一种应用向量聚合技术的KNN中文文本分类方法

Vector-Combination-Applied KNN Method for Chinese Text Categorization

针对 KNN文本分类方法中不考虑特征词关联的问题 ,提出一种改进方法 .这种方法基于对体现词和类别间相关程度的 CHI统计值分布的分析 ,应用向量聚合技术很好地解决了关联特征词的提取问题 .其特点在于 :聚合文本向量中相关联的特征词作为特征项 ,从而取代传统方法中一个特征词对应向量一维的做法 ,这样不但缩减了向量的维数 ,而且加强了特征项对文本分类的贡献 .实验表明该方法明显提高了分类的准确率和召回率 .

On account of the fact that the traditional method lacks for the consideration of words association, this paper proposes an improved KNN (k-Nearest Neighbor) method for Chinese Text Categorization. This method applies vector-combination technology to extract the associated discriminating words according to the CHI statistic distribution, which indicates the relationship between words and classes. One of the merits of this method is to combine the associated discriminating words to be one feature and abandon the traditional method -- one word per dimension. It not only decreases the dimensions of the text vector, but also strengthens the contribution to categorization of each feature. The experiment shows that this improvement augments the categorization recall and precision obviously.