tedster - 6:42 pm on Sep 30, 2012 (gmt 0)

No doubt judging something like "quality" is a huge job, even with the data volume Google has and the time they put into the machine-learning before going live. Panda was 1.5 years in the works.

I have some experience with machine learning in another web-related context, and it took about three years total over a massive data set to get to maybe 90% accuracy - that's with ongoing human feedback, too. A 10% failure rate on Panda's quality scoring would not be close to acceptable, IMO - because that algo has the ability to destroy businesses. I do think Panda is probably much better than a 10% failure rate, but just how much better is a big question.

----

On another note, it seems from comments here that the last three days have seen really major flux in the SERPs. We've got the Exact Match Domain update [webmasterworld.com], the Image Search update [webmasterworld.com] and something else as well - unless it's just fall-out from integrating the EMD quality scores into the rankings.