摘 要:近年来,利用深度学习技术实现古诗自动生成逐渐火热。研究者多采用基于语句或基于图片作为古诗自动生成系统的输入,完成古诗的自动生成。研究发现,目前基于语句作为输入时多受限于输入字数的限制,导致无法满足想要进行自由创作的需要。为此,文章基于深度学习利用 LSTM 神经网络实现了一种古诗自动生成系统,该系统可基于任意长度语句作为输入实现古诗自动生成;为了方便操作和查看,利用 Tkinter 实现了系统功能的可视化;最后通过人工评估方法论证了系统的可行性。
关键词:深度学习;古诗自动生成;LSTM;任意长度
DOI:10.19850/j.cnki.2096-4706.2021.19.024
中图分类号:TP18 文献标识码:A 文章编号:2096-4706(2021)19-0097-04
Research on Automatic Generation System of Ancient Poetry Based on Improved Deep Learning
LYU Jing, CHU Lili, GONG Ruixue
(Liaoning University of Technology, Jinzhou 121001, China)
Abstract: In recent years, the use of deep learning technology to realize the automatic generation of ancient poetry is becoming more and more popular. Researchers often use sentences or pictures as the input of the automatic generation system of ancient poetry to complete the automatic generation of ancient poetry. It is found that the current sentence based input is limited by the number of words, which can not meet the needs of free creation. Therefore, based on deep learning, this paper realizes an automatic generation system of ancient poetry by using LSTM neural network. The system can realize the automatic generation of ancient poetry based on arbitrary length sentences input; in order to facilitate operation and viewing, Tkinter is used to realize the visualization of system functions; finally, the feasibility of the system is demonstrated by manual evaluation method.
Keywords: deep learning; automatic generation of ancient poetry; LSTM; arbitrary length
参考文献:
[1] 马玉波,尹志刚 . 试论诗歌特征与读诗的着眼点 [J]. 教学与管理,2007(6):71-72.
[2] IKA J,DAENA F,SVOBODA A. Deep Learning [J].Nature Methods,2019.
[3] ZHANG C,WANG Z Y,LYU J. Research on early warning of agricultural credit and guarantee risk based on deep learning [J/OL]. Neural Computing and Applications,2021,33:[2021-08-03].https:// doi.org/10.1007/s00521-021-06114-3.
[4] The Pocketa, Pocketa School [N],Time,1962-05-25.
[5] ZHANG X X,LAPATA M. Chinese Poetry CGeneration with Recurrent Neural Networks [C]//Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing(EMNLP). Doha:Association for Computational Linguistics,2014:670-680.
[6] 黄文明,卫万成,邓珍荣 . 基于序列到序列神经网络模型的古诗自动生成方法 [J]. 计算机应用研究,2019,36(12):3539- 3543.
[7] 梁健楠,孙茂松,矣晓沅,等 . 基于神经网络的集句诗自动生成 [J]. 中文信息学报,2019,33(3):126-135.
[8] 陈佳舟,黄可妤,封颖超杰,等 . 基于古代绘画的古诗自动生成方法 [J]. 计算机辅助设计与图形学学报,2021,33(7): 1038-1044.
[9] 卫万成,黄文明,王晶,等 . 基于多任务学习的古诗和对联自动生成 [J]. 中文信息学报,2019,33(11):115-124.
[10] YI X Y,LI R Y,SUNMAO S. Generating Chinese Clasical Poems with RNN Encoderdecoder [J].Computation and Language, 2017.
[11] HOCHREITER S,SCHMIDHUBER J. Long Short-Term Memory [J]. Neual computation,1997,9(8):1735-1780.
[12] MESSINA R,LOURADOUR J. Segmentation-free handwritten Chinese text recognition with LSTM-RNN [C]//2015 13th International Conference on Document Analysis and Recognition (ICDAR).Tunis:IEEE, 2015:171-175.
[13] LIU C W,LOWE R,SERBAN L V,et al. How NOT To Evaluate Your Dialogue System:An Empirical Study of Unsupervised Evaluation Metrics for Dialogue Response Generation [J/OL].arXiv:1603.08023 [cs.CL].[2021-09-01].https://arxiv.org/ abs/1603.08023v2.
作者简介:吕敬(1996—),男,汉族,贵州兴义人,硕士研究生在读,研究方向:自然语言处理;褚丽莉(1970—),女,汉 族,辽宁锦州人,教授,博士,研究方向:现代通信网络理论与技术、数据通信与网络;龚瑞雪(1997—),女,汉族,辽宁沈阳人, 硕士研究生在读,研究方向:通信技术及其应用工程。