Algorithm 1. Pseudo-code of our proposed ConBGAT model.

Input: Scanned Image I

Output: Extracted Information

for k = 1 to I do

Text Feature Extraction using DistilBert in Eq. (1)

end for

Extract Text Feature TFE in Eq. (2)

for k = 1 to I do

Image Feature Extraction using ResNet50 in Eq. (3)

end for

Extract Image Feature IFE in Eq. (4)

Concat Text Feature and Image Feature $\ddot{\mathrm{\Upsilon }}$ in Eq. (5)

for t = 1, 2, … , T do

Training model using GAT: ConBGAT = GAT( $\ddot{\mathrm{\Upsilon }}$, p)

Calculate Loss Function in Eq. (10)

Optimize Loss Function with AdamW

end for

Return Extracted Information