Skip to main content

Table 2 Performance metrics ± standard deviations (Maximum mean columnar values in bold)

From: Tackling racial bias in automated online hate detection: Towards fair and accurate detection of hateful users with geometric deep learning

Model

Accuracy

Precision (%)

Recall (%)

F1-Score (%)

AUC (%)

LR

83.1 ± 1.78

37.6 ± 1.16

82.1 ± 1.32

51.6 ± 1.54

89.9 ± 1.20

SVM

81.6 ± 0.89

34.8 ± 0.77

77.6 ± 0.13

48.0 ± 0.64

87.6 ± 0.41

\(\mathit{ANN}_{\mathrm{text}+\mathrm{user}}\)

84.1 ± 2.58

40.6 ± 2.39

75.7 ± 2.14

51.7 ± 3.21

88.2 ± 2.45

\(\mathit{ANN}_{\mathrm{text}+\mathrm{user}+\mathrm{network}}\)

87.2 ± 2.02

45.3 ± 2.47

68.8 ± 2.05

54.2 ± 3.08

87.2 ± 1.72

\(\mathit{GraphSAGE}_{\mathrm{maxpool}}\)

84.2 ± 1.59

40.3 ± 1.46

80.9 ± 1.08

53.4 ± 1.16

90.2 ± 1.34

\(\mathit{GraphSAGE}_{\mathrm{attention}}\)

84.8 ± 1.11

40.6 ± 1.25

82.3 ± 1.18

54.3 ± 1.14

90.8 ± 1.07

\(\mathit{GraphSAGE}_{\mathrm{meanagg}}\)

87.4 ± 1.04

46.1 ± 1.15

76.8 ± 1.11

57.5 ± 1.12

90.8 ± 1.03