KG-NSF: Knowledge Graph Completion with a Negative-Sample-Free Approach

TL;DR

KG-NSF introduces a negative-sampling-free framework for knowledge graph completion that uses cross-correlation matrices, achieving comparable performance to traditional methods with faster convergence.

Contribution

It presents a novel negative sampling-free approach for KG embedding learning, reducing computational complexity and bias.

Findings

Achieves similar link prediction accuracy as negative sampling methods.

Converges significantly faster during training.

Reduces bias associated with the closed world assumption.

Abstract

Knowledge Graph (KG) completion is an important task that greatly benefits knowledge discovery in many fields (e.g. biomedical research). In recent years, learning KG embeddings to perform this task has received considerable attention. Despite the success of KG embedding methods, they predominantly use negative sampling, resulting in increased computational complexity as well as biased predictions due to the closed world assumption. To overcome these limitations, we propose \textbf{KG-NSF}, a negative sampling-free framework for learning KG embeddings based on the cross-correlation matrices of embedding vectors. It is shown that the proposed method achieves comparable link prediction performance to negative sampling-based methods while converging much faster.