基于Tabu的Deep Web特征选择算法  

Feature selection of deep web based on Tabu

作　　者：谭春亮[1] 甘丹[1] 陈丽娜[1] 蒋运承[1] 

机构地区：[1]广西师范大学计算机学院,广西桂林541004,中山大学计算机科学系,广东广州510275

出　　处：《计算机工程与设计》2008年第13期3358-3361,3473,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(60663001);中国博士后科学基金项目(20060400226);广西青年科学基金项目(桂科青0640030)

摘　　要：Deep Web分类的小样本、高维特征的特点限制了分类算法的选择,影响分类器的设计和准确度,降低了分类器的"泛化"能力,出现分类器"过拟合",所以需要进行特征选择,降低特征的维数,避免"维数灾难"。目前,没有Deep Web特征选择自动算法的相关研究。通过对Deep Web分类的特征选择进行研究,提出了基于类别可分性判据和Tabu搜索的特征选择算法,可以在2的时间复杂度内得到次优的特征子集,减小了分类器设计的难度,提高了分类器分类准确率。根据特征选择前后的特征集,利用KNN分类算法进行Deep Web分类,结果表明提高了分类器的分类准确率,降低了分类算法的时间复杂度。Classification of deep web has characteristic of small sample and high dimensional, which restricts choice of classification algorithm and makes the classifier hardly design, also lower the ＂ Generalization Ability＂ and makes the classifier ＂ overfitting ＂. Feature selection is necessary to avoid ＂curse of dimensionality＂. There is no research about automatic classification algorithm at present. Feature selection algorithm of deep web based on Tabu search algorithm and separative criterion is put forward through the research about feature selection, which can quickly fmd feature subset in the time complexity of O（N^2）. Feature selection algorithm based on Tabu and separative criterion makes design of classifier easily, also increases accuracy of classifier. Experimentation based on classifier indicates that feature selection algorithm based on Tabu and separative criterion increases accuracy of classifier and reduces computation complexity.

关 键 词：特征选择 TABU搜索算法 深层网络 信息检索 分类算法 分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]