Like the “mechanical turk” , the service that Amazon offers .. see mturk.com or google around for explanations and links .. the idea being to employ (usually low paid) people to make judgments that algorithms cannot, or cannot do so well, perhaps as inputs to ML procedures.

As in real people to assess the quality of their search results.

It would seem so .. see

http://anand.typepad.com/datawocky/2008/06/how-google-measures-search-quality.html

I guess there is nothing new about this, the machine learning community has long been interested in “assisted” and “on line” learning.

I’d love to be a fly on the wall at the GooglePlex to see how they integrate the human classifications/ratings with their own proprietary algorithms . My guess is “informal” . if the humans outperform the algorithm, then they tweak the algorithm.

Interesting.

For previous posts reflecting my ongoing interest in googlestuff, see

• http://dsanalytics.com/dsblog/honest-search-engine-optimization_132
• http://dsanalytics.com/dsblog/is-pagerank-any-use_124
• http://dsanalytics.com/dsblog/website-optimization-and-data-analytics_90