Research on the MuIti-LabeI Learning ModeI Based on Shared Subspaces
In the multi-label classification problems, multiple labels share the same input space, but there are some correlations between different instances of the same label, so in the study of such problems, correlation studies between labels become particularly important. Existing multi-label learning for relevant research between the labels are on the original data, and we hope to re-represent the original data, high-level semantic information extracted from high-dimensional data, mapping from the original input space to a low-dimensional subspace, in the case of class standard reflects the characteristics of the information as a guide to share information between class standard. Then uses the existing multi-label classification method to classify. Multiples Web classification tasks, experimental results show that this method is to some extent improve the classification results.
NETL
NSTL
万方数据
维普
