Visualizing Documents based on Topic Models

1We propose a method based on a topic model for visualizing documents withthe latent topic structure. Our method assumes that both documents and topicshave latent coordinates in a two-dimensional Euclidean space, or visualizationspace, and visualizes documents by considering a generative process of docu-ments as a mapping from the visualization space into the space of documents.A visualization, i.e. latent coordinates of documents, can be obtained by flttingthe model to given documents using the EM algorithm. In the experiments,we demonstrate that the proposed model can locate related documents closertogether than conventional visualization methods.