文本分类中用于协同训练的特征集分割[EB/OL]
北京:中国科技论文在线
苏金树 1
) 摘要: 用于文本分类领域的协同训练往往需要特征集的一个自然独立分割,但对大多数语料而言这种分割都很难获取或不存在
本文给出了特征子集间在类别下条件独立性的定量描述,并在此意义下提出了局部特征集分割的策略以及基于样本聚类和基于图分块的两种以独立性为前提的特征集分割算法
在两个语料库上的分类实验证明在我们方法所获得的特征集分割下,协同训练方法能有效利用未标注样本提高分类器的综合效果,从而有效扩展了协同训练的可用性
关键词: 文本分类;半监督学习;协同训练;特征集分割;局部分割;局部自适应聚类;图分块 Zhang Bo-Feng 1,*
Xu Xin 2, ( 1、 School of Computer, National University of Defense Technology
2、 School of Mechantronics Engineering and Automation, National University of Defense Technology
( 1、 国防科学技术大学计算机学院
2、 国防科学技术大学机电工程与自动化学院
苏金树
总览 评价 张博锋 1
) Abstract: In the area of text categorization, co-training method always needs a naturally independent split of the feature set which is hardly to obtain or rarely exists for most of the corpus
This paper presents the quantitative description of the conditional independence of feature subsets given the class and suggests a strategy for splitting feature set locally in this sense
Two algorithms respectively based on sample clustering and graph partitioning for feature set splitting in the precondition of independence are proposed
Experimentation on two corpuses shows that, in the feature divisions produced by our methods, the combined effectiveness of the co-trained classifiers is improved by applying the unlabeled samples
As a result, the applicability of the co-training method is extended
Keywords: text categorization; semi-supervised learning; co-training; feature set splitting; local splitting; locally adaptive clustering; graph partitioning 下载PDF阅读器 PDF全文下载: 初稿 ( 167 ) 作者简介: 通信联系人: 【收录情况】 中国科技论文在线: 张博锋
【详情见下载】