PAIR, a post-stratification-based replication method, improves calibration of NLP models trained on non-representative annotator pools in simulations, but the effect is not uniform for rare hate-speech labels.
Berinsky, Gregory A
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ME 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Aligning NLP Models with Target Population Perspectives using PAIR: Population-Aligned Instance Replication
PAIR, a post-stratification-based replication method, improves calibration of NLP models trained on non-representative annotator pools in simulations, but the effect is not uniform for rare hate-speech labels.