摘要
针对文本分类中几种典型的特征选择方法存在的不足,提出基于优化文档频和信息量的特征选择方法。该方法首先使用优化的文档频方法进行特征选择以降低文本向量的稀疏性,然后利用所提属性的约简算法消除冗余,从而获得较具代表性的特征子集。实验结果表明:该方法同3种经典特征选择方法相比,"互信息"和"统计量"以及"信息增益"都要好。
We firstly analyzed several classic feature selection methods and summarized their deficiencies, and then combined word frequency with document frequency and presented an optimal document frequency method, next, introduced rough sets and presented an attribute reduction algorithm based on information quantity, finally, combined the attribute reduction algorithm with the optimal document frequency method and proposed a comprehensive feature selection method. The comprehensive method firstly uses the optimal document frequency method to select feature to reduce the sparsity of feature spaces, and then uses the attribute reduction algorithm to eliminate redundancy, so can acquire the feature subset which are more representative. Experimental results show that the comprehensive method is better than mutual information, chi-square statistic and information gain which are three best conventional feature selection measures.
出处
《济南大学学报(自然科学版)》
CAS
北大核心
2009年第4期363-367,共5页
Journal of University of Jinan(Science and Technology)
基金
四川省科技计划(2008GZ0003)
国家重点基础研究发展规划(973-2004CB318003)
中国科学院知识创新工程重要方向资助(KJCX-YW-S02)
关键词
特征选择
词频
文档频
粗糙集
信息量
属性约简
feature selection
word frequency
document frequency
rough set
information quantity
attribute reduction