基于地标选择与标签相关性的类属特征多标签学习OA

The real world contained a large number of objects that could be assigned multiple labels simultaneously.Being a crucial component of machine learning,multi-label learning algorithms had significant advantages in dealing with these complex semantic entities.Most of the existing multi-label learning algorithms directly predicted all labels.However,in practical applications,due to the redundancy of the label space,it was quite difficult to predict all the labels at the same time,so it was necessary to select a representative subset of labels as landmarks.Landmarks were correlated with other labels,which was crucial to improve the performance of the model.Based on this,this paper proposed a multi-label learning algorithm of label-specific features based on landmark selection and label correlation.Firstly,a linear regression model for multi-label learning was established and the label-specific features were obtained by using Li-norm.Then,the representative landmarks were selected by introducing a landmark selection matrix with soft constraints.Finally,the label correlation between landmarks and other labels was mined to further improve the performance of multi-label learning algorithms.Experiments demonstrated that the proposed algorithm competes favorably with other advanced multi-label learning algorithms.