Table 2:

Testing accuracies (%) of GAT-GC variants (the original one and the ones applied with each of our 4 CPA models) on social network datasets. We highlight the result of the best performed model per dataset. The proportion P of multisets that hold the properties in Theorem 1 among all multisets is also reported.

Datasets P(%)	RE-B 100.0	RE-M5K 100.0
Original	50.00 ± 0.00	20.00 ± 0.00
Additive	93.07 ± 1.82	57.39 ± 2.09
Scaled	92.36 ± 2.27	56.76 ± 2.26
f-Additive	93.05 ± 1.87	56.43 ± 2.38
f-Scaled	92.57 ± 2.06	57.22 ± 2.20