The roles and interactions which nodes take part in has a significant impact on the structure of a network. In order to estimate this underlying latent structure, its numbers and composition from real data, flexible treatment of the structural uncertainty and efficient use of available information becomes a key issue. I take a Bayesian nonparametric approach, jointly modeling sparse network, node textual information and potentially unbounded number of components to handle the aforementioned task. I show using synthetic and real datasets that my model successfully learns the underlying structure utperforming previous method.
View on arXiv