The Study Of Classical Poetry Automatic Generation Based On Neural Network

With the development of Internet and the acceleration of life pace,people are generating and processing a large amount of text information every day.On the one hand,facing massive text information,people need to efficiently extract the most important information from it so as to avoid wasting time;on the other hand,a lot of jobs,such as news reports,need to quickly generate text so as to be able to pass out information in a timely manner.If the machine can automatically generate the text meeting the above requirements,it will greatly save manpower and material resources.Therefore,the research of automatic text generation is of great significance.This paper mainly studies the automatic text generation technology based on neural network.Appling the technology to the field of Chinese classical poetry,we put forward the automatic generation model of poetry title and the automatic generation model of poetry content.The automatic generation model of poetry title in this paper can automatically generate semantic related title given poetry content.First,we use the word vector model to represent the content of poetry as a matrix;then,the convolutional neural network(CNN)is used to encode the matrix into a vector;finally,a recurrent neural network(RNN)is used to generate the poetry title character by character.In order to capture the intra-sentence and inter-sentence semantic relevance of the poetry content at the same time,this paper designs two kinds of filters in the CNN,one covering two adjacent characters in one sentence while another covering two characters in the same position of two adjacent sentences.The experimental results show that the proposed model can capture the semantic meaning of poetry content more accurately than the RNN model,and can generate more semantic related titles.The automatic generation model of poetry content in this paper can automatically generate semantic related poetry content given keywords.First,we expand the input keywords and generate the first sentence using a language model;then,the partially generated poem is encoded into a vector using convolutional operation and RNN;finally,we use the RNN to generate the next sentence.In order to make the theme of poetry more relevant to keywords,this paper transforms the keywords to a vector,and uses it as the input of RNN to generate every sentence of the poetry.The experimental results show that,compared with the traditional machine writing model,the proposed model can generate more semantic related poetry to the input keywords.