Skip to main content

Table 3 The overall performance of different approaches on the share-disorder dataset in detecting 7 attributes of given disorders: negation (neg), subject (sub), conditional (con), severity (sev), course (cou), uncertainty (unc), body location (bdl). best results are shown in boldface

From: Applying a deep learning-based sequence labeling approach to detect attributes of medical concepts in clinical text

Attribute

NEG

SUB

CON

SEV

COU

UNC

BDL

1.1.1.Baseline

(Bi-LSTM-CRF + SVM)

Acc.

0.9323

0.9929

0.9669

0.9655

0.9576

0.9445

0.7524

P

0.7931

0.7374

0.6990

0.6421

0.5068

0.4091

0.5887

R

0.7768

0.6348

0.5987

0.7568

0.6437

0.4172

0.7516

F

0.7849

0.6822

0.6449

0.6948

0.5671

0.4131

0.6602

1.1.1.Baseline

(Bi-LSTM-CRF + Bi-LSTM)

Acc.

0.9146

0.9900

0.9632

0.9707

0.9597

0.9308

0.7859

P

0.8387

0.8158

0.7872

0.7609

0.6340

0.4380

0.7218

R

0.7277

0.5391

0.6054

0.8213

0.6322

0.3819

0.784

F

0.7793

0.6492

0.6844

0.7900

0.6331

0.4080

0.7516

1.1.1.Sequence Labeling

Acc.

0.9542

0.9937

0.9718

0.9817

0.9697

0.955

0.8695

P

0.8142

0.8222

0.7583

0.7812

0.6150

0.4854

0.7887

R

0.8310

0.6435

0.6682

0.8859

0.7529

0.4393

0.7991

F

0.8225

0.7220

0.7104

0.8302

0.6770

0.4612

0.7939