偏多标记分类学习

Partial Multi-Label Classification Learning

摘要: 在偏多标记学习中，以往研究者常将偏多标记分类和标记消歧过程独立进行，并且未考虑标记相关性的先验知识难以事先获得。为此，本文提出了偏多标记分类和标记消歧联合学习的算法(JPMLLD)。JPMLLD建立一个从样本特征到标记的映射；其次通过将观测标记矩阵分解的方式进行消歧；最后建立了一个偏多标记分类和标记消歧统一的联合学习框架。大量实验结果验证了本文提出的JPMLLD算法的有效性。

Abstract: In partial multi-label learning, researchers used to separate partial multi-label learning classification and label disambiguation processes, and prior knowledge without considering label correlation is difficult to obtain in advance. In this paper, we propose a joint learning method of partial multiple label classification and label disambiguation (JPMLLD). First, a mapping from sample features to labels is established. Secondly, the disambiguation is carried out by the decomposition of observation label matrix. Finally, a unified learning framework for partial multiple label classification and label disambiguation is established. The experimental results justify the effectiveness of the proposed JPMLLD algorithm.