The paper proposes a novel hierarchical Bayesian model for text data that can capture both intra-sequence structure with convolution and inter-sentence dependency via a gamma dynamic model. The paper is well written and the authors provided extensive comparisons to other methods, showing the usefulness of the approach.