基于标签正负相关性的多标签类属特征学习  

Multi-label learning with label-specific features based on positive and negative label correlation

作　　者：黄睿[1] 亢浏越 HUANG Rui;KANG Liu-yue(School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《计算机工程与设计》2021年第5期1271-1277,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61671283);上海市自然科学基金项目(16ZR1411100)。

摘　　要：提出一种基于标签正负相关性的多标签类属特征学习方法(multi-label learning with label-specific features based on positive and negative label correlation,LIFTPNL)。基于k近邻的思想构建全局和局部的标签信息矩阵,根据此矩阵计算成对标签的正负相关性;对每个类别标签,基于属于相同和不同类簇的样本构建连接矩阵,联合该标签正负相关性计算样本相似度;采用谱聚类方法获得聚类中心,将原有特征转换成类属特征;通过二分类器得到分类结果。实验结果表明,所提算法优于多种多标签分类方法。A multi-label classification method named multi-label learning with label-specific features based on positive and negative label correlation(LIFTPNL)was proposed.A global and local label information matrix was constructed based on the concept of k-nearest neighbors(kNN),and the positive and negative pairwise label correlations were calculated according to the matrix.For each class label,the adjacency matrix was constructed by samples from the same and different clusters and the positive and negative label correlations were used to calculate the sample similarities.The clustering centers were obtained by spectral clustering and applied to convert the original features into label-specific features.The final classification results were obtained by a series of binary classifiers.Experimental results show that the proposed method outperforms several multi-label classification methods.

关 键 词：多标签学习 分类 类属特征 标签正负相关性 聚类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]