Drawing on percolation theory from physics, we assess why some diffusion trajectories produce social big bangs (i.e., high percolation efficiency), while others generate highly fragmented social networks (i.e. low percolation efficiency). Specifically ...
Drawing on percolation theory from physics, we assess why some diffusion trajectories produce social big bangs (i.e., high percolation efficiency), while others generate highly fragmented social networks (i.e. low percolation efficiency). Specifically, we examine how different characteristics of nodes (i.e., users in an online social network, e.g., influential, hidden influential, broadcasters, and peripherals) and links (e.g., interactions among the users, e.g., incoming/outgoing, bridging, overlap) moderate percolation efficiency. Moreover, we identify different effects of each node and link types on the percolation efficiency for social and business issues.
For the empirical analysis in social network environments, we first select 19 and 8 most popular issues in business and social contexts respectively. Then, Twitter data for the 27 selected issues was provided by SocialMetrics which is a social trend exploration service operated by Daumsoft. Using the data, we construct networks for each issue on each day on the basis of each user’s retweet behavior. Then we model percolation efficiencies of networks as a function of their node and link characteristics. To operationalize the measurements of percolation efficiency and identify the effects of different node and link types, the two-stage data envelopment analysis (DEA) approach is adopted. In addition, we conduct subsample analyses to identify the differences between social and business issues. We analyze 1,052,619 retweet messages to measure the variation of percolation efficiencies across networks. The key findings suggest that (1) percolation varies significantly on message-based networks, for example, social and business threads; (2) bridge and incoming/outgoing links increase the efficiency of percolation, but overlap and adding links have no observable effect; (3) peripherals play a more significant role in percolation than influentials; (4) the moderating effects of links and nodes are more pronounced for business threads than social threads. On the basis of a new theoretical perspective, this study extends research on information diffusion and word-of-mouth and provide stylized insights into effective social media targeting, advertising and public opinion formation. The research was presented and discussed at various seminars at Korea University and KAIST Business school. It was also presented at International Conference on Internet Studies (NETs) which was held in Osaka, Japan from July 22 to 24, 2016.