Empirical Evaluation of Pretraining Strategies for Supervised Entity Linking
Thibault Févry, Nicholas FitzGerald, Livio Baldini Soares, Tom Kwiatkowski
Abstract
In this work, we present an entity linking model which combines a Transformer architecture with large scale pretraining from Wikipedia links. Our model achieves the state-of-the-art on two commonly used entity linking datasets: 96.7% on CoNLL and 94.9% on TAC-KBP. We present detailed analyses to understand what design choices are important for entity linking, including choices of negative entity candidates, Transformer architecture, and input perturbations. Lastly, we present promising results on more challenging settings such as end-to-end entity linking and entity linking without in-domain training data.
Results
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Entity Linking | AIDA-CoNLL | Micro-F1 strong | 76.7 | Févry et al. (2020b) |
Related Papers
LEMONADE: A Large Multilingual Expert-Annotated Abstractive Event Dataset for the Real World2025-06-01Distilling Closed-Source LLM's Knowledge for Locally Stable and Economic Biomedical Entity Linking2025-05-26Evaluation of LLMs on Long-tail Entity Linking in Historical Documents2025-05-06KGMEL: Knowledge Graph-Enhanced Multimodal Entity Linking2025-04-21Cross-Document Contextual Coreference Resolution in Knowledge Graphs2025-04-08Explainable ICD Coding via Entity Linking2025-03-26Entity-aware Cross-lingual Claim Detection for Automated Fact-checking2025-03-19Leveraging Knowledge Graphs and LLMs for Context-Aware Messaging2025-03-12