Research On Automatic Generation Of Chinese Classical Poems Based On Thematic Attention Mechanism

Classical poetry is a treasure of Chinese traditional culture.However,Ordinary people want to learn and write poetry very difficult.Therefore,the use of natural language processing technology for the automatic generation of classical poetry research to carry forward and inherit Chinese traditional culture and explore machine art creation and other aspects are of great significance.This paper adopts deep learning and natural language processing technology to study the automatic generation of Chinese classical poetry.Our main work and innovations are:(1)In order to solve the automatic construction of poetry-generating training corpora,it is difficult for many successful systems to generate corresponding poetry based on modern concepts.We first adopted a poetic dataset augmentation strategy,using the existing classical Chinese machine translation system to built a "parallel corpus".Considering that text generation does not require the input and the output to be completely equivalent,the input is generally a high summary of the content of the output.We propose a keyword extraction method based on a structured semantic matching framework.First,we express the words and ancient poetry sentences in the translation as the distributed semantic vector,then regard the words and corresponding verses in the translation as the semantic vector matching problem.The innovation here is:we have migrated the dataset augmentation to the poetry generation task for the first time;secondly,our proposed structured semantic matching model provides a new idea for keyword extraction.(2)At present,many systematically generated poems do not have a unified theme for many times,and the semantic coherence of verses has not been well solved.Every word in our speech is different in expressing the importance of our thoughts.We usually call a theme that summarizes the intent of the idea.The theme is represented in the text by a number of keywords.Inspired by such a common sense of life,we propose a poetic generation model based on thematic attention mechanism.We have modified the traditional encoder-decoder framework so that it can simultaneously encode keyword sequences and historically generated content.At the encoder side,we use the keyword weighted average method based on the theme of attention.Our model solves the problem of semantic coherence of the theme drift and verses of poetry generation to a certain extent.In general,we have well solved the problem of the lack of model training corpus and the difficulty of the poetry generation system to generate corresponding poetry based on modern concepts.The experimental results show that thematic attention mechanism has a significant increase in the quality of generated poetry.At the same time,in the comparative experiments of the PPG model,our model also showed better performance.