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 | | LEWL | MLML | LRML | CPLST | BR |
---|
Emotions | 0.3 | 0.8820.009 | 0.9050.006 | 0.8570.048 | 0.8470.013 | 0.8470.013 | 0.7 | 0.7750.010 | 0.8010.009 | 0.8140.035 | 0.6660.011 | 0.6660.011 | Yeast | 0.3 | 0.8830.007 | 0.8740.006 | 0.8050.015 | 0.8500.003 | 0.850.003 | 0.7 | 0.7490.008 | 0.7380.007 | 0.7170.023 | 0.6600.002 | 0.660.002 | CAL500 | 0.3 | 0.8450.004 | 0.7760.005 | 0.7550.021 | 0.8490.008 | 0.8490.008 | 0.7 | 0.6670.007 | 0.6330.007 | 0.6430.020 | 0.6570.012 | 0.6570.012 | Medical | 0.3 | 0.8040.053 | 0.7830.029 | 0.7550.005 | 0.7960.049 | 0.7960.049 | 0.7 | 0.5660.042 | 0.5930.029 | 0.5650.006 | 0.5570.038 | 0.5570.038 | Langlog | 0.3 | 0.8590.002 | 0.8560.002 | 0.8550.041 | 0.850.001 | 0.850.001 | 0.7 | 0.6850.003 | 0.6830.003 | 0.6820.066 | 0.6630.005 | 0.6630.005 | Enron | 0.3 | 0.8450.013 | 0.8220.006 | 0.7540.016 | 0.8220.015 | 0.8220.015 | 0.7 | 0.6330.033 | 0.6480.028 | 0.630.013 | 0.6510.014 | 0.6510.014 | Corel5k | 0.3 | 0.8170.012 | 0.7850.007 | 0.8010.025 | 0.8130.017 | 0.8130.017 | 0.7 | 0.6190.011 | 0.6050.013 | 0.6140.021 | 0.6120.016 | 0.6120.016 |
|
|
|