Utilizing Context in Generative Bayesian Models for Linked Corpus