Topic Models Conditioned on Arbitrary Features with Dirichlet-multinomial Regression

Although fully generative models have been successfully used to model the contents of text documents, they are often awkward to apply to combinations of text data and document metadata. In this pap