Latent space models for multidimensional network data

Network data are any relational data recorded among a group of individuals, the nodes. When multiple relations are recorded among the same set of nodes, a more complex object arises, which we refer to as “multidimensional network”, or “multiplex”, where different relations corresponding to different networks. In the past, statistical analysis of networks has mainly focused on single-relation network data, referring to a single relation of interest. Only in recent years statistical models specifically tailored for multiplex data begun to be developed. In this context, only a few works have been introduced in the literature with the aim at extending the latent space modeling framework to multiplex data. Such framework postulates that nodes may be characterized by latent positions in a p-dimensional Euclidean space and that the presence/absence of an edge between any two nodes depends on such positions. When considering multidimensional network data, latent space models can help capture the associations between the nodes and summarize the observed structure in the different networks composing a multiplex. This dissertation discusses some latent space models for multidimensional network data, to account for different features that observed multiplex data may present. A first proposal allows to jointly represent the different networks into a single latent space, so that average similarities between the nodes may be captured as proximities in such space. A second work introduces a class of latent space models with node-specific effects, in order to deal with different degrees of heterogeneity within and between networks in multiplex data, corresponding to different types of node-specific behaviours. A third work addresses the issue of clustering of the nodes in the latent space, a frequently observed feature in many real world network and multidimensional network data. Here, clusters of nodes in the latent space correspond to communities of nodes in the multiplex. The proposed models are illustrated both via simulation studies and real world applications, to study their perfomances and abilities.