Graph-based semi-supervised multi-label learning method

Abstract

The problem of multi-label classification has attracted great interest in the last decade. However, most multi-label learning methods only focus on supervised settings, and can not effectively make use of relatively inexpensive and easily obtained large number of unlabeled samples. To solve this problem, we put forward a novel graph-based semi-supervised multi-label learning method, called GSMM. GSMM characterize the inherent correlations among multiple labels by Hilbert-Schmidt independence criterion. It's expected to derive the optimal assignment of class membership to unlabeled samples by maximizing the consistency of class label correlations and simultaneously as smooth as possible on sample feature graph. The experiments comparing GSMM to the state-of-the-art multi-label learning approaches on several real-world datasets show GSMM can effectively learn from the labeled and unlabeled samples. Especially when the labeled is relatively rare, it can improve the performance greatly.