Research And Application Of Ancient Chinese Poetry Generation Method Based On Topic Control

Ancient poetry is one of the greatest art forms in the brilliant history of Chinese civilizati on,it has graceful rhythm,rigorous structure and with rich emotions.From ancient to modern times,it has attracted countless loyal followers with its prodounded and less demotic style ae sthetic and elegance,but there’s a lot of barriers for the average person who want to learn and write poetry.With the rapid development of neural network in recent years,the natural langu age processing has also reached a new level and provided technical support for the automatic generation of poetry.It is of great significance to carry forward and inherit the traditional Chi nese culture and explore the art creation of artificial intelligence through the relevant technol ogy to research the automatic generation of poetry.This article studies the automatic creation of traditional Chinese poetry,through reading the relevant literature,it is found that almost all poetry which generated by the existing metho ds have the problem of Poor continuity and lack of thematic.For this purpose,this article pro poses a method of poetry generation based on theme control,which will introduces the idea o f first conceiving and then writing into the process of poetry automatic generation,and before generation,plan a writing outline for the poetry to guide the theme.Considering the outline e xpression ability is limited.On this basis,this article also uses LDA thematic model to deeply dig out the potential thematic information as the thematic reference knowledge of poetry,wh ich is used to dynamically expand the relevant thematic information in the process of generati ng poetry.Eventually,the outline information and related thematic information dynamically a cting on the process of poetry generation,so that makes the whole poetry theme shared,and then,generated the poetry which is meanin g clear and the contact between the sentences is close.The work of this article revolves around the method of poetry generation based on topic contr ol,there are some major innovations as follow:1.In previous studies,the task of participle in ancient poetry is accomplished by the tools of word segmentation which based on the modern context,in this way,the artistic conception of ancient poetry is often destroyed and produced a large number of garbage fragments,which is not conducive to the construction of the training data in this article.Therefore,based on the c ontext of ancient poetry,this article constructed a participle word list,in the participle task wi th the help of word table to achieve the high quality participle,enhance the quality of training date,thus,improves the poetry generation effect.2.The poetry generation model of thematic control in this article is integrates the additional i nput of outline information and relevant theme knowledge,which is based on the foregoing i nformation,and these additional input with prior knowledge through the Attention Mechanis m,it dynamically affects the generation process of poetry,enhances the thematic nature of po etry generation.3.In order to make the model to capture and learn the thematic information related to poetry generation of poetry effectively,this article put forward a method of thematic information bas ed on hierarchical structure.and this method code the thematic information on the basis of ch aracter and subject terms of Self-attention Mechanism which is maximized retention of thema tic information.Finally,in view of this model,a multi-models comparative test is carried out,The experiment shows that,Compared with other models,the proposed poetry generation mo del based on thematic control has some improvement,Especially in the aspect of coherence a nd significance,the validity of the work and the model in this paper is verified.