Google's "Edit Search results" experiment(justinhileman.info)
justinhileman.info
Google's "Edit Search results" experiment
http://justinhileman.info/blog/2008/07/googles-edit-search-results-experiment
5 comments
A whole new game would begin of "X'ing" competitors search results. They need to tread very very lightly and put many checks and balances in place.
Yeah see Wikia for an example of this. Monitor any semi-high volume keyword and you'll witness a parade of buries/promotions until 1 side finally gives up.
The newest "AI on the web" strategy seems to center around this stuff -- computers do 95% of the work and rely on actual people when the law of diminishing returns kicks in.
We have so many warm bodies on the internet with this marvelous (and mysterious) sense of intuition that we can't seem to make the computer machines replicate. Why not?
See also: the super cool reCAPTCHA project. Or the FoldIt! protein folding game.
We have so many warm bodies on the internet with this marvelous (and mysterious) sense of intuition that we can't seem to make the computer machines replicate. Why not?
See also: the super cool reCAPTCHA project. Or the FoldIt! protein folding game.
Isn't monitoring where people actually surf more accurate than allowing those same people to 'vote' on a search result's relevance?
I would think many people would vote one way, and surf another.
I would think many people would vote one way, and surf another.
When I'm searching for a solution to a problem, I often have to click through to several results before finding it (assuming I find it at all). In that case, clicking through is not an endorsement; clicking through and then not clicking anywhere else might be, but it might mean I've given up, or it might mean I opened several results at once and found the solution in any one of them.
It might be more accurate in other search domains though.
It might be more accurate in other search domains though.
as a wise man once said "everything that can be socially powered, will be"
we were doing this stuff for a couple of years at a search company I worked for.
If you're allowed to talk about it, the data you get back is sparse, right? So how useful does the user feedback end up being? It seems like one of those areas where owning search share actually means something.
There's a type of machine learning called semi-supervised learning that can make use of sparse datasets like you suggest.
I think Google is finally realizing that the algorithms still have a way to go and that the massive amount of good stuff in human's heads is quite relevant.