基于RNN模型与LSTM模型的机器作诗研究  

作　　者：武丽芬[1] 严学勇[2] 赵吉 

机构地区：[1]晋中学院数学系,山西晋中030619 [2]中国联通晋中分公司,山西晋中030601 [3]晋中学院信息技术与工程系,山西晋中030619

出　　处：《科技创新与应用》2021年第27期48-50,共3页Technology Innovation and Application

基　　金：山西省教育科学“十二五”规划课题(编号:GH15052);晋中学院“大数据技术应用”创新团队(编号:jzxyjscxtd202105)。

摘　　要：现如今自然语言处理在人类生产生活中起到极其重要的作用,随着各类机器学习算法以及深度神经网络的发展,各类写诗机器人频频出现。文章通过机器写诗系统研究了循环神经网络(RNN)与长短期记忆网络(LSTM)在唐诗写作方面的模型效果。本系统基于谷歌开源的深度学习框架TensorFlow,Python作为开发语言,交叉熵损失作为实验结果评价依据进行研究,其结果有力证明了LSTM模型在作诗方面的优越性,并阐释了其具有优越性的具体原因。Nowadays,natural language processing plays an extremely important role in human production and life.With the development of various machine learning algorithms and deep neural networks,various poetry writing robots appear frequently.This paper studies the model effect of Recurrent Neural Networks(RNN)and Long Short-Term Memory(LSTM)network in poetry writing by means of machine poetry writing system.This system is based on Google's open source deep learning framework TensorFlow and Python as the development languages,and with the cross-entropy loss as the basis for the evaluation of the experimental results.The results strongly prove the superiority of the LSTM model in poetry,and explain the specific reasons for its superiority.

关 键 词：深度学习 循环神经网络 长短期记忆网络 TensorFlow 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]