基于主题的SE-TextRank情感摘要方法  

SE-TextRank Opinion Summarization Method Based On Topic Model

作　　者：刘志明[1] 于波[1] 欧阳纯萍[1] 余颖[1] 阳小华[1] 翟云[2] LIU ZhiMing YU Bo OUYANG ChunPing YU Ying YANG XiaoHua ZHAI Yun(School of Computer Science and Technology, University of South China, Hengyang 421001, China E-Government Research Center, Chinese Academy of Governance, Beijing 100089, China)

机构地区：[1]南华大学计算机科学与技术学院,衡阳421001 [2]国家行政学院电子政务研究中心,北京100089

出　　处：《情报工程》2017年第3期97-104,共8页Technology Intelligence Engineering

基　　金：国家自然科学基金(61402220;61672178);湖南省哲学社会科学基金(14YBA335;16YBA323);湖南省教育厅科研项目(15C1186)的资助

摘　　要：技术的目的是以简洁的形式准确表达文章的核心情感内容。为解决不同的文档结构及内容特征等问题对摘要结果的影响,提出了一种基于主题的SE-TextRank情感摘要方法。通过LDA模型自动获取收敛后的文本主题,利用余弦距离算法进行主题句子分组,使用传统多特征融合以及SE-TextRank情感摘要算法对组内中心句抽取,最终获取目的摘要。实验表明,采用此方法能够更为高效的获取新闻文本摘要结果。The purpose of the text sentiment summarization is to express the content of the article in a concise form. A topic-based SE-TextRank emotional abstract method was proposed in this study to solve the influence of different document structure and content characteristics on abstract results. This study obtained the convergent text theme automatically through LDA model, grouped the sentence topic through the cosine distance algorithm, applied traditional multi feature fusion and SE-TextRank sentiment summary algorithm to extract the central sentence within the group, and ultimately get the purpose summary. Results of the experiment showed that this method can be used to obtain the results d news text more efficiently.

关 键 词：文本摘要 LDA模型 余弦距离 SE-TextRank 特征融合 

分 类 号：G35[文化科学—情报学]