Relation Extraction on NYT-10 (test)
0.424AUCREDSandT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| REDSandTmodel_type=Transformer (BERT), features=label embeddings, attention over instance tokens2021.02 | 0.424 | 78 | 73 | 67.6 | |
| DISTREmodel_type=Transformer (GPT), training=fine-tuned with auxiliary objective2021.02 | 0.422 | 68 | 65.3 | 65 | |
| RESIDEmodel_type=Graph-CNN, side_information=entity types, relational phrases2021.02 | 0.415 | 81.8 | 74.3 | 69.7 | |
| PCNN+ATTmodel_type=CNN model, attention=instance-level2021.02 | 0.341 | 73 | 67.3 | 63.6 | |
| Mintzmodel_type=Multi-class logistic regression2021.02 | 0.107 | 52.3 | 45 | 39.7 |