Spectral Evolution of Twitter Mention Networks

This papers applies the spectral evolution model presented in [5] to networks of mentions between Twitter users who identified messages with the most popular political hashtags in Colombia (during the period which concludes the disarmament of the Revolutionary Armed Forces of Colombia). The model characterizes the dynamics of each mention network (i.e., how new edges are established) in terms of the eigen decomposition of its adjacency matrix. It assumes that as new edges are established the eigenvalues change, while the eigenvectors remain constant. The goal of our work is to evaluate various link prediction methods that underlie the spectral evolution model. In particular, we consider prediction methods based on graph kernels and a learning algorithm that tries to estimate the trajectories of the spectrum. Our results show that the learning algorithm tends to outperform the kernel methods at predicting the formation of new edges.