Research On Summary Generation Methods For Chinese-Vietnamese News Documents

With the promotion of the “The Belt and Road” construction,the bilateral cooperation and exchanges between China and Vietnam have become closer and deeper,and the scope of the field has been extensive.Relevant news reports have been increasing.It is very important for bilateral cooperation to realize major news items of common concern from diffe rent countries.This paper studies the summarization method for the Chinese and Vietnamese bilingual news documents which aim at summarizing and refining the Chinese and Vietnamese bilingual news and providing a summary system.Due to text analysis involving different languages,we focus on the impact of associations between multilingual texts on summarization task.The key issue is how to map Chinese and Vietnamese bilingual texts to the same space for association analysis;how to identify and summarize important information from a large number of news documents.In recent years,deep learning model has shown great potential in many fields.Therefore,this paper focuses on the promotion of tex t summary performance under the deep learning framework,and mainly completed the foll owing research work:(1)Abstractive text summarization based on attention mechanism of sentences correlation.The task of single text summarization aims at summarizing t he contents of document and generating short summari zation to express the most important information in the original text.The key is the validity of the model to evaluate the importance of the sentence in the source document.In order to enhance the ability of text generation model to recognize sentence-level information,this paper studies a framework of summarization generation based on neural network model with sentence-relevance attention mechanism.Firstly,the document is encoded using a hierarchical Bi-LSTM to obtain sentences vectors.Secondly,the correlation between sentences is analyzed by gate network to achieve the importance and redundancy evaluation of the sentence s.Finally,it is proposed to integrate the sentence correlation analysis into the attention mechanism to generate summarization.Experiments show tha t the proposed method achieves good results on several rouge evaluation indexes.(2)Chinese-Vietnamese news documents summarization based on elements-related attention mechanism.The task of Chinese and Vietnamese news documents summarization is designed to extract key sentences by the score of every sentence in the bilingual news document set des cribing the same event.The score of a sentence depends on the significance of the sentence in the multilingual document set.Therefore,it involves cross-language documents correlation analysis.Considering the consistency of elements information in the s ame news event,this paper proposed to combine bilingual elements with the attention mechanism of neural network for association analysis t o guide the summarization generation.Firstly,some statistical features can be analyzed by the vector representation of word embedding and the Chinese-Vietnamese dictionary,such as elements co-occurrence degree,word frequency,sentences position and sen tences relevance.Then,these features are integrated into the neural network,and the attention mechanism based on bilingual elements is pr oposed to set the salient scores of sentences.Finally,the sentences with high scores are selected and the summary is generated by redundant screening based on similarity.Experiments show that the method achieves good results.(3)Building a Chinese-Vietnamese bilingual news document summarization prototype system.The system c an analyze and summarize bilingual news documen t set at the same time,automatically select the most relevant content in the source documents,and set it into a short text to presen t a brief summary for users.