Research On The Application Of Chinese-Burmese Bilingual Sentence-level Embedding Semantic Representation Method Based On Neural Network

Burmese sentiment analysis is the basic work of Burma language public opinion analysis and text mining.But there is no artificial annotated sentiment analysis dataset in Burmese,so this work cannot be carried out.For Chinese,there are many mature methods and resourcesBased on the representation of bilingual word vectors and bilingual sentence vectors,this paper applies Chinese sentiment analysis resources and methods to Burman language to complete Burma emotional analysis.1)The Chinese Burmese bilingual corpus is constructed in this paper.Burmese Language is a resource scarcity language.Its scarcity lies in the lack of open corpus and data both at home and abroad,which are the basis of Sino Burmese bilingual Natural Language Processing.The effect of corpus is very important.The quality of corpus influences the progress and quality of subsequent research and further experiments.In this paper,we introduce the methods to obtain the Burmese Language materials,the source of the Burmese Language materials and how to store the acquired corpus,and obtain the bilingual parallel corpus in the Burmese Language.2)This paper proposes a vector representation method for Burmese words,which integrates the grammatical features of Burmese.In this paper,the vector training model of Burmese words is used to extract the syllabic features of Burmese words by using the convolution neural network(CNN)and the gate structure network for the reasons of the complexity of the Burmese word formation,the complexity of the grammar and the few reasons for the Burmese Language training.The model of this paper can not only solve the normal word vector training,but also make a correct representation of the unfamiliar words and unappeared words,and the model of this paper has a stronger ability to characterize the grammar of the Burmese.3)this paper proposes vector representation of Chinese and Burma bilingual words and vector representation of Chinese and Burma bilingual sentences.In order to make the Burmese language use the rich resources and methods of Chinese in Natural Language Processing,by using space mapping,the mapping relationship between Burmese words Vector and Chinese word vector is established by minimizing the space distance between the translated words in the bilingual dictionary,and the iterative algorithm is used to update the dictionary repeatedly,and the optimal mapping relation is obtained.The words in Burmese sentences are transformed into Burmese word vectors,and the Burmese word vectors are mapped into the Chinese semantic space to get the sentence level representation of Chinese Burma bilingual.4)This paper proposes a bilingual expression based emotion classification method for Burmese sentences.Through large-scale Chinese tagging data,a good model is trained in Chinese emotional classification.The Burmese Language is mapped to the semantic space of Chinese in the Burmese Language training,and the characteristics of the Chinese language are used to make up the problem of the lack of characteristics of the Burmese Language.Then the Burma language after mapping is put into the model for further training.Through the constraint of regular items,we can further adjust the characteristics of Burmese data after mapping the small annotation set,and get the Burma sentiment classification model.5)A Burmese sentiment classification prototype system was implemented.The Burmese sentence emotion classification system based on bilingual representation is designed and implemented based on the theoretical results of this paper.The system can be used to emotionally mark Burmese sentences and supplement the Burmese sentiment classification corpus.