From: Exploiting document graphs for inter sentence relation extraction
Component removed/changed | Precision | Recall | F1 | Change of F1 |
---|---|---|---|---|
Full model | 60.13 | 65.89 | 62.88 | Â |
Without subgraph | 57.68 | 55.16 | 56.39 | -6.49 |
Without TITLE | 61.12 | 54.12 | 57.41 | -5.47 |
Without NEXT-SENT | 62.36 | 58.33 | 60.28 | -2.60 |
Without instance merging technique | 52.40 | 69.26 | 59.66 | -3.22 |
Without swCNN and top-k paths | 59.92 | 62.19 | 61.03 | -1.84 |
Choose top-k by highest frequency (instead of length) | 58.56 | 66.96 | 62.48 | -0.40 |
Use w=2 for both training and testing (instead of different w) | 61.25 | 61.26 | 61.25 | -1.62 |
Without using class weight | 59.60 | 65.92 | 62.60 | -0.28 |
Without attention mechanism | 59.13 | 64.85 | 61.86 | -1.02 |