Abstract: Aiming at the effective utilization of label specific features and label correlation, a new multi-label algorithm LSFLC is proposed, which can effectively integrate label specific features and label correlation. Firstly, for each label, new positive class instances are generated by re-sampling technology to expand the number of positive class instances. Secondly, the original feature space is transformed into a specific feature space by feature mapping function, and the label specific feature set of each label is obtained. Then, for each label, its most relevant label is found to expand its specific feature set by copying the positive class instances of the label, which not only enriches the information of the label, but also solves the problem of class imbalance to a certain extent. Finally, experiments on different data sets show that the classification effect of the proposed algorithm is better.

Key words: multi-label learning, local label correlation, label-specific features, related instances insertion

摘要： 针对标签特有特征和标签相关性的有效利用，提出了一种新的多标签算法LSFLC，它可以有效地集成标签特有特征和标签相关性。首先，对于每个标签，通过重采样技术生成新的正类实例以扩充其正类实例的数目；其次，通过特征映射函数将原始特征空间转换为特定的特征空间，得到每个标签的标签特征集；然后，对于每个标签，找到与其最相关标签，通过复制该标签的正类实例来扩大标签特征集，这不仅丰富了标签的信息，而且在一定程度上改善了类不平衡的问题；最后，对于不同的数据集进行实验分析，实验结果表明该算法的分类效果更好。

关键词: 多标签学习, 局部标签相关性, 标签特有特征, 相关实例补充