Table 2.

Ablation experiments of different architectures.

Test model	AUC (%)	AUPR (%)
Weighted DeepWalk + Dot product Predictor	73.48	77.97
Weighted DeepWalk + GAT (untrained) + Dot product Predictor	89.78	88.59
Weighted DeepWalk + GAT + Dot product Predictor	92.17	91.52
Raw features + GAT + Dot product Predictor	72.68	69.27
Weight DeepWalk + Raw features + GAT + Dot product Predictor	93.52	93.09
Weight DeepWalk + Raw features + GAT + FC Predictor	94.03	93.51

The variants of the model are listed in the first column. We test each model variant for 10 times under five-fold CV on HMDD v2.0 dataset and report averages. Bold values indicates highlight our results.