Previous chapters focused on the models of static networks, which consider a relational network at a given point in time. However, real-world social networks are dynamic in nature; for example, friends of friends become friends. Social network research has, in recent years, paid increasing attention to dynamic and longitudinal network analysis in order to understand network evolution, belief formation, friendship formation, and so on. This chapter focuses mainly on the dynamics and evolutional patterns of social networks. The chapter introduces real-world applications and reviews major theories and models of dynamic network mining.
TopData Collection For Dynamic Networks
The data for dynamic network mining have usually been collected by means of questionnaires, observations, self-reporting, and simulations. For example, Nordle and Newcomb housed together 17 University of Michigan students who were initially unknown to each other. Each person was asked to rank each of his fraternity members with regard to positive feeling. Rankings were recorded each week and continued for a period of 15 weeks (Newcomb, 1961; Nordle, 1958). Some studies generate small sets of synthetic data to validate their dynamic frameworks (Lin, Chi, Zhu, Sundaram, & Tseng, 2010; Tantipathananandh, Berger-Wolf, & Kempe, 2007). With the increasing accessibility of digitized data through electronic databases and the Internet, researchers have collected longitudinal social network data on a large scale from e-mail interaction records (Carley & Skillicorn, 2005; Freeman, 1979), Digital Bibliography and Library Project (DBLP) citation data (Huang, Zhuang, Li, & Giles, 2008), protein interaction records (Ratmann, Wiuf, & Pinney 2009; Wagner, 2001), online social networks (Falkowski, 2009; Kumar, Novak, & Tomkins, 2006), and Web blogs (Lin et al., 2010). Furthermore, some studies attempt to automatically collect dynamic networks for given entities from news articles or from the entire Web through natural language processing (NLP) and machine learning (ML) techniques (Bernstein, Clearwater, Hill, Perlich, & Provost, 2002; Hu, Xu, Shen, & Fukushima, 2009; Ma, Pant, & Sheng, 2009, Tetlock, Saar-Tsechansky, & Macskassy, 2008). For example, Bernstein et al. use name co-occurrence frequency to calculate relational strength between companies, and extract inter-company network from public news; Ma et al. observed that a company is more likely to co-occur with its competitors on Web pages than with noncompetitors; Hu et al. use publishing time in news articles to extract temporal company networks from the Web. Other description of the sample data of dynamic social networks can be found in Goldenberg, Zhang, Fienberg, and Airoldi (2009).