We do this by accounting for the different spectral footprints (i.e., sizes) of the possible connections, their relative frequency, and the lengths of the shortest paths in the network weighted by the likelihood that a connection is requested between a pair of nodes.