Poet-based Poetry Generation: Controlling Personal Style with Recurrent Neural Networks

Abstract

Poetry style is not only a high-level abstract semantic information but also an important factor to the success of poetry generation system. Most work on Chinese poetry generation focused on controlling the coherence of the content of the poem and ignored the poetic style of the poem. In this paper, we propose a Poet-based Poetry Generation method which generates poems by controlling not only content selection but also poetic style factor (consistent poetic style expression). The proposed method consists of two stages: Capturing poetic style embedding by modeling poems and high-level abstraction of poetic style in Poetic Style Model, and generating each line sequentially using a modified RNN encoder-decoder framework. Experiments with human evaluation show that our method can generate high-quality poems corresponding to the keywords and poetic style.