|
Abstract:
|
In this work, we develop a parsimonious model that captures dynamic local structure in large-scale social networks. We first develop a method for quantifying the local structure by defining ego network representation through collections of isomorphic subgraphs, network motifs. We then utilize a scalable Bayesian Poisson factorization algorithm to provide interpretable descriptions of structural properties from the local subgraph counts while adjusting for the original social network information. The model is domain-independent and could be broadly applied to other networks where multiple nodal events are observed, e.g., social recommendation systems. We demonstrate how our model can be used to model patterns of technology and information diffusion across a large mobile phone network using the mobile phone call detail records (CDR) in an East African nation.
|
