Multi-label Learning By exploiting Correlations of Label Subsets

Abstract

Multi-label learning studies exist in a wide range of diverse scenes such as text processing, image mining, emotion analysis, etc. Feature selection technologies are proposed to be a vital factor in the scene of multi-label learning, which can relieve the influence of curse of dimensionality, enhance the classification accuracy, and reduce the time of consumption of learning process. At present, a great number of multi-label feature selection algorithms are proposed by the researchers who are focused on the machine learning or other research fields. It is demonstrated that considering the label correlation when the method choices the importance feature subset will improve the multi-label learning methods performance. On this account, a method that based on correlations of label is proposed in this work.