一种具有保序性的带权多标记学习算法  

A weighted multi-label learning algorithm with rank preservation

作　　者：宋中山[1] 周玮瑜 孙翀[1] 艾勇[1] 刘越 SONG Zhongshan;ZHOU Weiyu;SUN Chong;AI Yong;LIU Yue(College of Computer Science,South-Central University for Nationalities,Hubei Provincial Engineering Research Center for Intelligent Management of Manufacturing Enterprises,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院,湖北省制造企业智能管理工程技术研究中心,武汉430074

出　　处：《中南民族大学学报（自然科学版）》2020年第3期309-314,共6页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：湖北省技术创新专项重大项目(2019ABA101);中央高校基本科研业务费专项资金资助项目(CZY19014)。

摘　　要：多标记学习是机器学习研究领域的热点问题之一.经典算法仅考虑了标记间的单一关系(序或权重),这使得在部分场景中多标记学习无法应用.为解决该问题,一种具有保序性的带权多标记学习算法WMLARP(Weighted Multi-label Learning Algorithm with Rank Preservation)被提出.通过在学习过程中引入“相关-无关”、“相关-相关”两种标记对来度量标记间的序和相对权重,WMLARP对基于SVM的多标记学习算法进行了扩展和优化.实验结果表明:WMLARP可充分挖掘标记间的相关性,有效提高分类模型的质量.Multi-label learning is one of the central topic in the field of machine learning.The classical algorithms considers only a single relationship(rank or weight)between labels,which makes the algorithms unable to be applied in some scenarios.To solve this problem,a new algorithm named weighted multi-label learning algorithm with rank preservation(abbrev.WMLARP)is proposed.WMLARP extends and optimizes the SVM-based multi-label learning algorithm by introducing two kinds of label pairs,which is called“related-unrelated”and“related-related”label pairs,to measure the rank and weight between labels.The experiment shows that WMLARP can mines the correlation between labels fully,improving the quality of the classification model effectively.

关 键 词：多标记学习 标记间关系 权重 序 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]