Abstract: Dimensionality reduction is an important and challenging task in multi-label learning. Feature selection is a highly efficient technique for dimensionality reduction by maintaining maximum relevant information to find an optimal feature subset. First of all, this paper proposes a multi-label feature selection method based on non-negative sparse representation by studying the subspace learning. This method can be treated as a matrix factorization problem, which is combined with non-negative constraint problem and  L_2,1 norm minimization problem. Then, this paper designs a kind of efficient iterative update algorithm to tackle the new problem and proves its convergence. Finally, the experimental results on six real-world data sets show the effectiveness of the proposed algorithm.

Key words: multi-label learning, feature selection, non-negative matrix factorization;L_2,1-norm

摘要： 在多标签学习中，数据降维是一项重要而又具有挑战性的任务。特征选择是一种高效的数据降维技术，它通过保持最大相关信息选取一个特征子集。通过对子空间学习的研究，提出了基于非负稀疏表示的多标签特征选择方法。该方法可以看成是矩阵分解问题，其融合了非负约束问题和L_2,1-范数最小优化问题。设计了一种高效的矩阵更新迭代算法求解新问题，并证明其收敛性。最后，对6个实际的数据集进行了测试，实验结果证明了算法的有效性。

关键词: 多标签学习, 特征选择, 非负矩阵分解, L_2, 1-范数