Journal of Xidian University

Previous Articles     Next Articles

New method for multi-label feature extraction

ZHANG Jujie1;FANG Min1;GUO Jin2,3   

  1. (1. School of Computer Science and Technology, Xidian Univ., Xi'an 710071, China;
    2. School of Computer Science and Engineering, Northwestern Polytechnical Univ., Xi'an 710072, China;
    3. College of Electronical and Information Engineering, Xi'an Technological Univ., Xi'an 710021, China)
  • Received:2015-09-07 Online:2016-12-20 Published:2017-01-19


Existing multi-label feature extraction methods are limited by not fully exploiting feature information. To tackle this problem, this paper proposes a new method for multi-label feature extraction. First, it maximizes the Hilbert-Schmidt independence criterion (HSIC) between labels and the features after reducing dimensionality to exploit label information, while it minimizes the information loss using principal component analysis (PCA). Experiments across Yahoo demonstrate the effectiveness and superiority of the proposed method to PCA and 3 state-of-art multi-label feature extraction methods.

Key words: multi-label classification, feature extraction, Hilbert-Schmidt independence criterion, principal component analysis