ResearchBenchmarksToxicity Classification on VOICED item-level (binary)Follow80AccuracyDiADEM77.836878.398478.9679.5216Apr 9, 2026Evaluation ResultsMethodMethodLinksAccuracyF1 (Macro)F1 (Weighted)Cohen's Kappa (κ)MCCJensen-Shannon Divergence (JSD)Mean Deviation (MD)Error Rate (ER)Expected Calibration Error (ECE)DiADEM2026.048055.7474.3816.924.112.520.719.111.44DisCo2026.0478.3557.3174.3517.5120.323.3859.120.892.6LeWiDi2026.0477.9245.9169.571.523.394.2962.7121.034.79