Conversational web navigation on MT-Mind2Web (Cross-Website)
48.3Element AccuracySelf-MAP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Self-MAPBackbone=Flan-T5-base2024.02 | 48.3 | 71.8 | 40.6 | 18.2 | |
| MINDACT + FixedBackbone=Flan-T5-large2024.02 | 46.2 | 69.7 | 37.6 | 15.2 | |
| SynapseBackbone=Flan-T5-large2024.02 | 45.1 | 69 | 37.1 | 13 | |
| Self-MAPBackbone=Flan-T5-large2024.02 | 44.8 | 68.8 | 36.8 | 15.7 | |
| MINDACTBackbone=Flan-T5-large2024.02 | 43.6 | 67.6 | 36.5 | 12.4 | |
| MINDACT + CARBackbone=Flan-T5-large2024.02 | 43.2 | 69.2 | 36.1 | 12.2 | |
| SynapseBackbone=Flan-T5-base2024.02 | 43.1 | 70.6 | 33.1 | 13.7 | |
| MINDACT + FixedBackbone=Flan-T5-base2024.02 | 42.4 | 70 | 35.4 | 15.3 | |
| MINDACTBackbone=Flan-T5-base2024.02 | 38.8 | 69.4 | 29.2 | 15.2 | |
| MINDACT + CARBackbone=Flan-T5-base2024.02 | 37 | 67.5 | 32.2 | 9.6 | |
| DeBERTaBackbone=DeBERTa2024.02 | 31.7 | — | — | — | |
| MINDACTBackbone=GPT-3.52024.02 | 6.7 | 22.2 | 2.1 | 1.7 |