Skip to main content

Table 3 Fairness Evaluation Metrics (African–American denoted as AA, False positive denoted as FP)

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

Model

Accuracy

FPs

FPs

FP rate

FP rate

(AA users)

(AA users)

(non-AA users)

(AA users)

(non-AA users)

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

80.9%

26

660

20.3%

14.8%

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

86.0%

13

465

10.2%

10.5%

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

91.7%

5

680

3.9%

15.7%

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

93.9%

1

662

0.7%

15.3%

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

94.2%

0

492

0.0%

11.1%