Identifying dynamics and collective behaviors in microblogging traces

Microblogging disseminates realtime information through dynamic user interactions. While it is intuitive that such interactions may generate patterns, it is difficult to identify and characterize them in satisfactory detail. In this paper, we propose using a combination of dynamic graphs and time-series to study the dynamics and collective behaviors in microblogging. To enable automatic pattern identification, a distance metric is developed to incorporate the heterogeneous aspects of the dynamical interactions. We demonstrate the effectiveness of the proposed approach using a month long Twitter dataset and show that the new representation and distance metric are both essential for discovering the patterns of collective microblogging, such as propagation of breaking news, advertisement, social movement, and interest group formation.