Skip to main content

Table 2 Article filtering performance with different features and classifiers

From: Mining physical protein-protein interactions from the literature

Model

Precision

Recall

F1 score

AUC

Mean

0.6642

0.7636

0.6868

0.7351

Standard deviation

0.0810

0.1926

0.1035

0.0741

Best reported in terms of AUC [8]

0.7080

0.8609

0.7770

0.8554

Our results in BioCreative 2006

0.7507

0.8107

0.7795

0.8471

Term (baseline)

0.7016

0.8213

0.7568

0.8037

String

0.7044

0.8960

0.7887

0.8416

Named entity (NE)

0.5815

0.9600

0.7243

0.7570

Template

0.7841

0.7653

0.7746

0.8239

String + NE

0.7360

0.8773

0.8005

0.8479

String + template

0.7416

0.8880

0.8082

0.8372

String + NE + template

0.7585

0.8373

0.7959

0.8507

String + term + NE + template

0.7432

0.8720

0.8025

0.8608

Naïve Bayes classifier

0.6321

0.8613

0.7291

0.7884

Multinomial classifier

0.6264

0.8720

0.7290

0.7770

Linear kernel SVM

0.7016

0.8213

0.7568

0.8037

p-spectrum kernel SVM (p = 7)

0.7352

0.8293

0.7794

0.8376

Integration of the above four classifiers (AdaBoost)

0.7995

0.8933

0.8438

0.8746

  1. This table shows the experimental results from article filtering. AUC, area under the receiving operator characteristic curve; SVM, support vector machine.