多标记数据的特征及标记降维方法综述

A Review of Features and Labels Dimensionality Reduction Methods of Multi-Label Data

多标记学习降维方法在实际应用问题中用以处理特征,标记或二者维度较高的数据集,已成为研究热点;针对目前多标记学习降维方法数量众多,种类繁杂而导致缺乏科学分类的问题,从多标记数据降维空间选择的角度,提出将多标记学习降维方法按照特征空间降维,标记空间降维和二者均降维的形式归纳为三类,其中特征空间降维又分为特征降维和特征选择两类问题,分别从独立于和依赖于彼此空间的角度对已有的40余篇文献中的典型多标记学习降维算法的研究现状进行了综述;最后,总结了多标记学习降维方法的研究现状和启示,并提出了未来进一步的研究方向。

Multi-label learning dimensionality reduction method has been a research hotspot in the practical application problem to deal with data sets with higher features dimensions,labels dimensions or both dimensions.In view of the large number of multi-label learning dimensionality reduction methods and the lack of scientific classification,from the perspective of the dimension reduction space selection of multi-label data,a multi-label learning dimension reduction method is proposed to be classified into three types according to feature space dimension reduction,label space dimension reduction and both.The feature space dimension reduction is divided into two categories:feature dimension reduction and feature selection.They are based on the independent and dependent space of each other.The research status of typical multi-label learning dimensionality reduction algorithms is summarized.Finally,the research status and inspiration of multi-label learning dimensionality reduction methods are reviewed,and further research directions are proposed for the future.