From: Exploiting document graphs for inter sentence relation extraction
Method/model | Precision | Recall | F1 | |
---|---|---|---|---|
NOT having the ability to extract inter sentence relations | ||||
hybridDNN (Zhou et al., 2016 [41]) | Syntactic features | 62.15 | 47.28 | 53.70 |
 | + Context | 62.39 | 47.47 | 53.92 |
 | + Position | 62.86 | 47.47 | 54.09 |
ASM (Panyam et al., 2018 [42]) | Dependency graph | 49.00 | 67.40 | 56.80 |
MASS (Le et al., 2018 [28]) | Multi channel CNN-LSTM | 58.90 | 54.90 | 56.90 |
 | + Ensemble | 56.80 | 57.90 | 57.30 |
 | + Post processing | 52.80 | 71.10 | 60.60 |
Having the ability to extract inter sentence relations | ||||
UET-CAM (Le et al., 2016 [23]) | SVM + coreference | 53.41 | 49.41 | 51.60 |
 | + Data | 57.63 | 60.23 | 58.90 |
SVM (Peng et al., 2016 [24]) | SVM + Rich feature set | 64.24 | 52.06 | 57.51 |
 | + Data | 65.59 | 56.94 | 61.01 |
CNN+ME (Gu et al., 2017 [25]) | Hybrid model | 60.90 | 59.50 | 60.20 |
 | + Post-processing | 55.70 | 68.10 | 61.30 |
LSTM-CNN (Zheng et al., 2018 [20]) | Sequence of sentences | 24.00 | 52.00 | 32.80 |
 | + Entity replacing | 54.30 | 65.90 | 59.50 |
BRAN (Verga et al., 2018 [17]) | CNN + abstract attention | 55.60 | 70.80 | 62.10 |
 | + Data | 64.00 | 69.20 | 66.20 |
 | + Ensemble | 65.40 | 71.80 | 68.40 |
Graph CNN (Sahu et al., 2019 [18]) | Document-level Graph | 52.80 | 66.00 | 58.60 |
Our results | Document subgraph | 60.13 | 65.89 | 62.88 |
 | + Data | 62.95 | 75.16 | 68.52 |
 | + Ensemble | 64.79 | 74.05 | 69.11 |