1 Oct: This is done, and it works: from my sample I found several unusual accounts, though no bots – e.g. some that were created for an internal promotion. I don’t think there’s much more I can do with the data to hand, but perhaps one day I’ll get a few million hand histories from a site and can expand the project.
1 Aug: Worked on this to answer some other questions – all the data structures are now in place to finish this without enormous effort.
20 Jul: At Relax, I wrote the specs, hired the team, and oversaw development of their bot detection systems. Detection rates rose seven-fold under the new system in the first six months. Probably that’s a more useful metric than longer term, because you will catch every bot on the site that is vulnerable to the methods you are using.
So while I know a lot about how to do it when you have a lot of resources, I’m interested in seeing if I can find any using fairly simple methods on some hand history metadata I happen to have. Obviously it isn’t something I’d be able to post about (though I would notify the site), but it’s interesting to see what’s possible.
This project is half-complete just out of my own interest, though it’s possible the site I have the data from would want to pay me for my time spent on it. So this should probably move down my priority list until I hear about that one way or another.
ANW Consulting OÜ 