An Emotion Summarization Method Based On Semantic And Affective Relations In Chinese Micro-blog

With the rapid development of the Internet and the arrival of the media age,the Internet news text guided by micro-blog has become an important way to generate online public opinion.Micro-blog news text has the characteristics of simple structure and rich content,which makes the content of the text more intuitive and convenient.The extraction of emotion abstracts from news texts can meet the needs of users that getting quickly access to current public opinion events,and makes users respond quickly to the content of the information.Therefore,this thesis analyzes the extraction of text sentiment abstracts from four aspects of text,association grouping between text topic and sentences,semantic similarity,emotional similarity and fusion of emotion and semantic.(1)The LDA topic model is used to extract the text topic information from Micro-blog news text,then association grouping is calculated by the similarity of the topic and sentences.(2)Hownet semantic similarity algorithm,based on sememe relation between words,is adopted to calculate semantic similarity in the association group.(3)Using dependency parser to obtain the relationship between topical word and emotional word,<topical word,emotional polarity>structure become the atom element for represent sentence emotion.According to emotional sememes in two sentences,emotional similarity weight is computed by the way both sentences emotional sememes in the count.(4)Text emotional abstract sorting algorithm is proposed in this paper,based on SE-TextRank.The graphic computation is processed in both semantic and emotional domain of the sentence respectively,and fuse the semantic weight with emotional weight in statistically way.We get the result of the semantic emotional weight of the sentence in the association groups at last.Finally,experiments are conducted to test the parameters of different factors and coefficients in order to obtain the optimal value.The method in this paper is deployed in NLPCC2015,micro-blog guided abstract evaluation method,and get satisfactory result.The comparison of the results among TF-IDF,PageRank shows that the method used in this paper are close to the best value,and the emotion summary is more effective and accurate.