Bayesian Social Subgraph Generative Models: Social Network Twins using Belief Networks and Ego Behavior Models

Abstract

A key assumption of a Subgraph Generative Model (SUGM) for sparse networks is that a subgraph is independent of lower order subgraphs in a sparse network. This is not entirely true especially for non-sparse networks. Additionally, the generated networks lack the typical properties of a social network because of an assumption of random growth for nodes and edges. Finally, there is no concept for explicit ego choice or bias when connecting to dyadic or triadic relationships. We develop a novel graph generative model referred to as the Bayesian Social Subgraph Generative Model (BASSUGM). We ground the BASSUGM in a proposed sociological model and leverage Bayesian tools like belief networks. We introduce novel concepts like the networks’ macro theme when combines with an ego’s individuality realizes the ego’s intent. We also demonstrate how the social network twin generated with BASSUGM outperforms SUGM for non-sparse, small, social, networks.