Ways of asking, structuring, and steering that measurably work better.
2 stories
Techniques
A proposed explainable framework combines DistilBERT, BiLSTM, and attention with LIME to detect hate speech in binary and multi-class settings, achieving high F1 scores on two datasets.
Why it matters: Provides a more transparent and robust approach to hate speech detection across binary and multi-class tasks.
Worth testing Results are preliminary and from a single study; further validation needed.
arXiv — cs.CL · 2d ago
Techniques
Researchers introduce a systematic approach for tag-aware structured text translation using LLMs, featuring a hybrid data synthesis strategy, multi-task fine-tuning, and multi-objective alignment, improving both fluency and tag fidelity across multiple language directions.
