一种基于增量式超网络的多标签分类方法

在层次多标签分类问题中，一个样本同时被赋予多个类别标签，并且这些类别标签被组织成一定的层次结构。层次多标签分类问题的主要挑战在于：①分类方法的输出必须符合标签的层次结构约束；②层次深的节点所代表的标签往往只有很少的样本与之相关，造成标签不平衡的问题。提出一种用于层次多标签分类问题的增量式超网络学习方法(hierarchical multi-label classification using incremental hypernetwork, HMC-IMLHN)，通过将超网络的超边组织成相应的层次结构，使输出的预测标签能够满足标签的层次约束。此外，超网络学习方法可以利用标签之间的关联减少标签不平衡问题对分类性能的影响。实验结果表明，与其他层次多标签分类方法相比，提出的增量式超网络方法能够取得较好的分类准确性。

Hierarchical multi-label classification using incremental hypernetwork

In the hierarchical multi-label classification, a sample can be associated with multiple class labels residing on a hierarchy. The main challenges of hierarchical multi-label classification lie in the following two aspects: 1) The predictions must meet the label hierarchy constraints; 2) Labels at lower levels usually have few related samples, which leads to the imbalanced problem among the labels. In this paper, we propose a Hierarchical Multi-label Classification using Incremental Hypernetwork (HMC-IMLHN). By organizing hyperedges of hypernetwork into corresponding hierarchy, the prediction of hypernetwork can meet label hierarchy automatically. Furthermore, the imbalance problem among labels can be mitigated by utilizing correlations among labels. The results of experimental studies demonstrate that our proposed method can achieve competitive classification performance when compared with many other existing methods.