Skip to main content

Table 4 Results with different choices of model architecture on the test set for PNEN

From: Knowledge-enhanced biomedical named entity recognition and normalization: application to proteins and genes

ModelMicro-averagedMacro-averaged
PRF1PRF1
LSTM0.4860.3880.4310.5370.4430.395
+ Self-attention0.4900.3910.4350.5500.4520.405
+ Knowledge-based attention0.4950.3950.4400.5590.4650.417
GRU0.4840.3870.4300.5480.4540.406
+ Self-attention0.4910.3930.4360.5580.4610.414
+ Knowledge-based attention0.4950.3950.4390.5510.4540.407
BiLSTM0.4860.3890.4320.5410.4460.398
+ Self-attention0.4930.3940.4380.5520.4560.408
+ Knowledge-based attention0.4990.4000.4440.5590.4610.414
BiGRU0.4870.3890.4330.5420.4460.398
+ Self-attention0.4970.3970.4410.5580.4620.415
+ Knowledge-based attention0.5010.4000.4450.5620.4640.416
Hierarchical-ConvNet0.4680.3750.4160.5220.4290.381
+ Self-attention0.4730.3780.4200.5290.4340.386
+ Knowledge-based attention0.4830.3860.4290.5320.4410.392
  1. Only the normalized IDs returned by the systems are evaluated on both micro-averaged and macro-averaged metrics. Micro-averaged calculates metrics globally by counting the total true positives, false negatives and false positives, macro-averaged calculates metrics for each label in documents and finds their unweighted mean. The highest scores are highlighted in bold. We tune the hyper-parameters through the validation set and use the official evaluation script to assess the performance of the final chosen model on the test set.