Unveiling the silent majority: stance detection and characterization of passive users on social media using collaborative filtering and graph convolutional networks

EPJ Data Science

Table 3 Performance including/excluding the Projection & Fusion (P&F) layers

	nDCG	MAP	Acc.	RMSE	Least	Middle	Most
Dataset	Entry_DS
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\)) w/o P&F	0.0854	0.0355	0.7513	0.4984	0.6180	0.7333	0.9026
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\))	0.0860^∗	0.0364^∗	0.7673^∗	0.4809	0.6391^∗	0.7590^∗	0.9038
WLGCN + (\(\mathcal{G}_{\mathrm{Soc}}\)) w/o P&F	0.0850	0.0353	0.7488	0.5009	0.6217	0.7294	0.8951
WLGCN + (\(\mathcal{G}_{\mathrm{Soc}}\))	0.0857^∗	0.0357^∗	0.7704^∗	0.4790^∗	0.6479^∗	0.7569	0.9064^∗
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\), \(\mathcal{G}_{\mathrm{Soc}}\)) w/o P&F	0.0832	0.0343	0.7476	0.5022	0.6180	0.7333	0.8914
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\), \(\mathcal{G}_{\mathrm{Soc}}\))	0.0838	0.0357	0.7757^∗	0.4720	0.6539^∗	0.7710	0.9022
Dataset	Exit_DS
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\)) w/o P&F	0.0741	0.0378	0.6569	0.5924	0.5083	0.6000	0.8624
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\))	0.0769⁺	0.0392^∗	0.6795⁺	0.5658	0.5757^∗	0.6276⁺	0.8352
WLGCN + (\(\mathcal{G}_{\mathrm{Soc}}\)) w/o P&F	0.0772	0.0392	0.6605	0.5901	0.5083	0.5923	0.8807
WLGCN + (\(\mathcal{G}_{\mathrm{Soc}}\))	0.0770⁺	0.0390	0.6896⁺	0.5578^∗	0.5900⁺	0.6346⁺	0.8440
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\), \(\mathcal{G}_{\mathrm{Soc}}\)) w/o P&F	0.0742	0.0380	0.6552	0.5860	0.5420	0.5914	0.8321
WLGCN + (\(\mathcal{G}_{\mathrm{PathSim}}\), \(\mathcal{G}_{\mathrm{Soc}}\))	0.0749^∗	0.0384^∗	0.6922⁺	0.5496⁺	0.5567⁺	0.6577	0.8624

Note: “w/o P&F” means the model doesn’t include the P&F layers. We marked WLGCN variants’ results that are statistically significant (t-test) compared to the same variant but without the P&F layers: ∗(p<0.05), +(p<0.1).