Message Passing for Hyper-Relational Knowledge Graphs

TL;DR

This paper introduces StarE, a message passing encoder for hyper-relational knowledge graphs like Wikidata, which effectively models qualifiers and improves link prediction performance over existing methods.

Contribution

StarE is a novel graph encoder capable of modeling arbitrary qualifiers in hyper-relational KGs, and a new dataset WD50K addresses flaws in previous benchmarks.

Findings

StarE outperforms existing link prediction models on multiple benchmarks.

Leveraging qualifiers significantly improves link prediction accuracy, with gains up to 25 MRR points.

A new dataset WD50K is introduced to better evaluate hyper-relational KG models.

Abstract

Hyper-relational knowledge graphs (KGs) (e.g., Wikidata) enable associating additional key-value pairs along with the main triple to disambiguate, or restrict the validity of a fact. In this work, we propose a message passing based graph encoder - StarE capable of modeling such hyper-relational KGs. Unlike existing approaches, StarE can encode an arbitrary number of additional information (qualifiers) along with the main triple while keeping the semantic roles of qualifiers and triples intact. We also demonstrate that existing benchmarks for evaluating link prediction (LP) performance on hyper-relational KGs suffer from fundamental flaws and thus develop a new Wikidata-based dataset - WD50K. Our experiments demonstrate that StarE based LP model outperforms existing approaches across multiple benchmarks. We also confirm that leveraging qualifiers is vital for link prediction with gains up to 25 MRR points compared to triple-based representations.