2025
Conservative Bias in Large Language Models: Measuring Relation Predictions
ACL 2025finding
Large language models (LLMs) exhibit pronounced conservative bias in relation extraction tasks, frequently defaulting to no_relation label when an appropriate option is unavailable. While this behavior helps prevent incorrect relation assignments, our analysis reveals that it also leads to significa…