Embedding Method for Knowledge Graph with Densely Defined Ontology

TL;DR

Knowledge graphs are often incomplete, and existing KG embedding methods underutilize ontology structure. The authors propose TransU, which unifies subjects, properties, and objects into a single set $E$ with $E_2 \subset E_1 \subset E$ and enforces that properties used as entities share the same vector, preserving compatibility with existing KGE methods through aligned dimensionality. Evaluations on FB15K and a property-rich speckled string dataset show that integrating TransU with methods like ComplEx can improve link prediction on ontology-rich graphs, though gains are dataset-dependent. The work highlights the importance of ontological richness for embedding quality and points to future directions in generalizability, evaluation rigor, nuanced property relations, data-type handling, and scalability.

Abstract

Knowledge graph embedding (KGE) is a technique that enhances knowledge graphs by addressing incompleteness and improving knowledge retrieval. A limitation of the existing KGE models is their underutilization of ontologies, specifically the relationships between properties. This study proposes a KGE model, TransU, designed for knowledge graphs with well-defined ontologies that incorporate relationships between properties. The model treats properties as a subset of entities, enabling a unified representation. We present experimental results using a standard dataset and a practical dataset.

Figures (4)

• Figure 1: bad example of learning
• Figure 2: Example expressed as a two-dimensional vector
• Figure 3: Expected vector representation
• Figure 4: Initialisation algorithm for vector representation