基于归一化割的主题划分算法研究  

Probe: Normalized Cuts Based Topic Partition

作　　者：王洋洋[1] 刘柏嵩[2] 刘薇[1] 

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211 [2]宁波大学网络中心,浙江宁波315211

出　　处：《宁波大学学报（理工版）》2013年第4期40-44,共5页Journal of Ningbo University：Natural Science and Engineering Edition

基　　金：国家社科基金(08CTQ014);浙江省教育厅科研项目(Y200908634)

摘　　要：为了保证抽取信息的全面性,主题划分成了不可或缺的工作.借助同义词词林,从词语的语义角度计算文本中各个段落间的相似度,建立段落文本关系图.基于文本关系图对归一化割分割准则中权值矩阵的构建做出调整,使之更能体现出段落间的相似程度,并使用该准则对文本进行主题划分.结果表明,该方法无论是对连续段落还是跨段落表达同一主题的主题划分均较为有效.To ensure the completeness of information extraction,the topic partition is one of the indispensable tasks.With the aid of Tongyici Cilin,we first seek the similarity between paragraphs from the point of semantic computing,based on which we then establish text relation map.Using and accordingly adjusting the weight matrix,the degree of similarity between paragraphs can be more accurately obtained,in which the Normalized Cut approach is adopted to complete the topic partition of text.The experimental results show that the method is effective either for consecutive paragraphs or for cross-paragraphs expressing a similar topic.

关 键 词：同义词词林 文本关系图 段落相似度 主题划分 归一化割 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]