Abstract
A new dynamic latent space eigenmodel (LSM) is proposed for weighted temporal networks. It accommodates integer-valued weights, time-varying sparsity and time-varying node latent features. Lagged and contemporaneous dependence across nodes and features are introduced, a characteristic neglected in the LSM literature. A Bayesian approach is used to address two inference challenges: latent feature estimation and choice of latent space dimension. We provide a point-process representation of the network weights, derive the joint posterior distribution of the features, and propose a new efficient sampler for posterior approximation in high dimensional networks. A Laplace approximation of the partial marginal likelihood is used to sample the latent dimension. Two illustrations, that are relevant in international economics and neuroeconomics, show the ability of the proposed framework to uncover dynamic latent structures in trade and neural connectivity. Overall, our procedure is general, as it can be easily adapted to static and dynamic settings, as well as to other discrete or continuous weight distributions.

This publication has 44 references indexed in Scilit: