摘要
在文本分类中,文本特征向量通常高达几千甚至上万维,给整个分类过程带来了相当庞大的计算量,因此进行有效的降维处理是非常重要的。在不完备信息系统理论的基础上,结合文本分类的特点,提出了一种量化容差关系和启发式的属性约简算法。实验证明该属性约简算法不仅能有效地降低文本特征向量的维度,同时能保证分类的正确率。
Document vectors are highly dimensional in text classification, possibly there are tens of thousands of dimension, which leads to a massive amount of calculation. Thus, it is important to decrease the dimension. In the paper, the authors present a quantitative tolerant relation and a heuristic algorithm for attribute reduction, combining theory of incomplete information systems with features of text classification. The experiment results illuminate the efficiency, for it can not only effectively reduce the dimension, but also maintain high accuracy of text classification.
出处
《重庆邮电学院学报（自然科学版）》
2006年第3期397-401,共5页
Journal of Chongqing University of Posts and Telecommunications(Natural Sciences Edition)
基金
国家自然科学基金(60373111
60573068)
重庆市教育委员会科学技术研究项目资助
重庆邮电大学科研基金(XJG0516)
关键词
文本分类
粗集
不完备信息系统
属性约简
text classification
rough set
incomplete information system
attribute reduetion
