QQ客服 官方微博 反馈留言

	Select strong information features to improve text categorization effectiveness
	Dejun Xue ; Maosong Sun
	2010-05-06 ; 2010-05-06
关键词	Practical Theoretical or Mathematical/ feature extraction information theory pattern classification text analysis/ strong information features text categorization high feature dimensionality weak information features constrained information gain measure feature selection measure Chi measure class centroid based classifier Chinese character bigram features Chinese documents large scale document collection/ C7240 Information analysis and indexing C1250 Pattern recognition C1260 Information theory
中文摘要	High feature dimensionality is one of the main obstacles in text categorization (TC). This paper focused on the solution of feature selection to reduce feature dimensionality in TC. We first classified the original features into three types according to their contributions to categorization, including strong information features, weak information features, and irrelevant features. Then, we put forward the constrained information gain (CIG) measure that preferred to low-frequency informative features for categorization by ignoring negative evidence in classic IG measure. By concentrating on the first type of feature, we further proposed a novel feature selection measure, Chi-CIG, by combining Chi and CIG measures. Based on class-centroid-based classifier and Chinese character bigram features, a TC system for Chinese documents was designed. Experimental results on a large-scale document collection (71,674 documents) indicated that Chi-CIG measure set up a more effective feature set for categorization than did classic Chi and IG measures.
语种	英语 ; 英语
出版者	Freund & Pettmen Publishers ; UK
内容类型	期刊论文
源URL	[http://hdl.handle.net/123456789/10064]
专题	清华大学
推荐引用方式 GB/T 7714	Dejun Xue,Maosong Sun. Select strong information features to improve text categorization effectiveness[J],2010, 2010.
APA	Dejun Xue,&Maosong Sun.(2010).Select strong information features to improve text categorization effectiveness..
MLA	Dejun Xue,et al."Select strong information features to improve text categorization effectiveness".(2010).

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。