针对多标签分类问题,建立一种考虑标签间相关性的多标签分类模型.首先,对属性值为数值型的多标签数据集,建立基于类和属性依赖度的离散化MLAIM(Multi-Label Attribute Inter-dependence Maximization)算法.其次,通过对标签集进行贝叶斯网结构学习,得到每个标签的父节点,提出标签相关的多标签分类模型,即MLLD(Multi-Label Classification algorithm based on Lael Dependency)算法,并给出MLLD算法的具体过程.通过数值实验将MLLD算法与二元关联(BR)算法等4种算法进行比较,结果表明MLLD算法分类效果更好.
Multi-label classification algorithm based on label dependency
For the multi-label classification problem ,a multi-label classification model based on label dependency is established .Firstly ,MLAIM discrete algorithm is proposed for multi-label dataset with attribute value as numerical ,in order to label dependency of multi-label classifica-tion .Secondly ,based on the structure learning of Bayesian networks on the label sets ,the par-ent of each label is obtained .Moreover ,the multi-label classification models and algorithms are given .Finally the efficiency of MLLD algorithms is proved by some numerical experiments .
国家科技期刊平台
NSTL
万方数据
维普
