Table 7.
The coreference performance of different models on identical mentions with different POS tags on BioNLP dataset
| NN/NNS/NP | PRP | WH | IN | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| P | R | F1 | P | R | F1 | P | R | F1 | P | R | F1 | |
| BioNeu | 62.5 | 17.8 | 27.7 | 39.0 | 37.8 | 38.4 | 68.3 | 73.0 | 70.6 | 65.6 | 64.3 | 64.9 |
| BioNeu-feature | 42.9 | 21.4 | 28.5 | 46.3 | 41.3 | 43.6 | 69.1 | 75.7 | 72.2 | 70.1 | 74.2 | 72.1 |
| BioNeu + SFA | 52.9 | 32.1 | 40.0 | 45.7 | 34.8 | 39.5 | 72.0 | 73.0 | 72.5 | 71.6 | 72.3 | 71.9 |
| KE-LSTM | 66.7 | 21.4 | 32.4 | 45.0 | 39.1 | 41.8 | 63.5 | 73.0 | 67.9 | 67.6 | 68.3 | 68.0 |
| KE-LSTM-feature | 75.0 | 21.4 | 33.3 | 47.0 | 34.9 | 40.1 | 71.8 | 75.6 | 73.6 | 72.0 | 76.2 | 74.0 |
| KE-LSTM + SFA | 92.3 | 80.0 | 85.7 | 47.5 | 50.0 | 48.7 | 66.3 | 77.5 | 71.5 | 60.4 | 63.4 | 61.9 |
KE-LSTM is the baseline in [10]