Multiview Embedding with Partial Labels to Recognize Users of Devices Based on Unified Transformer

Table 7

Comparisons with GNNs for embedding of devices.

Type	Models	Input	Evaluation
			Accuracy	Precision	Recall	F1	Inductive
—	Concat	F	0.6123	0.1235	0.5353	0.2007	×
Homogeneous graph models	Node2Vec	G	0.6936	0.1622	0.5692	0.2525	×
	Struc2Vec	G	0.7212	0.1900	0.6332	0.2923	×
	GCN	F + G	0.8312	0.3092	0.6943	0.4279	×
	GAT	F + G	0.8545	0.3513	0.7095	0.4699	√
	GraphSAGE	F + G	0.8014	0.2675	0.6814	0.3842	√
	UniMP	F + G	0.8952	0.4526	0.7289	0.5584	×
Heterogeneous graph models	Metapath2Vec	G	0.6531	0.1326	0.5084	0.2103	×
	RGCN	F + G	0.7751	0.2249	0.6026	0.3276	×
	RGAT	F + G	0.7924	0.2470	0.6264	0.3543	×
	HAN	F + G	0.8354	0.3197	0.7189	0.4426	×
	HGT	F + G	0.8527	0.3550	0.7597	0.4839	×
	Concat	F	0.6123	0.1235	0.5353	0.2007	×
Multiview graph models	MVEPL	F + G + PL	0.9158	0.5265	0.7337	0.6131	√