基于二进制烟花算法的特征选择方法

发布时间：2018-02-12 02:18

本文关键词： 词文本分类特征选择二进制烟花算法信息增益卡方统计量　出处：《情报学报》2017年03期 　论文类型：期刊论文

【摘要】：文本分类中的特征选择方法对分类性能有重要的影响。烟花算法是一种解决优化问题的群体智能优化方法,而特征选择的本质是离散空间的优化组合问题。本文采用二进制编码方式,将烟花算法应用到特征选择上,通过改进二进制烟花算法及其参数设置,从而达到提高分类性能的目的。实验数据使用中英文两种语料库,其中英文数据使用路透社的21578经典文本分类语料库(Reuters21578)的R8单标签语料集;中文数据使用复旦语料库,每种语料库随机抽取训练文本和测试文本各为1800篇。分别使用KNN、SVM分类器在预选维数为300维、600维、900维、1200维、2000维、3000维、4000维、5000维、10000维中进行实验。结果表明:采用二进制烟花算法对特征词选择进行优化后,其效果优于传统的特征选择方法,且具有良好的稳定性。
[Abstract]:The feature selection method in text classification has an important effect on the classification performance. Fireworks algorithm is a swarm intelligence optimization method to solve the optimization problem. The essence of feature selection is the optimal combination problem in discrete space. In this paper, the binary coding method is used to apply the fireworks algorithm to feature selection, and the binary fireworks algorithm and its parameter setting are improved. In order to improve the classification performance, the experimental data used two kinds of corpus in Chinese and English, the R8 single label corpus of Reuters21578 (21578 classic text classification corpus of Reuters), and the Fudan corpus of Chinese data. Each corpus randomly extracted 1 800 training texts and 1 800 test texts. Experiments were carried out with KNN SVM classifier in the pre-selected dimension of 300 D, 600 D, 900 D, 1200 D, 2000 D, 3000 D, 3000 D, 3000 D, 5000 D, 5000 D and 10000 D, respectively. The results show that the binary dimension is adopted. After the fireworks algorithm optimizes the selection of feature words, Its effect is superior to the traditional feature selection method and has good stability.
【作者单位】：中山大学资讯管理学院;
【基金】：国家自然科学基金项目“面向文本分类的多学科协同建模理论与实验研究”(71373291) 广东省科技计划项目“面向主题的中文语料库构建方法与技术”(2015A030401037)
【分类号】：TP391.1

【相似文献】