BMC Bioinformatics

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

Model	Micro-averaged			Macro-averaged
Model	P	R	F1	P	R	F1
LSTM	0.486	0.388	0.431	0.537	0.443	0.395
+ Self-attention	0.490	0.391	0.435	0.550	0.452	0.405
+ Knowledge-based attention	0.495	0.395	0.440	0.559	0.465	0.417
GRU	0.484	0.387	0.430	0.548	0.454	0.406
+ Self-attention	0.491	0.393	0.436	0.558	0.461	0.414
+ Knowledge-based attention	0.495	0.395	0.439	0.551	0.454	0.407
BiLSTM	0.486	0.389	0.432	0.541	0.446	0.398
+ Self-attention	0.493	0.394	0.438	0.552	0.456	0.408
+ Knowledge-based attention	0.499	0.400	0.444	0.559	0.461	0.414
BiGRU	0.487	0.389	0.433	0.542	0.446	0.398
+ Self-attention	0.497	0.397	0.441	0.558	0.462	0.415
+ Knowledge-based attention	0.501	0.400	0.445	0.562	0.464	0.416
Hierarchical-ConvNet	0.468	0.375	0.416	0.522	0.429	0.381
+ Self-attention	0.473	0.378	0.420	0.529	0.434	0.386
+ Knowledge-based attention	0.483	0.386	0.429	0.532	0.441	0.392

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.

Back to article page

ISSN: 1471-2105

Contact us

General enquiries: journalsubmissions@springernature.com