Research On The Method Of Differential Summarization Of Bilingual News

Summarizing the Differences analysis in Bilingual News is one of the important research directions in the field of natural language processing.With the development of the Belt and Road Initiative,China's exchanges with other countries have become increasingly close,and we need to understand more accurately the different views of other countries and our country on a certain topic.Various news media at home and abroad will publish a large number of news reports,which can help us to better understand the difference in the treatment of a topic by different countries.The research goal of this article is to obtain the difference of news reports under the same topic in the two languages through bilingual news reports,according to the designed summary extraction method.The previous research on multi-document summary did not make full use of topic information and semantic information to extract multi-document summary.The research on differential summary currently only stays in the stage of relying on machine translation for bilingual processing.The analysis and research of the bilingual news summary the differences method are as follows:1.Multi-document summary extraction based on multi-information sentence graph modelAiming at the problem that the existing multi-document extraction method can not make good use of the topic information and semantic information of sentences,a multi-document summary extraction method combining multi-information sentence graph model is proposed.First,take the sentence as a node to construct a sentence graph model;then,the sentence topic probability distribution and sentence semantic similarity based on the sentence Bayesian topic model and the word vector model are fused to obtain the final relevance of the sentence,combined with the topic Information and semantic information are used as the edge weights of the sentence graph model;finally,the summary method of the multi-document is described by the summary method of the minimum dominating set of the sentence graph.This method combines the multi-information sentence graph model to combine the topic information,semantic information and relationship information between sentences.Compared with the traditional method,this method can effectively improve the comprehensive performance of summary extraction.2.Summarizing the Differences in multi-document Bilingual News based on graph convolutionIn order to extract the summarizing the differences efficiently,the summarizing the differences are further extracted on the basis of the summary that have been extracted,and the existing methods use machine translation to translate the bilingual texts.Based on the translation of bilingual news,there is a semantic conversion bias in the extraction of summarizing the differences in bilingual News Problem,and better based on the presentation ability of bilingual news sentences and sentence relationship information,a method for summarizing the differences in multi-document bilingual news based on graph convolution is proposed.First,construct the structure of the summarizing the differences graph based on the extracted summary information.Secondly,the graph convolution neural network is used to obtain the sentence node representation after the difference summary graph is aggregated.Then,using the GRU model,a vector representation of the bilingual summary document is obtained.Then,by calculating the bilingual summary document vector and the sentence node representation,the significance score of the summary sentence is obtained.Finally,according to the saliency scores,the summaries of Chinese and English are extracted.This method avoids the bias of using bilingual translation in extracting differential summary,and improves the quality of summarizing the differences in multi-document bilingual news.The experimental results show the effectiveness of the method and prove the direction of processing summarizing the differences in bilingual news.The advantages of combining the sentence relationship in the figure with the representation ability of the neural network.