LLMs consistently judge value statements and news headlines more negatively in binary response formats than in continuous rating scales when simulating human respondents.
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
-
Systematic Bias in Large Language Models: Discrepant Response Patterns in Binary vs. Continuous Judgment Tasks
LLMs consistently judge value statements and news headlines more negatively in binary response formats than in continuous rating scales when simulating human respondents.