基于主题特征的关键词抽取  

Keyphrase extraction based on topic feature

作　　者：刘俊[1] 邹东升[1] 邢欣来[1] 李英豪[1] 

机构地区：[1]重庆大学计算机学院,重庆400044

出　　处：《计算机应用研究》2012年第11期4224-4227,共4页Application Research of Computers

基　　金：中国博士后科学基金资助项目(20110490807);中央高校基金科研业务资助项目(CDJXS10181131)

摘　　要：为了使抽取出的关键词更能反映文档主题,提出了一种新的词的主题特征(topic feature,TF)计算方法,该方法利用主题模型中词和主题的分布情况计算词的主题特征。并将该特征与关键词抽取中的常用特征结合,用装袋决策树方法构造一个关键词抽取模型。实验结果表明提出的主题特征可以提升关键词抽取的效果,同时验证了装袋决策树在关键词抽取中的适用性。Keyphrase extraction is a process for extracting a set of terms from a document.This paper proposed a novel topic feature for extracting keyphrase.This topic feature was computed based on topic model which modeled the topic-word distributions and the topic distributions of document.Moreover,it proposed a keyphrase extraction approach based on bagged decision trees.This approach jointed common features and the proposed topic feature.Experimental results demonstrate that the proposed topic feature can make an improvement for keyphrase extraction.At the mean time,an effective performance can be achieved by the bagged decision trees based approach.

关 键 词：关键词抽取 主题特征 主题模型 装袋决策树 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]