Improving Entity Disambiguation by Reasoning over a Knowledge Base

NAACL 2022  ·  Tom Ayoola, Joseph Fisher, Andrea Pierleoni ·

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.

PDF Abstract NAACL 2022 PDF NAACL 2022 Abstract

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Entity Disambiguation ACE2004 KBED Micro-F1 93.4 # 1
Entity Disambiguation AIDA-CoNLL KBED In-KB Accuracy 90.4 # 16
Entity Disambiguation AQUAINT KBED Micro-F1 92.6 # 2
Entity Disambiguation MSNBC KBED Micro-F1 94.8 # 2
Entity Disambiguation ShadowLink-Shadow KBED Micro-F1 47.6 # 1
Entity Disambiguation ShadowLink-Top KBED Micro-F1 64.2 # 1
Entity Disambiguation WNED-CWEB KBED Micro-F1 78.2 # 3
Entity Disambiguation WNED-WIKI KBED Micro-F1 90.4 # 1