notesum.ai

Published at November 18

ZeFaV: Boosting Large Language Models for Zero-shot Fact Verification

cs.CL

cs.AI

Released Date: November 18, 2024

Authors: Son T. Luu¹, Hiep Nguyen², Trung Vo², Le-Minh Nguyen²

Aff.: ¹VNU University of Information Technology, Ho Chi Minh City, Vietnam; ²Japan Advanced Institute of Science and Technology, Ishikawa, Japan

Arxiv: http://arxiv.org/abs/2411.11247v1

Refer to caption

	HoVer (2-hop)	HoVer (3-hop)	HoVer (4-hop)	FEVEROUS-S
Number of claims	1,126	1,835	1,039	2,962
ZeFaV (w context)	77.85	70.61	67.47	86.74
ZeFaV (w/o context)	71.31	67.41	61.62	75.97
ProgramFC [14]	76.53	70.84	66.88	86.34
QACheck [13] (InstructGPT)	55.67	54.67	52.35	59.47
InfoRE [5] (llama2-70B)	52.40	51.21	50.07	67.96