A relative labeling importance estimation algorithm based on global-local label correlations for multi-label learning
method
同时考虑全局和局部标签相关性来估计相对标签重要性。
objective function
y
i
′
\text{y}_i'
yi′表示相对标记重要性矩阵的第
i
i
i行,
Y
i
′
Y_i'
Yi′表示相对标记重要性矩阵的第
i
i
i列,
∣
y
i
∣
|\text{y}_i|
∣yi∣表示
x
i
x_i
xi的相关标签数,
G
=
[
g
i
l
]
n
×
l
G=[g_{il}]_{n\times l}
G=[gil]n×l,其中当
y
i
l
=
1
\text{y}_{il}=1
yil=1时,
g
i
l
=
∣
y
i
∣
g_{il}=|\text{y}_i|
gil=∣yi∣,当
y
i
l
=
−
1
\text{y}_{il}=-1
yil=−1时,
g
i
l
=
−
1
g_{il}=-1
gil=−1
Construction of matrix S S S
Construction of matrix C C C and C ′ C' C′