Reproduce “Something Just Like TRuST : Toxicity Recognition of Span and Target” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Berk Atil, Namrata Sureddy, Rebecca J. Passonneau
TRuST creates a shared benchmark for toxicity detection, target prediction, and toxic-span identification by re-annotating five existing datasets under one framework. It includes 298,255 examples, with human annotations for 11,498; most labels are machine-assigned. On its human-annotated test set, fine-tuned RoBERTa reaches 0.79 accuracy and 0.79 F1, tying o4-mini on F1. The benchmark reflects the authors’ toxicity definition and setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.