基于多种特征选择的NB组合文本分类器设计

针对朴素贝叶斯（NB）分类器在分类过程中存在诸如分类模型对样本具有敏感性、分类精度难以提高等缺陷,提出一种基于多种特征选择方法的NB组合文本分类器方法。依据Boosting分类算法,采用多种不同的特征选择方法建立文本的特征词集,训练NB分类器作为Boosting迭代过程的基分类器,通过对基分类器的加权投票生成最终的NB组合文本分类器。实验结果表明,该组合分类器较单NB文本分类器具有更好的分类性能。

Design of NB Combination Text Classifier Based on Various Feature Selection

There are some shortcomings when it uses single Naïve Bayes(NB) classifier to classify text. For example, the classification model is sensitive to samples, and the precision is always hard to be improved. This paper proposes a method that creates different feature set which is used in training NB classifier using different method to extract text features in each iteration of Boosting procedure. An NB combination classifier for text categorization is designed based on different feature selection methods. Experimental result shows that the combination classifier is more effective than single NB classifiers.