在多标签学习中,人工标注标签的主观性和不稳定性往往造成标签缺失,无法形成完备的标签空间,从而对监督学习算法的训练产生误导.标签相关性可在一定程度上弥补缺失标签对算法分类性能造成的不利影响.但缺失标签也会导致对标签相关性的估计不准确.针对该问题,提出一种增强标签相关性矩阵的不完备多标签学习(multi-label learning with incomplete labels via augmented label correlation matrix,ML-ALC)方法.首先,通过拉普拉斯映射构造数据的低维流形；然后,使用标签向量计算原始标签相关矩阵;接着,构造一个校正矩阵对原始标签相关矩阵进行增强,并通过回归系数矩阵和增强标签相关性矩阵将原始特征空间和标签空间分别映射到低维流形;最后,经过迭代学习获得优化的回归系数矩阵和增强标签相关性矩阵,并应用于多标签分类.实验结果表明,ML-ALC方法的分类性能优于其他针对缺失标签的多标签分类方法.
In multi-label learning, the subjectivity and instability of manual labeling often result in the absence of partial class labels and an incomplete label space. Missing labels are likely to be misleading in the training of supervised learning algorithms. The use of label correlation can alleviate the negative effect of missing labels on the algorithm classification performance to a certain extent. However, missing labels can lead to inaccurate estimations of label correlations. To address this problem, a method called multi-label learning with incomplete labels using an augmented label correlation matrix (ML-ALC) was proposed. First, a low-dimensional manifold of the data is constructed using Laplace mapping, and the original label correlation matrix is calculated using the label vectors. Subsequently, a correction matrix is constructed to augment the original label correlation matrix, and the original feature space and label space are mapped to the low-dimensional manifold through a regression coefficient matrix and augmented label correlation matrix, respectively. Finally, the optimized regression coefficient matrix and augmented label correlation matrix are obtained through iterative learning and applied to multi-label classification. The experimental results demonstrate that the proposed algorithm performs better than other state-of-the-art methods in multi-label learning with incomplete labels.
[1] 余鹰. 多标记学习研究综述[J]. 计算机工程与应用, 2015, 51(17): 20-27.
[2] Chen Z, Ren J. Multi-label text classification with latent word-wise label information [J]. Applied Intelligence, 2021, 51: 966-979.
[3] Boutell M R, Luo J, Shen X, et al. Learning multi-label scene classification [J]. Pattern Recognition, 2004, 37(9): 1757-1771.
[4] Qi G J, Hua X S, Yong R, et al. Correlative multi-label video annotation [C]// Proceedings of the 15th ACM International Conference on Multimedia. 2007: 17-26.
[5] Liu S M, Chen J H. A multi-label classification based approach for sentiment classification [J]. Expert Systems with Applications, 2015, 42(3): 1083-1093.
[6] Turnbull D, Barrington L, Torres D, et al. Semantic annotation and retrieval of music and sound effects [J]. IEEE Transactions on Audio, Speech and Language Processing, 2008, 16(2): 467-476.
[7] Zhang M, Zhou Z. A review on multi-label learning algorithms [J]. IEEE Transactions on Knowledge and Data Engineering, 2014, 26(8): 1819-1837.
[8] Zhang M L, Zhou Z H. ML-KNN: a lazy learning approach to multi-label learning [J]. Pattern Recognition, 2007, 40(7): 2038-2048.
[9] Furnkranz J, Hullermeier E, Mencia E L, et al. Multilabel classification via calibrated label ranking [J]. Machine Learning, 2008, 73(2): 133-153.
[10] Huang J, Li G, Wang S, et al. Multi-label classification by exploiting local positive and negative pairwise label correlation [J]. Neurocomputing, 2017, 257: 164-174.
[11] 亢浏越, 黄睿, 孙广玲. 基于类属特征的多标签流形学习分类方法[J]. 上海大学学报(自然科学版), 2021, 27(3): 525-534.
[12] Read J, Pfahringer B, Holmes G, et al. Classifier chains for multi-label classification [J]. Machine Learning, 2011, 85(3): 333-359.
[13] Kai W S, Chong H L, Wang J. Multilabel classification via co-evolutionary multilabel hypernetwork [J]. IEEE Transactions on Knowledge and Data Engineering, 2016, 28(9): 2438-2451.
[14] Wu B, Lyu S, Hu B G, et al. Multi-label learning with missing labels for image annotation and facial action unit recognition [J]. Pattern Recognition, 2015, 48(7): 2279-2289.
[15] Hao X Y, Huang J, Qin F, et al. Multi-label learning with missing features and labels and its application to text categorization [J]. Intelligent Systems with Applications, 2022, 14: 200086.
[16] Zhang L, Cheng Y, Wang Y, et al. Feature-label dual-mapping for missing label-specific features learning [J]. Soft Computing, 2021, 25(14): 9307-9323.
[17] Zhu Y, Kwok J T, Zhou Z H. Multi-label learning with global and local label correlation [J]. IEEE Transactions on Knowledge and Data Engineering, 2018, 30(6): 1081-1094.
[18] Huang J, Qin F, Zheng X, et al. Improving multi-label classification with missing labels by learning label-specific features [J]. Information Sciences, 2019, 492: 124-146.
[19] Cheng Z, Zeng Z. Joint label-specific features and label correlation for multi-label learning with missing label [J]. Applied Intelligence, 2020, 50(11): 4029-4049.
[20] Kumar S, Rastogi R. Low rank label subspace transformation for multi-label learning with missing labels [J]. Information Sciences, 2022, 596: 53-72.
[21] Wang Y, Zheng W, Cheng Y, et al. Two-level label recovery-based label embedding for multi-label classification with missing labels [J]. Applied Soft Computing, 2021, 99(6): 106868.
[22] Belkin M, Niyogi P. Laplacian eigenmaps and spectral techniques for embedding and clustering [J]. Advances in Neural Information Processing Systems, 2001, 14(6): 585-591.
[23] Beck A, Teboulle M. Fast gradient-based algorithms for constrained total variation image denoising and deblurring problems [J]. IEEE Transactions on Image Processing, 2009, 18(11): 2419-2434.
[24] Gibaja E L, Ventura S. A tutorial on multi-label learning [J]. ACM Computing Surveys, 2015, 47(3): 1-38.
[25] Tsoumakas G, Katakis I, Vlahavas I. Mining multi-label data [M]//Maimon O, Rokach L. Data mining and knowledge discovery handbook, Boston: Springer, 2010: 667-685.
[26] Huang J, Li G, Huang Q, et al. Learning label-specific features and class-dependent labels for multi-label classification [J]. IEEE Transactions on Knowledge and Data Engineering, 2016, 28(12): 3309-3323.
[27] Demiar J, Schuurmans D. Statistical comparisons of classifiers over multiple data sets [J]. Journal of Machine Learning Research, 2006, 7(1): 1-30.
