Slashdot Mirror
Microsoft Uses Human Computing Game To Tune Bing
Al writes "Microsoft researchers have come up with a novel way to fine-tune the algorithms behind the company's new search engine, Bing: a game that harnesses human computing power to improve the results. Called Page Hunt, the game (which of course requires Silverlight to run) shows users a web page and asks them to figure out a search query that should produce the page within the first five results. The idea is to better understand user behavior and expectations and ultimately improve its search algorithms. Other human-computing projects have sought to digitize out-of-print text (reCAPTCHA) and image labeling (Google Image Labeler). Can Microsoft use a similar approach to gain the edge over its rival? Or does Google already have the edge with SearchWiki, which lets searchers re-rank its results?"
If users have the ability to tailor search results, won't page rank "fixers" (aka spammers) have an easier time? Or am I missing something?
Rhymes that keep their secrets will unfold behind the clouds.There upon the rainbow is the answer to a neverending story
Maybe they can use the real world data to fix this issue
http://www.bing.com/search?q=why+is+microsoft+word+so+expensive&form=QBLH&qs=n
http://www.google.com/search?hl=en&q=why+is+microsoft+word+so+expensive&aq=f&oq=&aqi=g1
Flooded with blog articles about the same query now, and yes, it looks like there's probably a technological reason (or at least viable excuse) for it, but it still seems pretty shady to me.
It's actually quite interesting that they're asking for search terms rather than simply labels. Search engines are the a form of machine learning, and a lot of ML research goes into improving them. So it's interesting to consider what Microsoft is asking, in the context of ML. For example, Google has a game where users play by tagging images. Obviously, they're using some sort of supervised classification algorithm under the hood. But with Bing they're not asking for 'tags,' which would imply a supervised classification system, but search queries which return the page. Now that suggests that Bing is actually built on a bayesian model, which is very different from Google's markov steady state (page rank) model.