Table 5. F1-score results of our proposed ConBGAT model with the fifteen compared baseline models on the SROIE, FUNSD and CORD datasets.

Model	SROIE	FUNSD	CORD
$BER{T}_{BASE}$	90.99	60.26	89.68
$BER{T}_{LARGE}$	92.00	65.63	90.25
$UniLMv{2}_{BASE}$	94.59	68.90	90.92
$UniLMv{2}_{LARGE}$	94.88	72.57	92.05
$LayoutL{M}_{BASE}$	94.38	78.66	94.72
$LayoutL{M}_{LARGE}$	95.24	78.95	94.93
$LayoutLMv{2}_{BASE}$	96.25	82.76	94.95
$LayoutLMv{2}_{LARGE}$	96.61	84.20	96.01
$LayoutLMv{2}_{LARGE}\left(excludingOCRmismatch\right)$	97.81	–	–
$LayoutLMv{3}_{BASE}$	–	90.29	96.56
$SpectralGraphConvolutionalNetwork$	91.50	-	–
$BROS$	94.93	81.21	95.58
$PICK$	96.79	–	–
$DocTr$		84.00	98.20
$ESP$	–	88.88	95.65
$GenKIE$	97.40	83.45	95.75
ConBGAT	97.40	89.61	96.72