ComparisonQA: Evaluating Factuality Robustness of LLMs Through Knowledge Frequency Control and Uncertainty

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Zong, Qing, Wang, Zhaowei, Zheng, Tianshi, Ren, Xiyu, Song, Yangqiu
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913856762150912
author Zong, Qing
Wang, Zhaowei
Zheng, Tianshi
Ren, Xiyu
Song, Yangqiu
author_facet Zong, Qing
Wang, Zhaowei
Zheng, Tianshi
Ren, Xiyu
Song, Yangqiu
contents The rapid development of LLMs has sparked extensive research into their factual knowledge. Current works find that LLMs fall short on questions around low-frequency entities. However, such proofs are unreliable since the questions can differ not only in entity frequency but also in difficulty themselves. So we introduce ComparisonQA benchmark, containing 283K abstract questions, each instantiated by a pair of high-frequency and low-frequency entities. It ensures a controllable comparison to study the role of knowledge frequency in the performance of LLMs. Because the difference between such a pair is only the entity with different frequencies. In addition, we use both correctness and uncertainty to develop a two-round method to evaluate LLMs' knowledge robustness. It aims to avoid possible semantic shortcuts which is a serious problem of current QA study. Experiments reveal that LLMs, including GPT-4o, exhibit particularly low robustness regarding low-frequency knowledge. Besides, we find that uncertainty can be used to effectively identify high-quality and shortcut-free questions while maintaining the data size. Based on this, we propose an automatic method to select such questions to form a subset called ComparisonQA-Hard, containing only hard low-frequency questions.
format Preprint
id arxiv_https___arxiv_org_abs_2412_20251
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle ComparisonQA: Evaluating Factuality Robustness of LLMs Through Knowledge Frequency Control and Uncertainty
Zong, Qing
Wang, Zhaowei
Zheng, Tianshi
Ren, Xiyu
Song, Yangqiu
Computation and Language
The rapid development of LLMs has sparked extensive research into their factual knowledge. Current works find that LLMs fall short on questions around low-frequency entities. However, such proofs are unreliable since the questions can differ not only in entity frequency but also in difficulty themselves. So we introduce ComparisonQA benchmark, containing 283K abstract questions, each instantiated by a pair of high-frequency and low-frequency entities. It ensures a controllable comparison to study the role of knowledge frequency in the performance of LLMs. Because the difference between such a pair is only the entity with different frequencies. In addition, we use both correctness and uncertainty to develop a two-round method to evaluate LLMs' knowledge robustness. It aims to avoid possible semantic shortcuts which is a serious problem of current QA study. Experiments reveal that LLMs, including GPT-4o, exhibit particularly low robustness regarding low-frequency knowledge. Besides, we find that uncertainty can be used to effectively identify high-quality and shortcut-free questions while maintaining the data size. Based on this, we propose an automatic method to select such questions to form a subset called ComparisonQA-Hard, containing only hard low-frequency questions.
title ComparisonQA: Evaluating Factuality Robustness of LLMs Through Knowledge Frequency Control and Uncertainty
topic Computation and Language
url https://arxiv.org/abs/2412.20251