Country and language personas degrade LLM hate speech detection F1 scores, and a custom reweighted fine-tuning loss reduces the degradation for Llama and Nemo, but less for Phi.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Personalisation or Prejudice? Addressing Geographic Bias in Hate Speech Detection using Debias Tuning in Large Language Models
Country and language personas degrade LLM hate speech detection F1 scores, and a custom reweighted fine-tuning loss reduces the degradation for Llama and Nemo, but less for Phi.