Server/article is down. Its a rather interesting problem and it generated lots of questions. One immediate question is how you determine the the nodes that sit on the rail voltages (i.e. the good and bad nodes)? Would be interesting to see if how they measured the success of this solution in reducing click fraud etc.
Meta-question: If the author is no longer able to publish the work, but HN still hosts the article comments with the title "Using an electrical circuit solver to track ad click fraud," then isn't the cat already out of the bag?
At this point of time, i honestly believe that it was an open problem in the public domain, that any high school student could have solved (at least, when nudged in the right direction)... And I am pretty sure that a lot of people have come up with a lot more useful/interesting solutions.
And that's why I openly shared my work, hoping that it would be just an interesting little observation...
So I don't believe that the cat was ever hiding in a bag ...
This can actually be improved by adding few capacitors at the nodes which would take into account, how many clicks a user has been better than the the black-listed ones or worse than the white-listed ones.
I'm not sure how capacitors would help at all. From what I can tell, they're just solving for the steady-state DC case; in that condition, capacitors are essentially open and have no effect on the analysis.
Ad-hoc as the use of a circuit solver might seem, there is some quite interesting machine learning literature on the use of resistive networks to do "graph transduction", e.g.:
