Salvato in:
Dettagli Bibliografici
Autori principali: Huang, Zhaoheng, Dou, Zhicheng, Zhu, Yutao, Wen, Ji-rong
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2402.14690
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913240308514816
author Huang, Zhaoheng
Dou, Zhicheng
Zhu, Yutao
Wen, Ji-rong
author_facet Huang, Zhaoheng
Dou, Zhicheng
Zhu, Yutao
Wen, Ji-rong
contents Large language models (LLMs) may generate text that lacks consistency with human knowledge, leading to factual inaccuracies or \textit{hallucination}. Existing research for evaluating the factuality of LLMs involves extracting fact claims using an LLM and verifying them against a predefined fact source. However, these evaluation metrics are task-specific, and not scalable, and the substitutability of fact sources in different tasks is under-explored. To address these challenges, we categorize four available fact sources: human-written evidence, reference documents, search engine results, and LLM knowledge, along with five text generation tasks containing six representative datasets. Then, we propose \texttt{UFO}, an LLM-based unified and flexible evaluation framework to verify facts against plug-and-play fact sources. We implement five evaluation scenarios based on this framework. Experimental results show that for most QA tasks, human-written evidence and reference documents are crucial, and they can substitute for each other in retrieval-augmented QA tasks. In news fact generation tasks, search engine results and LLM knowledge are essential. Our dataset and code are available at \url{https://github.com/WaldenRUC/UFO}.
format Preprint
id arxiv_https___arxiv_org_abs_2402_14690
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle UFO: a Unified and Flexible Framework for Evaluating Factuality of Large Language Models
Huang, Zhaoheng
Dou, Zhicheng
Zhu, Yutao
Wen, Ji-rong
Computation and Language
Large language models (LLMs) may generate text that lacks consistency with human knowledge, leading to factual inaccuracies or \textit{hallucination}. Existing research for evaluating the factuality of LLMs involves extracting fact claims using an LLM and verifying them against a predefined fact source. However, these evaluation metrics are task-specific, and not scalable, and the substitutability of fact sources in different tasks is under-explored. To address these challenges, we categorize four available fact sources: human-written evidence, reference documents, search engine results, and LLM knowledge, along with five text generation tasks containing six representative datasets. Then, we propose \texttt{UFO}, an LLM-based unified and flexible evaluation framework to verify facts against plug-and-play fact sources. We implement five evaluation scenarios based on this framework. Experimental results show that for most QA tasks, human-written evidence and reference documents are crucial, and they can substitute for each other in retrieval-augmented QA tasks. In news fact generation tasks, search engine results and LLM knowledge are essential. Our dataset and code are available at \url{https://github.com/WaldenRUC/UFO}.
title UFO: a Unified and Flexible Framework for Evaluating Factuality of Large Language Models
topic Computation and Language
url https://arxiv.org/abs/2402.14690