基于多层挖掘策略的特征选择及在科技文献分类中的应用

Feature selection based on multi-level mining strategy and its application in classification of scientific and technological literatures

提出一种应用于科技文献分类的文本特征选择方法.该方法运用了聚类与关联规则的思想,对文本特征进行逐层选择.同时为提高K-means算法的聚类性能,对K-means算法做了相应的改进,通过为算法的终止条件设定标准值来减少算法迭代次数,减少学习时间;通过删除由信息动态变化而产生的冗余信息,来减少动态聚类过程中的干扰.采用KNN分类器进行对比实验,实验结果表明,该特征选择方法在科技文献分类方面有较高的准确率.

A text feature selection method was proposed for classification of scientific and technological literatures.In this method,the idea of clustering and association rules was used to select the text feature layer by layer.Meantime,the K-means algorithm was modefied to improve its clustering performance,the amount of iteration and the time of self-learning in this algorithm was reduced by means of setting a standard allowance to terminate the tedious iterative computation,and the interference in the process of dynamic clustering was reduced by eliminating the redundant information induced by dynamic change of the information.It was shown by the test for comparison with KNN classifier that the feature selection method presented in this article has a higher accuracy for classification of scientific and technological literature.