Re: [igraph] interpretation of hub/auth scores

[Matthew Galati]

Agreed. This said, adding a small (not epsilon, but rather 0.1 or
something like that) constant to the graph still makes sense, the same
way as it does for PageRank on graphs that are not strongly connected.
But then this would be a parameter, like for PR.

It is the small epsilon, added only if double eigenvalues are
detected, that I am against.

Gabor

Hmm. OK. That is also an option - to always add some constant. This whole thread might be a good topic for Prof Kleinberg to consider. I didn't see anything about this in the original paper - I might have missed something in the proofs.