Table 3 Performance of the association prediction methods using glyph-based embeddings learned from the glyph-based embedding module

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

0.5276

0.1055

0.5508

0.0994

0.4884

0.1651

  

SVM

0.5383

0.1051

0.7427

0.1118

0.2635

0.1570

  

XGBoost

0.6957

0.3203

0.9137

0.7632

0.0747

0.1362

  

LightGBM

0.6406

0.2762

0.9112

0.9091

0.0258

0.0501

  

MLP

0.7384

0.3788

0.9155

0.7789

0.0985

0.1746

 

uSIF

LR

0.5385

0.1121

0.8942

0.1476

0.0341

0.0555

  

SVM

0.5621

0.1162

0.7834

0.1356

0.2570

0.1775

  

XGBoost

0.7049

0.3492

0.9162

0.8211

0.1005

0.1791

  

LightGBM

0.6647

0.3010

0.9115

0.9184

0.0290

0.0562

  

MLP

0.7488

0.3940

0.9157

0.7735

0.1036

0.1827

End-to-End

TextCNN

0.8823

0.5890

0.9260

0.6057

0.5506

0.5752

 

Transformer Encoder

0.9240

0.6333

0.9275

0.6023

0.6001

0.6010

 

LSTM

0.8943

0.5634

0.9268

0.6095

0.5457

0.5755

 

BiLSTM

0.9167

0.6333

0.9350

0.6680

0.5675

0.6135

 

SGBSAP

0.9587

0.7883

0.9537

0.7584

0.7206

0.7390

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