B1663
Title: A joint mixture model for the analysis of heterogeneous clusters in the alter group in interconnected ego-networks
Authors: Isabella Gollini - University College Dublin (Ireland) [presenting]
Abstract: Ego-networks are a particular type of network structure in which both nodes and links are collected from the perspective of a single node, called ego. The collection of nodes specified by the egos is called the alter group. The ego-networks are interconnected if there are links between the egos. A new latent variable approach is proposed in order to analyse the uncertainty associated with the presence of heterogeneous clusters in the altered group in interconnected ego-networks. A joint mixture model is introduced to describe that variability in a very natural way by taking into account the dependence within the ego-group. From a computational point of view, an efficient variational algorithm is implemented to overcome the issue of estimating the intractable likelihood function of the model. This new methodology is illustrated by exploring a complex network based on the wiretaps acquired by the Italian police during an investigation into human smuggling out of Libya.