一种基于嵌入式的弱标记分类算法

一种基于嵌入式的弱标记分类算法

李亚重,杨有龙,仇海全

Label embedding for weak label classification

Yachong Li,Youlong Yang,Haiquan Qiu

表3 缺失标签在Macro F₁上的恢复结果

Table 3 Recovery results for missing labels on Macro F₁

Data set	$ρ$	LEWL	MLML	LRML	CPLST	BR
Emotions	0.3	0.882 $\pm$ 0.009	0.905 $\pm$ 0.006	0.857 $\pm$ 0.048	0.847 $\pm$ 0.013	0.847 $\pm$ 0.013
Emotions	0.7	0.775 $\pm$ 0.010	0.801 $\pm$ 0.009	0.814 $\pm$ 0.035	0.666 $\pm$ 0.011	0.666 $\pm$ 0.011
Yeast	0.3	0.883 $\pm$ 0.007	0.874 $\pm$ 0.006	0.805 $\pm$ 0.015	0.850 $\pm$ 0.003	0.85 $\pm$ 0.003
Yeast	0.7	0.749 $\pm$ 0.008	0.738 $\pm$ 0.007	0.717 $\pm$ 0.023	0.660 $\pm$ 0.002	0.66 $\pm$ 0.002
CAL500	0.3	0.845 $\pm$ 0.004	0.776 $\pm$ 0.005	0.755 $\pm$ 0.021	0.849 $\pm$ 0.008	0.849 $\pm$ 0.008
CAL500	0.7	0.667 $\pm$ 0.007	0.633 $\pm$ 0.007	0.643 $\pm$ 0.020	0.657 $\pm$ 0.012	0.657 $\pm$ 0.012
Medical	0.3	0.804 $\pm$ 0.053	0.783 $\pm$ 0.029	0.755 $\pm$ 0.005	0.796 $\pm$ 0.049	0.796 $\pm$ 0.049
Medical	0.7	0.566 $\pm$ 0.042	0.593 $\pm$ 0.029	0.565 $\pm$ 0.006	0.557 $\pm$ 0.038	0.557 $\pm$ 0.038
Langlog	0.3	0.859 $\pm$ 0.002	0.856 $\pm$ 0.002	0.855 $\pm$ 0.041	0.85 $\pm$ 0.001	0.85 $\pm$ 0.001
Langlog	0.7	0.685 $\pm$ 0.003	0.683 $\pm$ 0.003	0.682 $\pm$ 0.066	0.663 $\pm$ 0.005	0.663 $\pm$ 0.005
Enron	0.3	0.845 $\pm$ 0.013	0.822 $\pm$ 0.006	0.754 $\pm$ 0.016	0.822 $\pm$ 0.015	0.822 $\pm$ 0.015
Enron	0.7	0.633 $\pm$ 0.033	0.648 $\pm$ 0.028	0.63 $\pm$ 0.013	0.651 $\pm$ 0.014	0.651 $\pm$ 0.014
Corel5k	0.3	0.817 $\pm$ 0.012	0.785 $\pm$ 0.007	0.801 $\pm$ 0.025	0.813 $\pm$ 0.017	0.813 $\pm$ 0.017
Corel5k	0.7	0.619 $\pm$ 0.011	0.605 $\pm$ 0.013	0.614 $\pm$ 0.021	0.612 $\pm$ 0.016	0.612 $\pm$ 0.016