notesum.ai

Published at October 22

Context-aware Inductive Knowledge Graph Completion with Latent Type Constraints and Subgraph Reasoning

cs.CL

cs.AI

cs.IR

Released Date: October 22, 2024

Authors: Muzhi Li¹, Cehao Yang¹, Chengjin Xu¹, Zixing Song², Xuhui Jiang¹, Jian Guo¹, Ho-fung Leung², Irwin King²

Aff.: ¹IDEA Research, International Digital Economy Academy; ²Dept. of Computer Science and Engineering, The Chinese University of Hong Kong

Arxiv: https://arxiv.org/abs/2410.16803v1

Refer to caption

		Transductive			Inductive
Metric	Method	WN18RR	FB15k-237	NELL-995	WN18RR	FB15k-237	NELL-995
MRR	RuleN	0.669	0.674	0.736	0.780	0.462	0.710
	GRAIL	0.676	0.597	0.727	0.799	0.469	0.675
	MINERVA	0.656	0.572	0.592	-	-	-
	TuckER	0.646	0.682	0.800	-	-	-
	KG-BERT	-	-	-	0.547	0.500	0.419
	BERTRL	0.683	0.695	0.781	0.792	0.605	0.808
	KRST	0.899	0.720	0.800	0.890	0.716	0.769
	APST	0.902	0.774	0.801	0.908	0.764	0.769
	CATS (TAR)	0.956	0.812	0.836	0.937	0.834	0.750
	CATS (SR)	0.972	0.829	0.869	0.992	0.875	0.906
	CATS (full)	0.978	0.843	0.885	0.982	0.882	0.861
Hits@1	RuleN	0.646	0.603	0.636	0.745	0.415	0.638
	GRAIL	0.644	0.494	0.615	0.769	0.390	0.554
	MINERVA	0.632	0.534	0.553	-	-	-
	TuckER	0.600	0.615	0.729	-	-	-
	KG-BERT	-	-	-	0.436	0.341	0.244
	BERTRL	0.655	0.620	0.686	0.755	0.541	0.715
	KRST	0.835	0.639	0.694	0.809	0.600	0.649
	APST	0.839	0.694	0.698	0.837	0.643	0.663
	CATS (TAR)	0.922	0.726	0.745	0.888	0.744	0.624
	CATS (SR)	0.951	0.752	0.792	0.984	0.804	0.849
	CATS (full)	0.962	0.776	0.820	0.965	0.805	0.783