基于对偶流形重排序的无监督特征选择算法  

Unsupervised Feature Selection Algorithm Based on Dual Manifold Re-ranking

作　　者：梁云辉 甘舰文 陈艳 周芃 杜亮[1,2] LIANG Yunhui;GAN Jianwen;CHEN Yan;ZHOU Peng;DU Liang(College of Computer and Information Technology,Shanxi University,Taiyuan 030006,China;Institute of Big Data Science and Industry,Shanxi University,Taiyuan 030006,China;College of Computer,Sichuan University,Chendu 610065,China;College of Computer Science and Technology,Anhui University,Hefei 230601,China)

机构地区：[1]山西大学计算机与信息技术学院,太原030006 [2]山西大学大数据科学与产业研究院,太原030006 [3]四川大学计算机学院,成都610065 [4]安徽大学计算机科学与技术学院,合肥230601

出　　处：《计算机科学》2023年第7期72-81,共10页Computer Science

基　　金：国家自然科学基金面上项目(61976129,62176001)。

摘　　要：在许多数据分析任务中,经常会遇到高维数据。特征选择技术旨在从原始高维数据中找到最具代表性的特征,但由于缺乏类标签信息,相比有监督场景,在无监督学习场景中选择合适的特征困难得多。传统的无监督特征选择方法通常依据某些准则对样本的特征进行评分,在这个过程中样本是被无差别看待的。然而这样做并不能完全捕捉数据的内在结构,不同样本的重要性应该是有差异的,并且样本权重与特征权重之间存在一种对偶关系,它们会互相影响。为此,提出了一种基于对偶流形重排序的无监督特征选择算法(Unsupervised Feature Selection Algorithm based on Dual Manifold Re-Ranking, DMRR),分别构建不同的相似性矩阵来刻画样本与样本、特征与特征、样本与特征的流形结构,并结合样本与特征的初始得分进行流形上的重排序。将DMRR与3种原始无监督特征选择算法以及2种无监督特征选择后处理算法进行比较,实验结果表明样本重要性信息、样本与特征之间的对偶关系有助于实现更优的特征选择。High dimensional data is often encountered in many data analysis tasks.Feature selection techniques aim to find the most representative features from the original high-dimensional data.Due to the lack of class label information,it is much more difficult to select suitable features in unsupervised learning scenarios than in supervised scenarios.Traditional unsupervised feature selection methods usually score the features of samples according to certain criteria in which samples are treated indiscriminately.However,these approaches cannot capture the internal structure of data completely.The importance of different samples should vary.There is a dual relationship between weight of sample and feature that will influence each other.Therefore,an unsupervised feature selection algorithm based on dual manifold re-ranking(DMRR)is proposed in this paper.Different similarity matrices are constructed to depict the manifold structures on samples and samples,features and features,and samples and features respectively.Then manifold re-ranking is carried out by combining the initial scores of samples and features.By comparing DMRR with three original unsupervised feature selection algorithms and two unsupervised feature selection post-processing algorithms,experimental results verify that importance information of different samples and the dual relationship between sample and feature are helpful to achieve better feature selection.

关 键 词：对偶 流形学习 重排序 特征选择 无监督学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]