Research On Text Generation Algorithm Based On Seq2seq Framework

Text generation is a very important field in natural language processing.It enables the computer to generate high-level text comparable to human beings.It is widely used in text retelling,automatic image description and emotional dialogue.With the birth of sequence to sequence(seq2seq)framework,there are many breakthroughs in the field of text generation.This paper studies the text generation algorithm based on seq2 seq framework.Compared with free language,ancient poetry and couplets have the characteristics of aesthetics and simplicity.Therefore,this paper improves the text generation model of ancient poetry and couplets to improve the quality of their generation.Firstly,this paper briefly introduces the research background and significance of automatic generation of ancient poetry and couplets,and expounds the research status of traditional text generation methods and text generation methods based on deep learning at home and abroad.Secondly,it introduces the related theories and models commonly used in ancient poetry and couplet generation,including language model,word vector model,neural network model and seq2 seq framework.In order to better capture the relevant information of ancient poetry and enhance contextual connectivity,this paper proposes a generation model of ancient poetry based on keyword group.The model is composed of the above encoder,keyword encoder,key phrase encoder and ancient poetry character decoder.The current line of poetry is generated by using keywords,keyword groups and generated poetry.Keyword group is composed of keywords of all lines,which is used to capture the context of ancient poetry.It is found that by introducing keyword groups into the generation process of ancient poems,the model can perceive the keyword information of subsequent poems in advance,so as to effectively improve the semantic coherence between poems.The experimental results show that the model based on keyword group is reasonable and superior.Finally,in order to improve the global perception ability of the model,this paper proposes a couplet generation model based on global features.The model consists of a content encoder,a global feature encoder and a couplet decoder.The global feature coder extracts the upstream information through multiple convolution checks,and then generates the global feature vector through maximum pooling and full join operations.It is found that by introducing global eigenvectors into the process of generating couplets,the overall coherence of generating couplets can be effectively improved.The experimental results show that the result of the model is reasonable and the performance is superior.In addition,this paper also designs and implements a web-based couplet generation system,which is convenient for couplet lovers.