The machine poetry generator imitating Du Fu's styles

摘要

Du Fu is one of the greatest poets in Tang Dynasty. His poems are depressed and sonorous, look gloomy and contemplative, but show desires for better life and brighter future. His poems depict the era of 1000 years ago. To understand and inherit his poems better, we propose a novel machine learning approach with LSTM to generate new Chinese poetry imitating Du Fu's styles automatically. Given the first character, the model will generate the rest characters of a poem according to the writing style of the author. In contrast with the previous methods, besides following the formats, satisfying tonal and rhythm requirements and having semantic coherence, our method could inherit Du Fu's styles to some extent. We build a prototype with Du Fu's poems and the evaluations have illustrated the effectiveness of our approach.