ReFinED: An Efficient Zero-shot-capable Approach to End-to-End Entity Linking
We introduce ReFinED, an efficient end-to-end entity linking model which uses fine-grained entity types and entity descriptions to perform linking. The model performs mention detection, fine-grained entity typing, and entity disambiguation for all mentions within a document in a single forward pass, making it more than 60 times faster than competitive existing approaches. ReFinED also surpasses state-of-the-art performance on standard entity linking datasets by an average of 3.7 F1. The model is capable of generalising to large-scale knowledge bases such as Wikidata (which has 15 times more entities than Wikipedia) and of zero-shot entity linking. The combination of speed, accuracy and scale makes ReFinED an effective and cost-efficient system for extracting entities from web-scale datasets, for which the model has been successfully deployed. Our code and pre-trained models are available at https://github.com/alexa/ReFinED
PDF Abstract NAACL (ACL) 2022 PDF NAACL (ACL) 2022 AbstractCode
Datasets
AIDA CoNLL-YAGO
AQUAINT
Results from the Paper
Ranked #1 on
Entity Linking
on WebQSP-WD
(using extra training data)
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Uses Extra Training Data |
Benchmark |
|---|---|---|---|---|---|---|---|
| Entity Disambiguation | ACE2004 | ReFinED | Micro-F1 | 91.6 | # 3 | ||
| Entity Typing | AIDA-CoNLL | ReFinED | Micro-F1 | 84.0 | # 1 | ||
| Entity Linking | AIDA-CoNLL | ReFinED | Micro-F1 strong | 84.0 | # 6 | ||
| Entity Disambiguation | AIDA-CoNLL | ReFinED | In-KB Accuracy | 93.9 | # 8 | ||
| Entity Disambiguation | AQUAINT | ReFinED | Micro-F1 | 91.8 | # 3 | ||
| Entity Linking | Derczynski | ReFinED | Micro-F1 | 50.7 | # 2 | ||
| Micro-F1 strong | 50.7 | # 2 | |||||
| Entity Linking | KORE50 | ReFinED | Micro-F1 strong | 64.7 | # 2 | ||
| Micro-F1 | 65.9 | # 1 | |||||
| Entity Linking | MSNBC | ReFinED | Micro-F1 | 71.8 | # 5 | ||
| Micro-F1 strong | 71.8 | # 2 | |||||
| Entity Disambiguation | MSNBC | ReFinED | Micro-F1 | 94.4 | # 3 | ||
| Entity Linking | N3-Reuters-128 | ReFinED | Micro-F1 | 58.1 | # 1 | ||
| Micro-F1 strong | 58.1 | # 1 | |||||
| Entity Linking | OKE-2015 | ReFinED | Micro-F1 | 65.0 | # 2 | ||
| Micro-F1 strong | 64.4 | # 1 | |||||
| Entity Linking | OKE-2016 | ReFinED | Micro-F1 | 59.5 | # 1 | ||
| Micro-F1 strong | 59.1 | # 2 | |||||
| Entity Linking | WebQSP-WD | ReFinED | F1 | 89.1 | # 1 | ||
| Entity Disambiguation | WNED-CWEB | ReFinED | Micro-F1 | 79.4 | # 1 | ||
| Entity Disambiguation | WNED-WIKI | ReFinED | Micro-F1 | 88.7 | # 3 |
