基于标记集合划分的多标记分类算法

A Multi- Label Classification Algorithm Based on Label Set Partition

在多标记学习中,标记之间往往既不是完全独立也不是完全排斥的,因此在构建多标记分类器时要充分利用标记之间的依赖关系.目前利用标记间关系的方法有将标记集合划分为子集和将各标记间关系表示为链式两种.本文提出了一种结合上述两种思想的算法,首先根据标记对间的依赖度量值来启发式地对标记集合进行划分,然后在最终的划分子集合间依次建立分类器组成分类器链.通过与其他算法的比较,实验结果表明该算法能提升分类器性能.

Labels in multi -label learning are usually neither completely independent nor mutually exclusive, therefore it is important to explore the dependencies among labels during building multi-label classifiers.There exist two methods to exploit label dependencies：dividing the label set into several sets and regarding the structure of labels as a chain.In this paper,a new multi-label classification algorithm,which combines the above two approaches,is proposed.First,it divides the label set into several subsets based on label dependencies heuristically.It builds thereafter a classifier on each subset.All these classifiers comprise a chain.It is shown through the experiment that this algorithm outperforms other algorithms and improves multi-label classification performance.