Automatic Generation Method Of Prose Poem Based On Deep Learning

Poetry is a kind of special charm of literary genre.It is a symbol of the development of human civilization.The automatic generation of computer for human poetry is a challenge of natural language generation.In this paper,through the establishment of large-scale prose poem corpus,we realize the creation of poem based on the deep learning models and achieve the computer assisted creation of prose poem.It can provide great convenience for the majority of lovers of prose poetry and literature,and it can also help spread and propagate Chinese language.In this paper,I focus on the creation of Chinese prose poems and studies the computer automatic generation of prose poems.First of all,it classifies and summarizes the research results in the area of automatic generation of verses,and discusses the basic concepts,specific structures and training methods of deep learning.On this basis,a method of poetry generation based on topic model is proposed.Through the segmentation of prose poems and each verse,it is applied to the corpus of a given topic,and the same subject clustering based on topic model features is used for word segmentation results.This method relies on the statistical information of auto-indexing of phrase phrases,which effectively solves the clustering problem of synonym and polysemy phrase phrases in prose poetry generation,and establishes a relatively complete knowledge corpus.In the process of automatic generation of prose poems,the method obtains the context structure model by improving the feature extraction layer of the convolutional neural network.This method is different from the traditional Naive Bayesian probability model,and realizes the calculation based on n-gram.By sequentially merging adjacent result vectors,the continuous vector representation of the verse text is obtained,and the characteristic information of the long sentences is well obtained.Then the context structure model output vector is used as the input to generate the model,the weighted LSTM model is used to input and the sentence is output in a loop to achieve the automatic generation of prose poetry with good contextual relevance.Finally,the BLEU automatic evaluation,manual evaluation are two complementary ways to design experiments and add turing test to establish a more standard evaluation system.According to the experimental results,the model established by the method and the designed system can realize the computer automatic generation of prose poems,which provides the theoretical and experimental basis for further research in the future.