Table 5 Performance of the association prediction methods using character context-based embeddings learned from secondary-character tags

From: A multi-modal dataset and method for bone-level association prediction in oracle bone inscriptions

Model Type

Method

Evaluation Metric

 

Sentence Embedding

Classifier

AUROC

AUPR

Accuracy

Precision

Recall

F1 score

Two-Stage

SIF

LR

0.5521

0.1337

0.5245

0.1013

0.5374

0.1705

  

SVM

0.5534

0.1358

0.6102

0.1120

0.4742

0.1812

  

XGBoost

0.7688

0.3865

0.9170

0.7237

0.1418

0.2371

  

LightGBM

0.7577

0.3414

0.9124

0.8132

0.0477

0.0901

  

MLP

0.7247

0.2587

0.9095

0.6106

0.0160

0.0311

 

uSIF

LR

0.5538

0.1495

0.9075

0.3452

0.0187

0.0355

  

SVM

0.5271

0.1100

0.8482

0.1497

0.1430

0.1463

  

XGBoost

0.7647

0.3892

0.9185

0.7368

0.1624

0.2661

  

LightGBM

0.7607

0.3749

0.9138

0.8293

0.0657

0.1218

  

MLP

0.7511

0.2884

0.9097

0.6226

0.0184

0.0357

 

BoW

LR

0.6072

0.1320

0.8113

0.1548

0.2410

0.1885

  

SVM

0.6157

0.1258

0.7569

0.1424

0.3331

0.1995

  

XGBoost

0.7888

0.4070

0.9164

0.7647

0.1173

0.2034

  

LightGBM

0.8053

0.4313

0.9142

0.7919

0.0760

0.1387

  

MLP

0.8570

0.4533

0.9182

0.6869

0.1848

0.2908

End-to-End

TextCNN

0.9090

0.5810

0.9278

0.6207

0.5579

0.5834

 

Transformer Encoder

0.7684

0.2527

0.9099

0.6294

0.0241

0.0462

 

LSTM

0.8903

0.5322

0.9218

0.5869

0.4814

0.5281

 

BiLSTM

0.9068

0.5701

0.9249

0.5936

0.5545

0.5733

 

SGBSAP

0.9496

0.7375

0.9454

0.7308

0.6344

0.6791

  1. Boldface indicates the best performance.
Back to article page