Table 4:

The effect of fine-tuning GaLM* with pre-trained GNNs on stitched application graphs.

Models	Search-CTR	ESCvsI	Query2PT
	(ROC-AUC)	(macro-F1)	(macro-F1)
${\text{GALM}}_{\text{rgat}}^{*}+$	20.37%	12.99%	80.06%
${\text{GALM}}_{\text{rgat}}^{**}+$	19.93%	11.34%	77.25%