The augmented dataset is generated with Neural Motif. To simplify, I represents internal transfer and E represents external transfer.
| Type | Link |
|---|---|
| I-10% | link |
| I-30% | link |
| I-70% | link |
| I-30% + E-100% | link |
| I-70% + E-100% | link |
The I-10% and 30% is much similar to the original dataset's predicate distribution.
| Task | Link |
|---|---|
| PREDCLS | link |
| SGCLS | link |
| SGDET | link |
| Task | Link |
|---|---|
| SGDET | link |
| Task | Link |
|---|---|
| SGCLS I-90% + E-100% | link |