Analogy recognition on SAT
60.78AccuracyDeBERTa-v3 + ANALOGYKB
Evaluation results
| Method | Links | |
|---|---|---|
| DeBERTa-v3 + ANALOGYKBBackbone=DeBERTa-v3, Knowledge Enhancement=ANALOGYKB, Filtering=Checked2023.05 | 60.78 | |
| RoBERTa-Large + ANALOGYKBBackbone=RoBERTa-Large, Knowledge Enhancement=ANALOGYKB, Filtering=Checked2023.05 | 59.05 | |
| Human2023.05 | 57 | |
| Word Embedding from RoBERTa-LargeBackbone=RoBERTa-Large, Embedding Type=Word2023.05 | 56.7 | |
| Word Embedding from InstructGPTBackbone=InstructGPT, Embedding Type=Word2023.05 | 55.38 | |
| DeBERTa-v3Backbone=DeBERTa-v32023.05 | 52.26 | |
| RoBERTa-LargeBackbone=RoBERTa-Large2023.05 | 51.6 | |
| DeBERTa-v3 + ANALOGYKB (w/o check)Backbone=DeBERTa-v3, Knowledge Enhancement=ANALOGYKB, Filtering=Unchecked2023.05 | 48.36 | |
| Sentence Embedding from SentenceBERTBackbone=SentenceBERT, Embedding Type=Sentence2023.05 | 47.7 | |
| Sentence Embedding from SimCSEBackbone=SimCSE, Embedding Type=Sentence2023.05 | 45.1 | |
| RoBERTa-Large + ANALOGYKB (w/o check)Backbone=RoBERTa-Large, Knowledge Enhancement=ANALOGYKB, Filtering=Unchecked2023.05 | 43.38 | |
| BERT-LargeBackbone=BERT-Large2023.05 | 41.3 | |
| ERNIEBackbone=ERNIE2023.05 | 34.92 | |
| T5-LargeBackbone=T5-Large2023.05 | 31.45 | |
| LUKEBackbone=LUKE2023.05 | 30.26 |