A Note on Jointly Modeling Edges and Node Attributes of a Network

TL;DR

This paper introduces a probabilistic model that jointly captures the interactions between network edges and node attributes, combining latent space and Gaussian graphical models to better understand complex network dynamics.

Contribution

It proposes a novel joint modeling framework integrating edge and node attribute distributions, linking latent space and Gaussian graphical models.

Findings

Connection to dynamical network processes

Marginal distribution as a random graph model

Existence of limiting distribution for edges

Abstract

We are interested in modeling networks in which the connectivity among the nodes and node attributes are random variables and interact with each other. We propose a probabilistic model that allows one to formulate jointly a probability distribution for these variables. This model can be described as a combination of a latent space model and a Gaussian graphical model: given the node variables, the edges will follow independent logistic distributions, with the node variables as covariates; given edges, the node variables will be distributed jointly as multivariate Gaussian, with their conditional covariance matrix depending on the graph induced by the edges. We will present some basic properties of this model, including a connection between this model and a dynamical network process involving both edges and node variables, the marginal distribution of the model for edges as a random graph model, its one-edge conditional distributions, the FKG inequality, and the existence of a limiting distribution for the edges in a infinite graph.