融合句子情感和主题相似性的中文新闻文本情感摘要

摘要

新闻文本情感摘要是指通过提炼、浓缩而产生表达文本全局情感意见的摘要,旨在帮助人们快速获取文本的情感倾向.现有的文本摘要方法仅考虑主题及句子特征等因素,无法获取带有情感意见的文本摘要.针对这一问题,提出了融合句子情感和主题相似性的中文新闻文本情感摘要.首先,对文本中的句子进行情感标注;然后,在LexRank算法中加入情感信息计算句子相似度;最后,根据新闻标题的特殊性计算句子与标题的相似性,再综合以上步骤的结果得到最终的情感摘要.实验结果表明,在ROUGE-1、ROUGE-2和ROUGE-W三个指标上,该方法比传统的LexRank算法均有提升,证明了同时考虑情感信息和主题信息能够更加有效地生成体现文本主要观点、情感的情感摘要.%News opinion summarization aims to produce opinions abstract via refining the text with emotional information,which helps people to know the theme content and tendency of opinions quickly.However,the existing methods only consider the theme and the characteristics of the sentence,which can not get a summary of the text with emotional comments.To address the above problem,this paper presented a method of integrating sentence emotion and topic similarity for Chinese news text opinion summarization.Firstly,it annotated the opinion information of sentences.Secondly,it added opinion information to the LexRank algorithm to compute sentence similarity.Finally,according to the special characteristics of the news title,it calculated the similarity between the sentence and the title.The results of the above three steps were taken into account to generate opinion summary.The results of experiment show that this method is more effective than the classic LexRank algorithm on ROUGE-1 、ROUGE-2 and ROUGE-W.In addition,it also represents that considering both the emotion and theme can help generating opinion summary effectively.