Multi-label learning with label-specific features by resolving label correlations

• Published in: Knowledge-based systems 2018-11, Vol.159, p.148-157
• Main Authors: Zhang, Jia, Li, Candong, Cao, Donglin, Lin, Yaojin, Su, Songzhi, Dai, Liang, Li, Shaozi
• Format: Article
• Language: English
• Subjects: Chinese medicine; Correlation; Correlation analysis; Datasets; Design optimization; Label correlations; Label-specific features; Labeling; Labels; Learning; Multi-label learning; Optimization; Optimization framework; Performance enhancement; Traditional Chinese medicine; Weight
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2018.07.003

•We propose to learn label-specific features using sparsity regularized optimization in multi-label setting, which cover the information of label correlations.•We model this multi-label learning problem by an optimization framework in which the weights of features and label correlations-based features are defined as two sets of unknown variables, and introduce a iterative optimization method to update these unknown variables.•Label correlations are represented by additional features generated in the optimization process, and a KNN-like method is designed to obtain label correlations-based features of test data.•Extensive experiments demonstrate the advantages of our proposed algorithm. In addition, two real-world data sets on TCM are collected, and our proposed algorithm is further validated on these two data sets in terms of the identification of health-state in TCM. In multi-label learning, different labels may have their own inherent characteristics for distinguishing each other, in the meanwhile, exploiting the correlations among labels is another practical yet challenging task to improve the performance. In this work, we present a new method for the joint learning of label-specific features and label correlations. The key is the design of an optimization framework to learn the weight assignment scheme of features, and the correlations among labels are taken into account by constructing additional features at the same time. Through iteratively optimizing the two sets of unknown variables, which are referred to feature weights and label correlations-based features, label-specific features of each label are available to achieve multi-label classification. Comprehensive experiments on various multi-label data sets including two collected traditional Chinese medicine data sets reveal the advantages of our proposed algorithm.