ResearchBenchmarksToxicity Detection on Chinese Toxicity Detection Emoji perturbation (test)Follow93.85Detection RateCA-CoT72.675678.172883.6789.1672May 30, 2025Evaluation ResultsMethodMethodLinksDetection RateCA-CoTBackbone=GPT-4o-mini,...Backbone=GPT-4o-mini, Prompting=Chinese-Aware Chain-of-Thought2025.0593.85ICLBackbone=GPT-4o-mini,...Backbone=GPT-4o-mini, Prompting=In-Context Learning2025.0588.36No ICLBackbone=GPT-4o-mini,...Backbone=GPT-4o-mini, Prompting=Zero-shot (No ICL)2025.0573.49