Papers › Relation Prediction as an Auxiliary Training Objective for Improving Multi-Relational...

Relation Prediction as an Auxiliary Training Objective for Improving Multi-Relational Graph Representations

6 Oct 2021AKBC 2021 10arXiv:2110.02834archive 2025-07-28

Yihong Chen, Pasquale Minervini, Sebastian Riedel, Pontus Stenetorp

Learning good representations on multi-relational graphs is essential to knowledge base completion (KBC). In this paper, we propose a new self-supervised training objective for multi-relational graph representation learning, via simply incorporating relation prediction into the commonly used 1vsAll objective. The new training objective contains not only terms for predicting the subject and object of a given triple, but also a term for predicting the relation type. We analyse how this new objective impacts multi-relational learning in KBC: experiments on a variety of datasets and models show that relation prediction can significantly improve entity ranking, the most widely used evaluation task for KBC, yielding a 6.1% increase in MRR and 9.9% increase in Hits@1 on FB15k-237 as well as a 3.1% increase in MRR and 3.4% in Hits@1 on Aristo-v4. Moreover, we observe that the proposed objective is especially effective on highly multi-relational datasets, i.e. datasets with a large number of predicates, and generates better representations when larger embedding sizes are used.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

facebookresearch/ssl-relation-prediction officialmentioned in paperpytorchNOASSERTION report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Graph Representation LearningKnowledge Base CompletionKnowledge Graph CompletionKnowledge Graph EmbeddingKnowledge Graph EmbeddingsLink PredictionLink Property PredictionRelation PredictionRelational Reasoning

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Link Prediction Aristo-v4 ComplEx-N3-RP Hits@1 0.24 #1 of 1 Archive leaderboard report
Link Prediction Aristo-v4 ComplEx-N3-RP Hits@10 0.447 #1 of 1 Archive leaderboard report
Link Prediction Aristo-v4 ComplEx-N3-RP Hits@3 0.336 #1 of 1 Archive leaderboard report
Link Prediction Aristo-v4 ComplEx-N3-RP MRR 0.311 #1 of 1 Archive leaderboard report
Link Prediction CoDEx Large ComplEx-N3-RP Hits@1 0.277 #1 of 6 Archive leaderboard report
Link Prediction CoDEx Large ComplEx-N3-RP Hits@10 0.473 #1 of 6 Archive leaderboard report
Link Prediction CoDEx Large ComplEx-N3-RP Hits@3 0.377 #1 of 6 Archive leaderboard report
Link Prediction CoDEx Large ComplEx-N3-RP MRR 0.345 #1 of 6 Archive leaderboard report
Link Prediction CoDEx Medium ComplEx-N3-RP Hits@1 0.277 #2 of 7 Archive leaderboard report
Link Prediction CoDEx Medium ComplEx-N3-RP Hits@10 0.490 #2 of 7 Archive leaderboard report
Link Prediction CoDEx Medium ComplEx-N3-RP Hits@3 0.386 #2 of 7 Archive leaderboard report
Link Prediction CoDEx Medium ComplEx-N3-RP MRR 0.352 #2 of 7 Archive leaderboard report
Link Prediction CoDEx Small ComplEx-N3-RP Hits@1 0.375 #1 of 6 Archive leaderboard report
Link Prediction CoDEx Small ComplEx-N3-RP Hits@10 0.663 #1 of 6 Archive leaderboard report
Link Prediction CoDEx Small ComplEx-N3-RP Hits@3 0.514 #1 of 6 Archive leaderboard report
Link Prediction CoDEx Small ComplEx-N3-RP MRR 0.473 #1 of 6 Archive leaderboard report
Link Prediction FB15k-237 ComplEx-N3-RP Hits@1 0.298 #5 of 75 Archive leaderboard report
Link Prediction FB15k-237 ComplEx-N3-RP Hits@10 0.568 #5 of 75 Archive leaderboard report
Link Prediction FB15k-237 ComplEx-N3-RP Hits@3 0.424 #5 of 75 Archive leaderboard report
Link Prediction FB15k-237 ComplEx-N3-RP MR 163 #5 of 75 Archive leaderboard report
Link Prediction FB15k-237 ComplEx-N3-RP MRR 0.389 #5 of 75 Archive leaderboard report
Link Prediction FB15k-237 TuckER-RP Hits@1 0.264 #26 of 75 Archive leaderboard report
Link Prediction FB15k-237 TuckER-RP Hits@10 0.535 #26 of 75 Archive leaderboard report
Link Prediction FB15k-237 TuckER-RP Hits@3 0.388 #26 of 75 Archive leaderboard report
Link Prediction FB15k-237 TuckER-RP MRR 0.354 #26 of 75 Archive leaderboard report
Link Prediction FB15k-237 CP-N3-RP Hits@10 0.55 #59 of 75 Archive leaderboard report
Link Prediction FB15k-237 CP-N3-RP MRR 0.366 #59 of 75 Archive leaderboard report
Link Prediction WN18RR ComplEx-N3-RP Hits@1 0.443 #29 of 75 Archive leaderboard report
Link Prediction WN18RR ComplEx-N3-RP Hits@10 0.578 #29 of 75 Archive leaderboard report
Link Prediction WN18RR ComplEx-N3-RP Hits@3 0.505 #29 of 75 Archive leaderboard report
Link Prediction WN18RR ComplEx-N3-RP MRR 0.488 #29 of 75 Archive leaderboard report
Link Property Prediction ogbl-biokg ComplEx-N3-RP Number of params 187750000 #7 of 16 Archive leaderboard report
Link Property Prediction ogbl-biokg ComplEx-N3-RP Test MRR 0.8494 #7 of 16 Archive leaderboard report
Link Property Prediction ogbl-biokg ComplEx-N3-RP Validation MRR 0.8497 #7 of 16 Archive leaderboard report
Link Property Prediction ogbl-wikikg2 ComplEx-N3-RP (100dim) Number of params 500334800 #13 of 30 Archive leaderboard report
Link Property Prediction ogbl-wikikg2 ComplEx-N3-RP (100dim) Test MRR 0.6481 #13 of 30 Archive leaderboard report
Link Property Prediction ogbl-wikikg2 ComplEx-N3-RP (100dim) Validation MRR 0.6701 #13 of 30 Archive leaderboard report
Link Property Prediction ogbl-wikikg2 ComplEx-N3-RP (50dim) Number of params 250167400 #15 of 30 Archive leaderboard report
Link Property Prediction ogbl-wikikg2 ComplEx-N3-RP (50dim) Test MRR 0.6364 #15 of 30 Archive leaderboard report
Link Property Prediction ogbl-wikikg2 ComplEx-N3-RP (50dim) Validation MRR 0.6594 #15 of 30 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: CP-N3-RP, ComplEx-N3-RP, RESCAL-RP, TuckER-RP

CP-N3CP-N3-RPComplEx-N3ComplEx-N3-RPRESCALRESCAL-RPTuckERTuckER-RP

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections