The simple answer is: You can’t…
The score is fine to compare different models of the same complex, but I would not correlate it to binding affinity.
See for example:
P.L. Kastritis, J.P.G.L.M. Rodrigues and A.M.J.J. Bonvin HADDOCK2P2I: A robust biophysical model for predicting the binding affinity of protein-protein interaction inhibitors J. Chem. Info. Model. 54, 826-836 (2014).
P.L. Kastritis and A.M.J.J. Bonvin Molecular origins of binding affinity: Seeking the Archimedean point. Curr. Opin. Struct. Biol., 23, 868-877 (2013).
P.L. Kastritis and A.M.J.J. Bonvin On the binding affinity of macromolecular interactions: daring to ask why proteins interact J. R. Soc. Interface, 10, doi: 10.1098/rsif.2012.0835 (2013).