Abstract: An instance’s label information may provide some useful information for the other instances, especially in the case of relatively scarce data. Using the relationship between the labeled and unlabeled instances can avoid the errors caused by insufficient data. Based on the research on instance correlation, an algorithm with local label information for multi-label learning is proposed. This algorithm gets the local label information of the instance firstly, and then takes them into the attribute space to build the new space. Finally, the algorithm classifies the instance based on the new space. Experimental results show that, the new algorithm outperforms the other commonly used multi-label algorithms most of the time.

Key words: multi-label learning, instance correlation, cluster, kNN

摘要： 一个样例的标记信息可能会对附近其他样例的学习提供有用信息，特别是在数据比较匮乏的情况下，利用已标记数据与未标记数据间的相关性，能够在一定程度上避免因数据不足所造成的误差。针对样例之间的相关性研究，提出基于局部标记信息的多标记学习算法，算法首先获取样例的局部标记信息，然后将样例的局部标记信息引入属性空间构造新的样例集合，并根据新的样例集合进行分类。实验结果表明，算法的分类性能得到较大提升，且优于其他常用多标记学习算法。

关键词: 多标记学习, 样例相关性, 聚类, k近邻