| People can get all kinds of news easily and quickly by abundant network resources,and there are lots of emotional news text in these network resources.Because of these positive or negative emotional information,people's attitudes to current events can be changed imperceptibly,and the development trend of news and public opinion can be influenced directly or indirectly.But in these vast amounts of data,most of the data is useless and meaningless to the user.It is an urgent problem to get the main information and the emotion expressed in texts from mass data quickly and efficiently by reorganizing and analyzing automatically and condensing,refining,and generating an emotional summary of the document's core content,it can help people understand the latest social trends and reduce the browsing burden caused by information overload.At the same time,it is also helpful for the public opinion supervision departments to take positive and effective measures to guide the public's attention reasonably,and reduce the spread of negative news in the network.In this paper,a research and analysis ofemotion abstracts in Chinese news texts by fusing sentence emotion with topic similarity is presented.The existing methods of text summarization just take the theme and sentence features into consideration,it is hard to obtain a text summary with affective views.To solve this problem and the comprehensive method in NLPCC2015 news summary task used based on the graph ranking model and sentence feature,a sentiment summarization method which combines sentence emotion and topic similarity in Chinese news text is proposed.In this paper,the current situation of emotion summarization both at home and abroad was analyzed.At the same time,the data set of test was preprocessed by separating parts of speech and sentences,removing stop words,and filtering sentences.Compared with the general text summarization,the thesis combines two aspects: sentence emotion and sentence topic relevance.Firstly,in order to integrate emotion information,a combination emotion dictionary was established,and the emotion of sentences in the news was tagged by using the emotional dictionary,and then the sentence emotion feature vector was constructed.Based on deep analysis of the LexRank algorithm,the emotional weight of sentence was acquired by the incorporation of emotion information,it helps improve the relationship between nodes in the model and the calculation of edge weight.Secondly,through statistics and analysis of the characteristics of news texts,we can select the topicsentences that can represent news subject and the topic similarity of sentences were calculate in the news.Finally,emotion abstract in news text was gotten under the help of sentence weights in terms of sentence emotion and topic similarity.This method is based on the method used in the NLPCC2015 news digest task by the author,through integrates the sentence emotion and the theme similarity to propose the news text emotion summarization method.In this paper,the affective feature weights and the weighting parameters of different factors were compared and analyzed,and the relative optimal values of each parameter were obtained by experiment.Then,this experiment based on sentence emotion and topic similarity was compared with the traditional methods based on statistical principle,LexRank based on graph model and LDA based on topic probability,it shows that this method has been improved in the evaluation index,and has achieved a certain effect on the extraction of the emotion summary of news text,and can get a more representative summary of the summary. |
PDF Full Text Request