Table 2.

Summary of link features. We denote the Twitter neighbourhood as ${\mathbf{\Gamma }}^{\mathit{T}}$ and the Foursquare neighbourhood as ${\mathbf{\Gamma }}^{\mathit{F}}$

Twitter features
mentions	$|{\mathit{\text{mentions}}}_{ij}|$
hashtags	$|{\mathit{\text{hashtags}}}_{ij}|$
overlap	$\frac{|{\mathrm{\Gamma }}_i^T\cap {\mathrm{\Gamma }}_j^T|}{|{\mathrm{\Gamma }}_i^T\cup {\mathrm{\Gamma }}_j^T|}$
aa_sim	${\sum }_{z\in {\mathrm{\Gamma }}_i^T\cap {\mathrm{\Gamma }}_j^T}\frac{1}{log(|{\mathrm{\Gamma }}_z^T|)}$

Foursquare features
colocs	$|{\mathit{\text{colocations}}}_{ij}|$
dist	$\mathit{\text{haversine}}({\mathit{\text{lat}}}_i,{\mathit{\text{lon}}}_i,{\mathit{\text{lat}}}_j,{\mathit{\text{lon}}}_j)$
overlap	$\frac{|{\mathrm{\Gamma }}_i^F\cap {\mathrm{\Gamma }}_j^F|}{|{\mathrm{\Gamma }}_i^F\cup {\mathrm{\Gamma }}_j^F|}$
aa_sim	${\sum }_{z\in {\mathrm{\Gamma }}_i^F\cap {\mathrm{\Gamma }}_j^F}\frac{1}{log(|{\mathrm{\Gamma }}_z^F|)}$

Multilayer features
${\mathit{\text{int}}}_{GNij}$	${\sum }_{\alpha }^M|in{t}_{ij}^{\alpha }|$
${\mathit{\text{sim}}}_{GNij}$	$\frac{{\mathit{\text{sim}}}_{ij}^a}{{\mathit{\text{dist}}}_{ij}^b}$
overlap	$\frac{|{\mathrm{\Gamma }}_{CNi}\cap {\mathrm{\Gamma }}_{CNj}|}{|{\mathrm{\Gamma }}_{CNi}\cup {\mathrm{\Gamma }}_{CNj}|}$
aa_sim	${\sum }_{z\in {\mathrm{\Gamma }}_{CNi}\cap {\mathrm{\Gamma }}_{CNj}}\frac{1}{log(|{\mathrm{\Gamma }}_{CNz}|)}$