Conservative Bias in Large Language Models: Measuring Relation Predictions

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Aguda, Toyin, Wilson, Erik, Anzagira, Allan, Kaur, Simerjot, Smiley, Charese
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910998146842624
author Aguda, Toyin
Wilson, Erik
Anzagira, Allan
Kaur, Simerjot
Smiley, Charese
author_facet Aguda, Toyin
Wilson, Erik
Anzagira, Allan
Kaur, Simerjot
Smiley, Charese
contents 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 significant information loss when reasoning is not explicitly included in the output. We systematically evaluate this trade-off across multiple prompts, datasets, and relation types, introducing the concept of Hobson's choice to capture scenarios where models opt for safe but uninformative labels over hallucinated ones. Our findings suggest that conservative bias occurs twice as often as hallucination. To quantify this effect, we use SBERT and LLM prompts to capture the semantic similarity between conservative bias behaviors in constrained prompts and labels generated from semi-constrained and open-ended prompts.
format Preprint
id arxiv_https___arxiv_org_abs_2506_08120
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Conservative Bias in Large Language Models: Measuring Relation Predictions
Aguda, Toyin
Wilson, Erik
Anzagira, Allan
Kaur, Simerjot
Smiley, Charese
Computation and Language
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 significant information loss when reasoning is not explicitly included in the output. We systematically evaluate this trade-off across multiple prompts, datasets, and relation types, introducing the concept of Hobson's choice to capture scenarios where models opt for safe but uninformative labels over hallucinated ones. Our findings suggest that conservative bias occurs twice as often as hallucination. To quantify this effect, we use SBERT and LLM prompts to capture the semantic similarity between conservative bias behaviors in constrained prompts and labels generated from semi-constrained and open-ended prompts.
title Conservative Bias in Large Language Models: Measuring Relation Predictions
topic Computation and Language
url https://arxiv.org/abs/2506.08120