Performance as a function of justification length in sentences (or, the number of graphlets in a TAG) for two models: one aware of connection-type, and one that is not. Bold font indicates the best score in a given column for each model group.
| Model . | P@1 . | P@1 Impr. . | MRR . | MRR Impr. . |
|---|---|---|---|---|
| Normal | ||||
| 1G | 35.33 | – | 59.26 | – |
| 2G | 38.16 | 8.0% | 61.33 | 3.5% |
| 3G | 37.78 | 6.3% | 61.14 | 3.2% |
| Connection-type aware (Daumé) | ||||
| 1GCT | 34.80 | – | 58.86 | – |
| 2GCT | 39.91 | 14.7% | 62.53 | 6.2% |
| 3GCT | 38.53 | 10.7% | 61.65 | 4.7% |
| Model . | P@1 . | P@1 Impr. . | MRR . | MRR Impr. . |
|---|---|---|---|---|
| Normal | ||||
| 1G | 35.33 | – | 59.26 | – |
| 2G | 38.16 | 8.0% | 61.33 | 3.5% |
| 3G | 37.78 | 6.3% | 61.14 | 3.2% |
| Connection-type aware (Daumé) | ||||
| 1GCT | 34.80 | – | 58.86 | – |
| 2GCT | 39.91 | 14.7% | 62.53 | 6.2% |
| 3GCT | 38.53 | 10.7% | 61.65 | 4.7% |