ResearchBenchmarksConversational Derailment Forecasting on CGA-CMV Large (randomly selected sample)Follow70Accuracy (Acc)Human61.88863.99466.168.206May 28, 2026Evaluation ResultsMethodMethodLinksAccuracy (Acc)False Positive Rate (FPR)Precision (P)Recall (R)F1 Score (F1)H Metric (H)HumanRound=2Round=22026.057015.675.955.663.92.1SOTAaggregation=averaged a...aggregation=averaged across 5 random seeds2026.057036.267.976.271.83.3HumanRound=1Round=12026.0562.224.467.848.954.62.6