A new LDA formulation with covariates

The Latent Dirichlet Location (LDA) model is a popular method for creating mixed-membership clusters. Despite having been originally developed text analysis, LDA has used wide range of other applications. We propose new formulation the which incorporates covariates. In this model, negative binomial regression embedded within LDA, enabling straight-forward interpretation coefficients and analysis quantity cluster-specific elements in each sampling units (instead being focused on modeling proportion cluster, as Structural Topic Models). use slice Gibbs algorithm to estimate parameters. rely simulations show how our able successfully retrieve true parameter values ability make predictions abundance matrix using information given by illustrated real data sets from three different areas: text-mining Coronavirus articles, grocery shopping baskets, ecology tree species Barro Colorado Island (Panama). This allows identification clusters discrete provides inference relationship between covariates these