A Social Contagion: An Empirical Study of Information Spread on Digg and Twitter Follower Graphs

Social scientists have long recognized the importance of social networks in the spread of information [Granovetter 1973], products [Brown and Reingen 1987; Watts and Dodds 2007], and innovation [Rogers 2003]. Modern communications technologies, notably email and more recently social media, have only enhanced the role of networks in marketing [Domingos and Richardson 2001; Kempe et al. 2003], information dissemination [Wu et al. 2004a; Gruhl and Liben-nowell 2004], search [Adamic and Adar 2005], and expertise discovery [Davitz et al. 2007]. The recent DARPA Network Challenge1 successfully tested the ability of online social networks to mobilize massive ad-hoc teams to solve real-world problems, which could potentially improve disaster response and coordination of relief efforts. In addition to making social networks ubiquitous, the Web has given researchers access to massive quantities of data for empirical analysis. These data sets offer a rich source of evidence for studying the structure of social networks [Cha et al. 2010] and the dynamics of individual [Vázquez et al. 2006] and group behavior [Hogg and Lerman 2009], efficacy of viral product recommendation [Leskovec et al. 2006], global properties of the spread of email messages [Wu et al. 2004a; Liben-Nowell and Kleinberg 2008] and blog posts [Gruhl and Liben-nowell 2004], and identification of influentials [Leskovec et al. 2007; Ghosh and Lerman 2010; Bakshy et al. 2011]. In most of these studies, however, the structure of the underlying network was not visible but had to be inferred from the flow of information from one individual to another. This posed a serious challenge to our efforts to understand how the structure of the network affects social dynamics and information spread. Social media sites Digg and Twitter offer a unique opportunity to study social dynamics on networks. Both sites have become important sources of timely information for people. The social news aggregator Digg allows users to submit links to news stories and vote on stories submitted by other users. On Twitter users tweet short text messages, that often contain links to news stories or retweet messages of others. Both sites allow users to link to others whose activity (i.e., votes and tweets) they want to follow. Both sites provide programmatic access both to data about user activity and social networks. This rich, dynamic data allows us to ask new questions about information spread on networks. How far and how fast does information spread? How deeply and how widely does it penetrate? How do people respond to new information? How does network structure affect information spread? Do some network topologies accelerate or inhibit information spread? We address some of these questions through a large scale empirical study of the spread of information on Digg and Twitter. For our study we collected activity data from these websites. As described in Section 2, the Digg data set contains all popular stories submitted to Digg over a period of a month, and who voted for these stories and when. Twitter data set contains tweets with embedded URLs posted over a period of three weeks. We use URLs as markers for how information diffuses through Twitter. In addition, we extracted the follower graphs of active users on these sites. These data sets allow us to empirically characterize individual and collective dynamics (Section 3) and trace the flow of information on the network (Section 4). We measure global properties of information flow on the two sites and compare them to each other. In addition to using standard measure such as size, depth and breadth of spread, we define a new metric that characterizes how closely knit the network is through which information is spreading. We find that while characteristics of