Regensburg 2025 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 26: Networks, From Topology to Dynamics (joint session SOE/BP/DY)
DY 26.6: Talk
Wednesday, March 19, 2025, 16:30–16:45, H45
Modelling retweet cascades using multivariate Hawkes processes on sparse networks — Alexander Kreiß1 and •Eckehard Olbrich2 — 1Leipzig University, Germany — 2Max Planck Institute for Mathematics in the Sciences, Leipzig, Germany
We apply a model that considers vertices in a network who are able to cast events, e.g. users of the online social media platform Twitter. Furthermore, there is a directed edge from vertex A to vertex B if A takes note of the events cast by B and changes its own behavior accordingly. More precisely, the model assumes that the activity of B increases the activity of A and likewise its other neighbors. This is called peer effects. However, there might also be other information, which also influences the activity of the vertices, e.g. the time of the day for social media posts. This is called global effects. We use a Hawkes model that incorporates, both, peer and global effects. This allows for the estimation of the network, that is, the influence structure while controlling for network effects or the estimation of the global effects while controlling for peer effects. The estimation is based on a LASSO strategy, which respects sparsity in the network. We apply this model to retweets on Twitter in order to reconstruct potential retweet cascades and identify accounts that are influential in sharing information.
Keywords: Hawkes process; networks; social media; information sharing; high-dimensional methods