REFIND at SemEval-2025 Task 3: Retrieval-Augmented Factuality Hallucination Detection in Large Language Models

Donggeon Lee, Hwanjo Yu


Abstract
REFIND is a retrieval-augmented framework for detecting hallucinated spans in LLM outputs by leveraging retrieved documents. It introduces Context Sensitivity Ratio, a metric quantifying LLM sensitivity to evidence. REFIND outperforms baselines across nine languages, including low-resource settings, achieving superior hallucination detection accuracy. These results demonstrate the effectiveness of context sensitivity quantification in improving hallucination detection.
Anthology ID:
2025.semeval-1.2
Volume:
Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025)
Month:
July
Year:
2025
Address:
Vienna, Austria
Editors:
Sara Rosenthal, Aiala Rosá, Debanjan Ghosh, Marcos Zampieri
Venues:
SemEval | WS
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
7–15
Language:
URL:
https://preview.aclanthology.org/transition-to-people-yaml/2025.semeval-1.2/
DOI:
Bibkey:
Cite (ACL):
Donggeon Lee and Hwanjo Yu. 2025. REFIND at SemEval-2025 Task 3: Retrieval-Augmented Factuality Hallucination Detection in Large Language Models. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 7–15, Vienna, Austria. Association for Computational Linguistics.
Cite (Informal):
REFIND at SemEval-2025 Task 3: Retrieval-Augmented Factuality Hallucination Detection in Large Language Models (Lee & Yu, SemEval 2025)
Copy Citation:
PDF:
https://preview.aclanthology.org/transition-to-people-yaml/2025.semeval-1.2.pdf