基于信息增益和萤火虫算法的文本特征选择  

Text feature selection based on information gain and firefly algorithm

作　　者：文武[1,3] 赵成 赵学华 刘颖 范荣妹 WEN Wu;ZHAO Cheng;ZHAO Xue-hua;LIU Ying;FAN Rong-mei(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Chongqing Institute of Quality and Standardization,Chongqing 400023,China;Chongqing Information Technology Designing Limited Company,Chongqing 401121,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]重庆市质量和标准化研究院,重庆400023 [3]重庆信科设计有限公司,重庆401121

出　　处：《计算机工程与设计》2019年第12期3457-3462,共6页Computer Engineering and Design

摘　　要：针对传统的文本特征选择方法所选特征子集精度不高的问题,提出一种结合信息增益和萤火虫算法的特征选择方法。计算所有特征词的信息增益并按从高到低排序,在排序靠前的特征集合上利用萤火虫算法寻优搜索得到最优特征子集。为避免萤火虫算法求解速度慢和陷入局部最优,对步长因子α进行改进,使其动态更新。在KNN上的文本分类实验结果表明,萤火虫算法选择的特征子集与信息增益和遗传算法相比,具有更好的分类效果。Aiming at the problem of low precision of feature subset selected using traditional text feature selection methods,a feature selection method combining information gain and firefly algorithm was proposed.The information gain of all feature words was computed and sorted from high to low.The optimal feature subset was obtained by searching the feature set using the firefly algorithm.To avoid slow solving speed and falling into local optimum,the step factor was improved to update dynamically.Experimental results on KNN show that the feature subset selected using firefly algorithm has better classification effects than information gain and genetic algorithm.

关 键 词：信息增益 萤火虫算法 特征选择 步长因子 文本分类 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]