Latent variable modelling of interdependent ego-networks

Speaker: Isabella Gollini, University College Dublin

Abstract

Ego-networks consist of a focal node ("ego") and the nodes to whom ego is directly connected to ("alters"). We present a Bayesian latent variable network modelling framework for describing the connectivity structure of interdependent ego-networks (network of ego-networks) by using a latent space approach. This allows us to visualise both the ego-ego and ego-alter relational structures by estimating the positions of all the individuals in a latent "social space".  We apply this new methodology using an efficient variational algorithm able to explore the structure of human smuggling out of Libya network (operation Glauco II) consisting of ego-networks based on the wiretaps acquired by the Italian police on 29 suspects during an investigation period lasting from January to October 2014. The statistical challenge with these ego-networks is that the large number of alters (more than 15k people) can potentially be members of several ego-nets.