Infinite Sparse Block Model with Text Using 2DCRP

A fundamental step in understanding the topology of a network is to uncover its latent block structure. To estimate the latent block structure with more accuracy, I propose an extension of the sparse block model, incorporating node textual information and an unbounded number of roles and interactions. The latter task is accomplished by extending the well-known Chinese restaurant process to two dimensions. Inference is based on collapsed Gibbs sampling, and the model is evaluated on both synthetic and real-world interfirm buyer-seller network datasets.