Table 2. Performance comparison between GNN-A² and baselines.

To enhance the clarity of the experimental results, the optimal results for each dataset are highlighted in bold, and the suboptimal ones are indicated with underlining.

Model	MovieLens 1M				Book-crossing				Taobao
	AUC	Logloss	NDCG@5	NDCG@10	AUC	Logloss	NDCG@5	NDCG@10	AUC	Logloss	NDCG@5	NDCG@10
FM	0.8761	0.4409	0.8143	0.8431	0.7417	0.5771	0.7616	0.8029	0.6171	0.2375	0.812	0.1120
NFM	0.8985	0.3996	0.8486	0.8832	0.7988	0.5432	0.7989	0.8326	0.6550	0.2122	0.0997	0.1251
W&D	0.9043	0.3878	0.8538	0.8869	0.8105	0.5366	0.8048	0.8381	0.6531	0.2124	0.0959	0.1242
Deep-FM	0.9049	0.3856	0.8510	0.8848	0.8127	0.5379	0.8088	0.8400	0.6550	0.2115	0.0974	0.1243
AutoInt	0.9034	0.3883	0.8619	0.8931	0.8130	0.5355	0.8127	0.8472	0.6434	0.2146	0.0924	0.1206
Fi-GNN	0.9063	0.3871	0.8705	0.9029	0.8136	0.5338	0.8094	0.8522	0.6462	0.2131	0.0986	0.1241
L0-SIGN	0.9072	0.3846	0.8849	0.9094	0.8163	0.5274	0.8148	0.8629	0.6547	0.2124	0.1006	0.1259
GMCF	0.9127	0.3789	0.9374	0.9436	0.8228	0.5233	0.8671	0.8951	0.6679	0.1960	0.1112	0.1467
CAN	0.9133	0.3773	0.9396	0.9442	0.8235	0.5143	0.8722	0.8996	0.6776	0.1919	0.1130	0.1494
GNN-A2	0.9101	0.3846	0.9511	0.9506	0.8400	0.4956	0.9003	0.9137	0.6715	0.1944	0.1159	0.1526
Improv	–	–	1.22%	0.68%	2.00%	3.64%	3.22%	1.57%	–	–	2.57%	2.14%