact with each other because they attend the same uni- versity; some actors form a group as they are enrolled in an