metadata

datasets:
  - cyanic-selkie/aida-conll-yago-wikidata
language:
  - en
metrics:
  - accuracy
  - f1
tags:
  - entity linking
  - entity disambiguation
  - EL
  - ReFinED
  - RoBERTa

Model Card for LLMAEL-ReFinED-FT

We introduce LLMAEL (LLM-Augmented Entity Linking), a pipeline method to enhance entity linking through LLM data augmentation. We release our customly fine-tuned LLMAEL-ReFinED-FT model, which is fine-tuned from the ReFinED EL model using an Llama-3-70b augmented version of the AIDA_train dataset. LLMAEL-ReFinED-FT yields new SOTA results across six standard EL benchmarks: AIDA_test, MSNBC, AQUAINT, ACE2004, WNED-CLUEWEB, and WNED-WIKIPEDIA, achieving an average 1.21% accuracy gain.

For more details, refer to our paper 📖 LLMAEL: Large Language Models are Good Context Augmenters for Entity Linking

Model Description

Developed by: Amy Xin, Yunjia Qi, Zijun Yao, Fangwei Zhu, Kaisheng Zeng, Bin Xu, Lei Hou, Juanzi Li
Model type: Entity Linking Model
Language(s): English
Finetuned from model [optional]: ReFinED