摘要
针对现有的嵌入式多标签特征选择方法只能分析有标签样本,无法利用大量“廉价”的无标签样本信息的问题,提出一种基于流形学习与L_(2,1)范数的无监督多标签特征选择方法。该算法在L_(2,1)范数回归的基础上,用特征流形和数据相似矩阵约束特征权重矩阵和伪标签矩阵,从而达到特征选择的目的。实验结果表明,所提方法的各指标性能优于SCLS、MDMR等特征选择方法,充分体现所提算法的可行性。
Embedded multi-label feature selection method is a kind of feature selection method with good performance and it is widely used at present.However,the existing methods can only analyze labeled samples and cannot take advantage of a large number of"cheap"unlabeled samples.Therefore,an unsupervised multi-label feature selection method based on manifold learning and L_(2,1)norm(UMLFS)was proposed.On the basis of L_(2,1)norm regression,the feature manifold and data similarity matrix were used to constrain the feature weight matrix and pseudo label matrix,so as to achieve the purpose of feature selection.Experimental results show that the performance of the proposed method is better than that of SCLS,MDMR and other feature selection methods in most indexes,which fully reflects the feasibility of the proposed algorithm.
作者
马盈仓
张要
张宁
朱恒东
MA Yingcang;ZHANG Yao;ZHANG Ning;ZHU Hengdong(School of Science,Xi’an Polytechnic University,Xi’an 710600,China;Basic Education Department,Shandong Huayu University of Technology,Dezhou 253034,Shandong,China)
出处
《纺织高校基础科学学报》
CAS
2021年第3期102-111,120,共11页
Basic Sciences Journal of Textile Universities
基金
国家自然科学基金(61976130)
陕西省重点研发计划项目(2018KW-021)
陕西省自然科学基金(2020JQ-923)。
