Table 5: The overall comparison of models using ${\mathrm{G}\mathrm{A}\mathrm{L}\mathrm{M}}_{\text{rgat}}$on Amazon-PQ and Product-Reviews.

The LMs in models without a “*” are not fine-tuned on applications.

Models	Amazon-PQ			Product-Reviews
	Search-CTR	ESCvsI	Query2PT	CoPurchase	Product2PT
	(ROC-AUC)	(macro-F1)	(macro-F1)	(MRR)	(macro-F1)
LM(Bert-base) +Gnn(rgat)	15.60%	8.51%	39.56%	0.1941	0.7038
${\text{LM}}_{\left(\text{BERT-BASE}\right)}^{*}+{\text{GNN}}_{\left(\text{rgat}\right)}$	15.47%	6.49%	44.84%	0.3230	0.7479
GaLMrgat	17.30%	10.39%	64.03%	0.3461	0.7317
${\text{GALM}}_{\text{rgat}}^{*}$	17.49%	10.89%	66.50%	0.3474	0.7491
${\text{GALM}}_{\text{rgat}}^{*}+$	20.37%	13.35%	80.06%	0.4542	0.7813