Duality-Induced Regularizer for Tensor Factorization Based Knowledge Graph Completion

TL;DR

This paper introduces DURA, a novel regularizer for tensor factorization in knowledge graph completion, addressing overfitting issues and improving model performance across various methods.

Contribution

The paper proposes DURA, a new regularizer based on duality principles, which enhances tensor factorization models for knowledge graph completion and is widely applicable.

Findings

DURA significantly improves performance on benchmark datasets.

DURA is effective across multiple tensor factorization models.

Experiments demonstrate consistent and substantial gains.

Abstract

Tensor factorization based models have shown great power in knowledge graph completion (KGC). However, their performance usually suffers from the overfitting problem seriously. This motivates various regularizers -- such as the squared Frobenius norm and tensor nuclear norm regularizers -- while the limited applicability significantly limits their practical usage. To address this challenge, we propose a novel regularizer -- namely, DUality-induced RegulArizer (DURA) -- which is not only effective in improving the performance of existing models but widely applicable to various methods. The major novelty of DURA is based on the observation that, for an existing tensor factorization based KGC model (primal), there is often another distance based KGC model (dual) closely associated with it. Experiments show that DURA yields consistent and significant improvements on benchmarks.