More threads by Margaret Ornsby

Thanks so much Margaret! Can't wait to watch it.
 
I've wondered the extent to which Google does human testing to determine quality of it's results. Cool to see a little bit about how they do it, and cool too to see that local intent is one of the things they explicitly test and refine.

Also interesting to see 'insufficient content' as a mark against. I know for those of us working with local businesses, it's very common to see home pages with like 50 words of content... looking like that's a good best practices thing to address.

Cool stuff, thanks for sharing!
 
Awesome, awesome insights. Wish I were there

Sent from my SM-G900V using Tapatalk
 
@james
I've often wondered the same thing myself about the human testing.

We've known for years they do (such as "leaked" quality guidelines). G's kind of cagey about how much they do, but it seems to me there's more human testing than you might think at first.
Personally I believe that element of human testing is absolutely essential now that they have rankbrain going. I know lots of people say RB gives better results, but I have a few friends who completely disagree. Lesser used software for instant, with a name that's commonly used for other things, has become almost impossible to find at times >4th page... I'd like to think somewhere there are humans as a safety-brake, and it seems that's still in place in some form.
 
I imagine some cases are always going to have ranking problems... a mostly unknown piece of software with a name shared by something much more popular seems like more of a branding mistake than a Google mistake when looking at poor show in the SERPs.

I do agree with you though, the more nebulous and obfuscated Google's algorithms become even to Google themselves (a really common side effect any kind of machine learning algorithm) the more they'll need to have some way at least of verifying that edge cases haven't gone all wonky.

I remember reading an interesting story once about a guy that used a machine learning algorithm to 'program' a 100 gate chip to identify a sound at a particular frequency. When it settled on a final iteration, the solution it found only used about 60% of the chip, it didn't work on any other chip (it relied on microscopic manufacturing defects in the original chip) and it took them a long time to even figure out how it worked in the first place... and that's with a comparatively incredibly simple thing. I'm sure with the direction Google's heading in with Rank Brain, those of us outside Google aren't the only ones speculating on how exactly the algorithm decides what to show and what not to show.
 

Login / Register

Already a member?   LOG IN
Not a member yet?   REGISTER

LocalU Event

Trending: Most Viewed

  Promoted Posts

New advertising option: A review of your product or service posted by a Sterling Sky employee. This will also be shared on the Sterling Sky & LSF Twitter accounts, our Facebook group, LinkedIn, and both newsletters. More...
Top Bottom