Table 4 Hyperparameters and pre-trained embeddings used for named entity recognition and relation extraction baseline results.

From: The Leaf Clinical Trials Corpus: a new resource for query generation from clinical trial eligibility criteria

Task

Architecture

Hyperparameter/Embeddings

Training Value

Named Entity Recognition

biLSTM + CRF

Character Dimensions

25

Token Embedding Dimensions

100

Learning Rate

0.005

Dropout

0.5

Pretrained Embeddings

GloVe28

Relation Extraction

BERT & R-BERT

Pretrained Model

SciBert

Learning Rate

0.00003

  1. For the NER task, the same architecture and hyperparameters were used for both general and fine-grained entity models. For the relation extraction task, the same hyperparameters were used with both the BERT and R-BERT architectures.
Back to article page