Table 3.

Overview of tested $\lambda$-values in weighted loss function $\mathcal{L}={\lambda }_1{\mathcal{L}}_{pred}+{\lambda }_2{\mathcal{L}}_{dist\left(\alpha \right)}+{\lambda }_3{\mathcal{L}}_{dist\left(\beta \right)}$ - with ${\mathcal{L}}_{pred}$ the prediction loss, ${\mathcal{L}}_{dist\left(\alpha \right)}$ the article–user distance loss, and ${\mathcal{L}}_{dist\left(\beta \right)}$ the user–user distance loss.

${\lambda }_1$	${\lambda }_2$	${\lambda }_3$	Description
1	0	0	No distance losses (= base)
0.5	0	0.5	No article-user distance loss
0.5	0.5	0	No user-user distance loss
0.33	0.33	0.33	All losses equally important
0.5	0.25	0.25	Prediction loss most important,
0.8	0.1	0.1	Distance losses are equally important
0.6	0.3	0.1	Article-user loss more important than user-user loss
0.6	0.1	0.3	User-user loss more important than article-user loss