WN18

18-relation knowledge-graph completion Knowledge graph link prediction on WordNet (full, 18 relations). Inputs are directed subject-relation-object triples with learned embeddings. Inverse relations make this legacy split highly vulnerable to leakage. Evaluated by MRR.

Dataset details

Task

Knowledge graph link prediction on WordNet (full, 18 relations). The primary catalog metric is MRR.

Data

Knowledge graph link prediction on WordNet (full, 18 relations).

Features

The raw input is a set of directed subject-relation-object triples. Entity and relation embeddings are learned inputs; reciprocal edges, inverse relations, and filtered candidate sets are evaluation or preprocessing choices.

Splits and evaluation

The catalog records these protocols or variants: Standard split. Evaluation uses MRR (higher is better). Exact masks or folds must come from the cited release.

Quirks and pitfalls

Filtered ranking removes other known true triples but the graph is incomplete, so sampled corruptions can be false negatives. WN18 and FB15k contain inverse-relation leakage; use WN18RR or FB15k-237 when evaluating relational generalization.

Sources

SOTA Progression

Target

new SOTA matches SOTA at that time other proposed result other baseline result

Leaderboard

#	Model	MRR ↑	Hit@10	Paper
1	Inverse Model code ↗ Trad baseline	0.9630	—	AAAI 2017 Knowledge Graphs code ↗ AAAI 2017 Jul 2017 Jul '17
2	RotatE code ↗ GNN ★ proposed	0.9490 ± 0.0000	—	ICLR 2018 RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space code ↗ ICLR 2018 Sep 2018 Sep '18
3	ConvE code ↗ GNN ★ proposed	0.9430	—	AAAI 2017 Knowledge Graphs code ↗ AAAI 2017 Jul 2017 Jul '17
4	ANALOGY code ↗ baseline	0.9420	—	AAAI 2017 Knowledge Graphs code ↗ AAAI 2017 Jul 2017 Jul '17
5	ComplEx GNN baseline	0.9410	—	Modeling Relational Data with Graph Convolutional Networks no code Mar 2017 Mar '17
6	HolE code ↗ GNN baseline	0.9380	—	ICML 2016 Complex Embeddings for Simple Link Prediction code ↗ ICML 2016 Jun 2016 Jun '16
7	LinkFeat Trad baseline	0.9380	—	Modeling Relational Data with Graph Convolutional Networks no code Mar 2017 Mar '17
8	DistMult code ↗ GNN baseline	0.8220	—	ICML 2016 Complex Embeddings for Simple Link Prediction code ↗ ICML 2016 Jun 2016 Jun '16
9	R-GCN GNN ★ proposed	0.8190	—	Modeling Relational Data with Graph Convolutional Networks no code Mar 2017 Mar '17
10	R-GCN code ↗ GNN baseline	0.8140	—	AAAI 2017 Knowledge Graphs code ↗ AAAI 2017 Jul 2017 Jul '17
11	DistMult GNN baseline	0.8130	—	Modeling Relational Data with Graph Convolutional Networks no code Mar 2017 Mar '17
12	TripleRE code ↗ Trad ★ proposed	0.7725	0.9541	TripleRE: Knowledge Graph Embeddings via Tripled Relation Vectors code ↗ Sep 2022 Sep '22
13	TransAt code ↗ Trad baseline	0.6446	0.9510	TripleRE: Knowledge Graph Embeddings via Tripled Relation Vectors code ↗ Sep 2022 Sep '22
14	TransE code ↗ GNN baseline	0.4540	—	ICML 2016 Complex Embeddings for Simple Link Prediction code ↗ ICML 2016 Jun 2016 Jun '16
15	CP code ↗ GNN baseline	0.0750	—	ICML 2016 Complex Embeddings for Simple Link Prediction code ↗ ICML 2016 Jun 2016 Jun '16
16	CP GNN baseline	0.0580	—	Modeling Relational Data with Graph Convolutional Networks no code Mar 2017 Mar '17

SOTA performance set a new SOTA at the time matched the SOTA at the time

Structure

Coverage