Conversational web navigation on MT-Mind2Web Cross-Subdomain
52Element AccuracySelf-MAP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Self-MAPBackbone=Flan-T5-large2024.02 | 52 | 77.1 | 43.6 | 25.4 | |
| MINDACT + FixedBackbone=Flan-T5-large2024.02 | 47.4 | 74.9 | 38.8 | 21.4 | |
| SynapseBackbone=Flan-T5-large2024.02 | 47.4 | 74.1 | 39.3 | 19.4 | |
| MINDACTBackbone=Flan-T5-large2024.02 | 46.8 | 74 | 38.9 | 21.8 | |
| Self-MAPBackbone=Flan-T5-base2024.02 | 46.4 | 79.1 | 38.3 | 20.8 | |
| MINDACT + CARBackbone=Flan-T5-large2024.02 | 44.5 | 75 | 40.2 | 15.6 | |
| MINDACT + FixedBackbone=Flan-T5-base2024.02 | 43.1 | 77.6 | 37.5 | 17.7 | |
| MINDACTBackbone=Flan-T5-base2024.02 | 41.9 | 77.2 | 35.5 | 15.7 | |
| SynapseBackbone=Flan-T5-base2024.02 | 41.7 | 77.8 | 35.9 | 16 | |
| MINDACT + CARBackbone=Flan-T5-base2024.02 | 41.2 | 75.3 | 35.4 | 13.2 | |
| DeBERTaBackbone=DeBERTa2024.02 | 27.7 | — | — | — | |
| MINDACTBackbone=GPT-3.52024.02 | 4 | 22.9 | 1.5 | 1.1 |