《计算机应用研究》|Application Research of Computers

基于标签相关性的类属属性多标签分类算法

Label-correlation based multi-label classification algorithm with label-specific features

免费全文下载 (已被下载 次)  
获取PDF全文
作者 牟甲鹏,蔡剑,余孟池,徐建
机构 南京理工大学 计算机科学与工程学院,南京 210094
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)09-018-2656-03
DOI 10.19734/j.issn.1001-3695.2019.04.0118
摘要 多标签学习中一个样本可同时属于多个类别标签,每个标签都可能拥有反映该标签特定特点的特征,即类属属性,目前已经出现了基于类属属性的多标签分类算法LIFT。针对LIFT算法中未考虑标签之间相互关系的问题,提出一种基于标签相关性的类属属性多标签分类算法CLLIFT。该算法使用标签距离度量标签之间的相关性,通过在类属属性空间附加相关标签的方式完成标签相关性的引入,以达到提升分类性能的目的。在四个多标签数据集上的实验结果表明,所提算法与LIFT算法相比在多个多标签评价指标上平均提升21.1%。
关键词 标签相关性; 类属属性; 多标签学习
基金项目 国家自然科学基金资助项目(61872186,61802205,91846104)
本文URL http://www.arocmag.com/article/01-2020-09-018.html
英文标题 Label-correlation based multi-label classification algorithm with label-specific features
作者英文名 Mu Jiapeng, Cai Jian, Yu Mengchi, Xu Jian
机构英文名 School of Computer Science & Engineering,Nanjing University of Science & Technology,Nanjing 210094,China
英文摘要 In multi-label learning, each instance can belong to multiple category labels simultaneously. Each label may have a feature that reflects the specific characteristics of the label, which could label-specific feature, and there propose multi-label learning algorithm with label-specific features named LIFT. Aiming at the problem that LIFT algorithm ignored the relevance among labels, this paper proposed a label-correlation based multi-label classification algorithm with label-specific features named CLLIFT. The CLLIFT algorithm used label-distance to measure the correlation between labels and introduced the label correlation by appending related labels to the label-specific features space. The experimental results on four multi-label datasets show that the proposed algorithm has an average increase of 21.1% over multiple multi-label metrics indicators compared with the LIFT algorithm.
英文关键词 label correlation; label-specific features; multi-label learning
参考文献 查看稿件参考文献
 
收稿日期 2019/4/30
修回日期 2019/6/10
页码 2656-2658,2673
中图分类号 TP301.6
文献标志码 A