结合标签相关性和不均衡性的多标签学习模型

针对现有多标签学习算法较少兼顾标签间关联性和不平衡性的问题,提出一种同时考虑多标签间相关性与多标签不平衡问题的学习模型（A Multi-label Learning Model based on Label Correlation and Imbalance,MLCI）. 该学习模型针对每个标签类别,通过耦合其他标签类别以考量标签间的关联性,并降低缓解标签间不均衡比率,MLCI是一个将当前标签的二类不平衡学习器和多个与其他标签耦合的多类不平衡学习器结合的集成分类器. 采用7种常用的多标签算法作为对比算法,针对yeast、scene、emotions和CAL500这4个开放数据集进行分类处理. 实验结果表明,MLCI相比其他对比算法,在精度均值（Average- Precision ）、排序损失（Ranking-Loss）、宏观平均AUC（Macro-Averaging AUC）和微观平均AUC（Micro-Averaging AUC）4个性能评估指标上总体占明显优势.
Since the existing multi-label learning algorithms pay less attention to the problem of correlation and imbalance between label sets, this paper proposes a multi-label learning model based on label correlation and imbalance (MLCI). By coupling other label categories to consider the correlation among labels and reducing the imbalance ratio between labels of instances, the learning model is for each label category, and it is an ensemble classifier that combines the current class of binary imbalance classifier with multiple imbalanced classifiers coupled to other labels. In this paper, seven commonly used multi-label algorithms are used as comparison algorithms to classify the four open datasets of yeast, scene, emotions and CAL500. The experimental results show that the MLCI has obvious advantages in accuracy precision, ranking-Loss, macro-averaging AUC and micro-averaging AUC.