Research On The Generation Of Tibetan Rhythmic Poems Based On Neural Network

In the ever-changing development of the Internet and artificial intelligence,if we can extract structured data from unstructured Tibetan poems texts,we can not only promote the ability of machines to automatically write poems,but also have a great deal of intelligence in Tibetan information processing.research value.This paper first introduces the research background and status quo of automatic generation of poems.Then it further understands the basic ideas and mathematics of Word2 Vec,RNN,LSTM,End-to-End models,and Attention Mechanisms in the field of Natural Language Processing.Finally,a method for obtaining Tibetan texts from Tibetan web pages or e-books was implemented.These included the process of obtaining Tibetan texts from ePub file types and their code,and the extracting algorithm of Tibetan poems texts from Tibetan texts,and its code.A total of 373,636 Tibetan poems were collected through the extraction algorithm.Each of them was selected as a validation set and a test set,and the remaining ones were used as training sets.The main production model of this paper draws on the automatic generation model of Chinese poems.There are three small models in the Tibetan poems generation model.They are a verse model for training a single verse,a poetic block model for training multiple verses,and a poetic word model for training the topic to generate the first sentence.The basic framework is based on The end-to-end model of the BiLSTM.The training data used by these three small models are not the same,so the tasks that they need to undertake are also different.The BiLSTM is used in the encoder of each model,mainly to core words in each Tibetan verse so that the model can capture the important information in the Tibetan verses without looking at the verses equally.All the word information.When training each model,both positive verses and target verses use positive sequences.This helps the LSTM and the attention mechanism to better learn the rules of using the same number of syllables in the same lexeme from the Tibetan verse,and at the same time guarantees the Tibetan language.The problem of rhythm consistency in new legal poems Usually in the Tibetan poems,the metaphorical words and the source metaphors are in the same position.Therefore,the local attention mechanism is used.This not only speeds up the calculation,but also does not significantly reduce the production results.In each model,the pre-trained Tibetan syllable vectors from the Tibetan poems corpus are used to initialize the input values instead of initializing them with random values.This not only speeds up the model convergence time,but also enhances the results of each model.The BLEU and ROUGE values of the Tibetan poems generation model can reach 67.43% and 68.81%,respectively,indicating that the Tibetan poems generation model can generate a new Tibetan poem with a certain level of flow and fidelity.