Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ReaLitE: Enrichment of Relation Embeddings in Knowledge Graphs using Numeric Literals

Antonis Klironomos, Baifan Zhou, Zhuoxun Zheng, Gad-Elrab Mohamed, Heiko Paulheim, Evgeny Kharlamov

TL;DR

ReaLitE introduces a relation-centric enrichment of knowledge graph embeddings by aggregating numeric literals from head and tail entities per relation and fusing these summaries into the relation embeddings. The approach decouples literal processing from the base KGE model, offering linear or gated fusion variants and a flexible aggregation mechanism (including a learnable combination of aggregations). Empirical results show ReaLitE improves link prediction and node classification across multiple datasets and base models, with the largest gains for long-tail relations and numerically correlated literals. The method demonstrates strong compatibility with existing KGE frameworks and highlights the value of leveraging numeric attributes to capture nuanced relational patterns in real-world graphs.

Abstract

Most knowledge graph embedding (KGE) methods tailored for link prediction focus on the entities and relations in the graph, giving little attention to other literal values, which might encode important information. Therefore, some literal-aware KGE models attempt to either integrate numerical values into the embeddings of the entities or convert these numerics into entities during preprocessing, leading to information loss. Other methods concerned with creating relation-specific numerical features assume completeness of numerical data, which does not apply to real-world graphs. In this work, we propose ReaLitE, a novel relation-centric KGE model that dynamically aggregates and merges entities' numerical attributes with the embeddings of the connecting relations. ReaLitE is designed to complement existing conventional KGE methods while supporting multiple variations for numerical aggregations, including a learnable method. We comprehensively evaluated the proposed relation-centric embedding using several benchmarks for link prediction and node classification tasks. The results showed the superiority of ReaLitE over the state of the art in both tasks.

ReaLitE: Enrichment of Relation Embeddings in Knowledge Graphs using Numeric Literals

TL;DR

ReaLitE introduces a relation-centric enrichment of knowledge graph embeddings by aggregating numeric literals from head and tail entities per relation and fusing these summaries into the relation embeddings. The approach decouples literal processing from the base KGE model, offering linear or gated fusion variants and a flexible aggregation mechanism (including a learnable combination of aggregations). Empirical results show ReaLitE improves link prediction and node classification across multiple datasets and base models, with the largest gains for long-tail relations and numerically correlated literals. The method demonstrates strong compatibility with existing KGE frameworks and highlights the value of leveraging numeric attributes to capture nuanced relational patterns in real-world graphs.

Abstract

Most knowledge graph embedding (KGE) methods tailored for link prediction focus on the entities and relations in the graph, giving little attention to other literal values, which might encode important information. Therefore, some literal-aware KGE models attempt to either integrate numerical values into the embeddings of the entities or convert these numerics into entities during preprocessing, leading to information loss. Other methods concerned with creating relation-specific numerical features assume completeness of numerical data, which does not apply to real-world graphs. In this work, we propose ReaLitE, a novel relation-centric KGE model that dynamically aggregates and merges entities' numerical attributes with the embeddings of the connecting relations. ReaLitE is designed to complement existing conventional KGE methods while supporting multiple variations for numerical aggregations, including a learnable method. We comprehensively evaluated the proposed relation-centric embedding using several benchmarks for link prediction and node classification tasks. The results showed the superiority of ReaLitE over the state of the art in both tasks.

Paper Structure

This paper contains 17 sections, 6 equations, 2 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Description
  4. Method: ReaLitE
  5. Aggregation of Numeric Literals
  6. Fusing Literals with Relation Embeddings
  7. Model Complexity
  8. Experimental Setup
  9. Datasets
  10. Models
  11. Training
  12. Evaluation Settings
  13. Results
  14. Link Prediction Results
  15. Node Classification Results
  16. ...and 2 more sections

Figures (2)

  • Figure 1: An example showing the potential benefit of the mathematical relationship between people's income and houses' rent during link prediction.
  • Figure 2: ReaLitE's pipeline for relation embedding enrichment (left) given a triple to be scored (right). ReaLitE aggregates the head and tail numeric literals (Step 1) and combines the aggregated literals with the relation embedding (Step 2).