Automated Thematic and Emotional Modern Chinese Poetry Composition

Topic and emotion are two essential elements in poetry creation, and also have critical impact on the quality of poetry. Inspired by this motivation, we propose a novel model to inject rich topics and emotions into modern Chinese poetry generation simultaneously in this paper. For this purpose, our model leverages three novel mechanisms including (1) learning specific emotion embeddings and incorporate them into decoding process; (2) mining latent topics and encode them via a joint attention mechanism; and (3) enhancing content diversity by encouraging coverage scores in beam search process. Experimental results show that our proposed model can not only generate poems with rich topics and emotions, but can also improve the poeticness of generated poems significantly.