Answer Sentence Selection on Alexa Virtual Assistant traffic accurate Sample 3 (test)
0.5814Prec@1TANDA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TANDABackbone=RoBERTa, Model Size=Large, Transfer Sequence=ASNQ → NAD2019.11 | 0.5814 | 0.514 | 0.699 | |
| TANDABackbone=RoBERTa, Model Size=Base, Transfer Sequence=ASNQ → NAD2019.11 | 0.5698 | 0.473 | 0.679 | |
| RoBERTaModel Size=Large, Training Strategy=Zero-shot from ASNQ2019.11 | 0.5465 | 0.478 | 0.674 | |
| RoBERTaModel Size=Large, Training Strategy=Fine-tuning on NAD2019.11 | 0.5291 | 0.49 | 0.651 | |
| TANDABackbone=BERT, Model Size=Large, Transfer Sequence=ASNQ → NAD2019.11 | 0.5116 | 0.439 | 0.616 | |
| TANDABackbone=BERT, Model Size=Base, Transfer Sequence=ASNQ → NAD2019.11 | 0.4942 | 0.391 | 0.613 | |
| RoBERTaModel Size=Base, Training Strategy=Fine-tuning on NAD2019.11 | 0.4826 | 0.43 | 0.612 | |
| BERTModel Size=Large, Training Strategy=Zero-shot from ASNQ2019.11 | 0.4593 | 0.399 | 0.585 | |
| RoBERTaModel Size=Base, Training Strategy=Zero-shot from ASNQ2019.11 | 0.4535 | 0.437 | 0.608 | |
| BERTModel Size=Base, Training Strategy=Zero-shot from ASNQ2019.11 | 0.4419 | 0.369 | 0.561 | |
| BERTModel Size=Large, Training Strategy=Fine-tuning on NAD2019.11 | 0.4361 | 0.395 | 0.558 | |
| BERTModel Size=Base, Training Strategy=Fine-tuning on NAD2019.11 | 0.4186 | 0.352 | 0.543 |