A Learning and Practicing System to Support Effective Poetry Generation Based on Neural Network

Abstract

Chinese ancient poetry has been a favorite literary form and is still very popular after thousands of years. As opposed to free language, poetry has the characteristics of aestheticism and conciseness. People can easily judge the quality of ancient poetry, so the generation of ancient poetry can be used as an important method for beginners to learn NLG models and judge the performance of the models. Therefore, we present a novel system called LiBai to facilitate the comprehensive generation and analysis of poetry. The system LiBai includes two main functionality modules - a versatile model library and a user-friendly and interactive studio. LiBai can help user: 1) learn popular poetry generation models systematically, including the model introduction, network structures and performances; 2) through adjusting various model parameters, interactive training, predicting and directly apply these models on real data easily and 3) simply and quickly analyse the generated poetry.