| 英文标题 | Chinese poetry generation model with multi-adversarial training |
| 作者英文名 | Huang Wenming, Ren Chong, Deng Zhenrong |
| 机构英文名 | a.College of Computer & Information Security,b.Guangxi Colleges & Universities Keys Laboratory of Cloud Computing & Complex Systems,Guilin University of Electronic Technology,Guilin Guangxi 541000,China |
| 英文摘要 | Many generation methods still have a large gap from the human creation, especially on the aspects of topical relevance and semantics of verses. To address these shortcomings of existing methods, this paper proposed a framework for generating poems with multiple adversarial training. The poetry generator used the sequence-to-sequence model with the attention mechanism and dual-encoding. Two discriminative model guided the poetry generation, included hierarchical RNN and TextCNN. Meanwhile, the framework used policy gradient for multi-adversarial training. Experiments show that the poetry generation method based on multi-adversarial training effectively improves the relevance between verses and vision keywords, and the semantic connotation of poem is more abundant. |
| 英文关键词 | sequence to sequence; attention mechanism; adversarial training; poetry generation |