Table 2. .

Classical, neighborhood-based link predictors for directed and weighted networks

Method (abbreviation)	Formula
Common neighbors v2 (CN2)	CN2(u, v) = ${\displaystyle \frac{1}{2}}{\displaystyle \sum _{z\in I}}$ [w(z, u) + w(z, v)]
Preferential attachment (PA2)	PA2(u, v) = $\left({\displaystyle \frac{{\sum }_{z\in {\Gamma }_o\left(u\right)}w\left(z,u\right)}{\left|{\Gamma }_o\left(u\right)\right|}}\right)\left({\displaystyle \frac{{\sum }_{z\in {\Gamma }_o\left(v\right)}w\left(z,v\right)}{\left|{\Gamma }_o\left(v\right)\right|}}\right)$
Adamic-Adar v2 (AA2)	AA2(u, v) = ${\displaystyle \frac{1}{2}}\left({\displaystyle \sum _{z\in I}}{\displaystyle \frac{w\left(z,u\right)+w\left(z,v\right)}{log{\sum }_{x\in {\Gamma }_o\left(z\right)}w\left(x,z\right)}}\right)$
Resource allocation v2 (RA2)	RA2(u, v) = ${\displaystyle \sum _{z\in I}}{\displaystyle \frac{w\left(z,u\right)+w\left(z,v\right)}{{\sum }_{x\in {\Gamma }_o\left(z\right)}w\left(x,z\right)}}$
Jaccard v2 (JA2)	JA2(u, v) = ${\displaystyle \frac{{\sum }_{z\in I}min\left(w\left(z,u\right),w\left(z,v\right)\right)}{{\sum }_{z\in I}max\left(w\left(z,u\right),w\left(z,v\right)\right)}}$

Note. The formulas have been adapted to be based on the out-link neighborhood information of the endpoints (u, v) of the directed link to be predicted. Each formula provides a prediction score s(u, v) for that directed link. Here I denotes the set of all target (injected) areas and Γ₀(u) denotes the neighbors of u, including itself.