结合局部标记序关系的弱监督标记分布学习

标记分布学习（label distribution learning，LDL）是一种用于解决标记多义性的新颖学习范式。现有的LDL方法大多基于完整数据信息进行设计，然而由于高昂的标注成本以及标注人员水平的局限性，很难获取到完整标注数据信息，且会导致传统LDL算法性能的下降。为此，本文提出了一种新型的结合局部序标记关系的弱监督标记分布学习算法，通过维持尚未缺失标记之间的相对关系，并利用标记相关性来恢复缺失的标记，在数据标注不完整的情况下提升算法性能。在14个数据集上进行了大量的实验来验证算法的有效性。

Weakly supervised label distribution learning by maintaining local label ranking

Label distribution learning (LDL) is a novel learning paradigm for solving labeling polysemy. Most existing LDL methods are designed based on complete data information; however, because of high labeling costs and the limitation of labelers’ level, complete labeling data information is difficult to obtain, which leads to performance degradation in traditional LDL algorithms. In this paper, we propose a novel weakly supervised LDL by maintaining a local label ranking (WSLDL-MLLR) algorithm. We improve algorithm performance under incomplete data labeling by maintaining relative relationships between the not-yet-missing labels and using label correlation to recover missing labels. Extensive experiments conducted on 14 datasets verified the effectiveness of the algorithm.