Improving Entity Disambiguation by Reasoning over a Knowledge Base

TL;DR

This paper presents a novel entity disambiguation model that leverages all knowledge base facts through differentiable reasoning, significantly improving accuracy especially for infrequent and ambiguous entities.

Contribution

The authors introduce a fully differentiable reasoning approach over KB facts for entity disambiguation, surpassing state-of-the-art performance and reducing reliance on entity popularity priors.

Findings

Outperforms baselines on six ED datasets by 1.3 F1 on average.

Improves performance on ShadowLink dataset by 12.7 F1.

Enables use of all KB facts, descriptions, and types in disambiguation.

Abstract

Recent work in entity disambiguation (ED) has typically neglected structured knowledge base (KB) facts, and instead relied on a limited subset of KB information, such as entity descriptions or types. This limits the range of contexts in which entities can be disambiguated. To allow the use of all KB facts, as well as descriptions and types, we introduce an ED model which links entities by reasoning over a symbolic knowledge base in a fully differentiable fashion. Our model surpasses state-of-the-art baselines on six well-established ED datasets by 1.3 F1 on average. By allowing access to all KB information, our model is less reliant on popularity-based entity priors, and improves performance on the challenging ShadowLink dataset (which emphasises infrequent and ambiguous entities) by 12.7 F1.