Table 7 Effect of different character contextual embedding strategies on the proposed method

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

Character Embedding

Character Tag

AUROC

AUPR

Accuracy

Precision

Recall

F1 score

VAE on Glyph Image

0.9587

0.7883

0.9537

0.7584

0.7206

0.7390

SGNS

Primary

0.9464

0.7337

0.9460

0.7329

0.6405

0.6835

 

Secondary

0.9496

0.7375

0.9454

0.7308

0.6344

0.6791

 

Both tags

0.9535

0.7592

0.9481

0.7305

0.6812

0.7050

CBOW-NEG

Primary

0.9353

0.6941

0.9398

0.7033

0.5856

0.6389

 

Secondary

0.9405

0.7106

0.9415

0.7152

0.5932

0.6482

 

Both tags

0.9469

0.7308

0.9446

0.7188

0.6416

0.6780

GloVe

Primary

0.9158

0.6336

0.9364

0.7247

0.4856

0.5815

 

Secondary

0.9118

0.6188

0.9345

0.7153

0.4656

0.5638

 

Both tags

0.9254

0.6631

0.9369

0.7086

0.5202

0.5999

  1. Boldface indicates the best performance.