A Benchmark for Semi-Inductive Link Prediction in Knowledge Graphs

18 Oct 2023  ·  Adrian Kochsiek, Rainer Gemulla ·

Semi-inductive link prediction (LP) in knowledge graphs (KG) is the task of predicting facts for new, previously unseen entities based on context information. Although new entities can be integrated by retraining the model from scratch in principle, such an approach is infeasible for large-scale KGs, where retraining is expensive and new entities may arise frequently. In this paper, we propose and describe a large-scale benchmark to evaluate semi-inductive LP models. The benchmark is based on and extends Wikidata5M: It provides transductive, k-shot, and 0-shot LP tasks, each varying the available information from (i) only KG structure, to (ii) including textual mentions, and (iii) detailed descriptions of the entities. We report on a small study of recent approaches and found that semi-inductive LP performance is far from transductive performance on long-tail entities throughout all experiments. The benchmark provides a test bed for further research into integrating context and textual information in semi-inductive LP models.

PDF Abstract

Datasets


Introduced in the Paper:

Wikidata5M-SI

Used in the Paper:

Wikidata5M
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Inductive Link Prediction Wikidata5M-SI KGT5 (descriptions) 0-shot MRR 0.470 # 1
Inductive Link Prediction Wikidata5M-SI KGT5-context (descriptions) 0-shot MRR 0.417 # 2
1-shot MRR 0.420 # 1
10-shot MRR 0.437 # 1
Inductive Link Prediction Wikidata5M-SI SimKGC (descriptions) 0-shot MRR 0.403 # 3
Inductive Link Prediction Wikidata5M-SI DistMult + ERAvg (descriptions) 1-shot MRR 0.278 # 2
10-shot MRR 0.292 # 4
Inductive Link Prediction Wikidata5M-SI KGT5-context (mentions) 1-shot MRR 0.217 # 3
10-shot MRR 0.311 # 3
Inductive Link Prediction Wikidata5M-SI KGT5 (mentions) 0-shot MRR 0.31 # 4
Inductive Link Prediction Wikidata5M-SI SimKGC (mentions) 0-shot MRR 0.22 # 5
Inductive Link Prediction Wikidata5M-SI DistMult + ERAvg (mentions) 1-shot MRR 0.187 # 4
10-shot MRR 0.28 # 5
Inductive Link Prediction Wikidata5M-SI HittER 0-shot MRR 0.019 # 7
1-shot MRR 0.105 # 7
10-shot MRR 0.221 # 6
Inductive Link Prediction Wikidata5M-SI DistMult + ERAvg 1-shot MRR 0.171 # 5
10-shot MRR 0.333 # 2
Inductive Link Prediction Wikidata5M-SI ComplEx + Bias + Fold in 1-shot MRR 0.151 # 6
10-shot MRR 0.206 # 7
Inductive Link Prediction Wikidata5M-SI ComplEx + Bias * FoldIn 0-shot MRR 0.124 # 6

Methods


No methods listed for this paper. Add relevant methods here