MLMLM: Link Prediction with Mean Likelihood Masked Language Model

TL;DR

MLMLM introduces a novel approach using masked language models for link prediction in knowledge bases, achieving state-of-the-art results and better interpretability compared to traditional methods.

Contribution

The paper proposes MLMLM, a new method that leverages mean likelihood in MLMs for scalable and interpretable link prediction in knowledge bases.

Findings

Achieves state-of-the-art results on WN18RR dataset.

Obtains best non-entity-embedding results on FB15k-237.

Performs well on link prediction with unseen entities.

Abstract

Knowledge Bases (KBs) are easy to query, verifiable, and interpretable. They however scale with man-hours and high-quality data. Masked Language Models (MLMs), such as BERT, scale with computing power as well as unstructured raw text data. The knowledge contained within those models is however not directly interpretable. We propose to perform link prediction with MLMs to address both the KBs scalability issues and the MLMs interpretability issues. To do that we introduce MLMLM, Mean Likelihood Masked Language Model, an approach comparing the mean likelihood of generating the different entities to perform link prediction in a tractable manner. We obtain State of the Art (SotA) results on the WN18RR dataset and the best non-entity-embedding based results on the FB15k-237 dataset. We also obtain convincing results on link prediction on previously unseen entities, making MLMLM a suitable approach to introducing new entities to a KB.