Slashdot Mirror
Google Turns On User-Tweakable Search Wiki
Barence writes "Google has launched a new service that allows users to tailor to their own search results. Called SearchWiki, the service allows Google account holders to move results up or off the rankings, or even add their own choice of site to the top of the search results. Google claims that any changes a user makes will only affect their results, and not those of fellow surfers, although it's difficult to believe that some of the feedback generated from the SearchWiki won't be used to fine tune the Google search algorithm. Is this a cunning way to encourage people to sign in while they search, thus providing Google with a richer set of data that can be mapped to specific user accounts?"
I would happily spend *all* my time downgrading each and every result pointing to experts(-)exchange.com
http://userscripts.org/scripts/show/1898 removes experts-exchange.com results from Google searches :)
Three words: User Relevance Feedback.
In 1986 I started work on the first commercial search engine that supported "similarity searching." This was based on the Cosine Coefficient of weighted attribute vectors. As we got deeper and deeper into what made one search successful and another a flop, it became obvious that there were two prime enemies of successful searching.
For example, a query of "man versus machine" (actually run by Esther Dyson in our offices), produced an apparently random mix of articles about chess-playing computers, labor strikes over factory automation, and some guy towing a locomotive with his teeth (it's been over 20 years, so I don't remember the exact results). She hesitated for a moment, then said, "Oh! They really are all about people and machines, but how do I narrow it down to what I meant, which was more along the lines of the labor articles?"
We already had an answer to this by the time she visited us. Our solution (based on the work of Salton and Buckley) was to allow the user to say, "Results 1 and 3 are irrelevant, but result 2 is highly relevant." We would then take the attribute vectors for the articles and raise or lower the importance of those terms in the new query vector. After the first round of relevance feedback, the results often got downright spooky. Why? Because a) we now had a lot more terms to work with, and b) the additional terms helped to disambiguate terms like "bank." E.g., mention of levee construction probably meant this was a "river bank" and not a financial institution.
What does this have to do with what Google is doing? By reordering your results you are, in effect, giving user relevance feedback. I don't know what they are going to do with this information, but just getting your hands on it is a very important first step. And having you bring things to the top is doubly important, because positive relevance feedback is several times more important than negative. E.g. Tell me to drive towards San Francisco, rather than just saying I should leave Chicago.
On the downside, as mentioned by several previous posters, this opens the door to deep understanding of the user and what s/he is interest in. (Of course, they already get some of this when you click on items in a result list. They have a little JavaScript goody that records each and every click. Fortunately, there are GreaseMonkey scripts to disable this "feature.") One of my (few) customers back in the 80's and 90's said, "If you know only the questions I ask, you know too much." The customer? The NSA.