Group Dynamics Modeling, Forecasting, and Validation

Building on our prior research in group discovery and tracking, we propose to develop new algorithms for predicting future behaviors of social media groups. These algorithms will be based on emerging social interaction theories and new computational models of group dynamics at multiple levels, from individual roles and interactions to aggregate trends within and across groups. Our approach takes advantage of latest advances in Bayesian mixture modeling to infer latent structure at every level such that discovered structure at one level serves as a reduced representation for the next. In addition to aggregate-level structure inference, we will extend theoretical models of temporal group behavior via an integer programming relaxation of graph-based optimization. We will assess the accuracy of our algorithms and refine our group dynamics models.