|
Abstract:
|
In this paper, we propose a bipartite graph modeling framework using latent features generated by dependent Indian Buffet Processes (dIBP). The model not only preserves the properties brought by marginal IBPs, but also accommodates the dependence between the two nodal regimes of the bipartite structure through a correlated feature generating process. We provide an MCMC inference algorithm and illustrate the enriched modeling capabilities through graph link prediction and graph co-clustering.
|
