« July 2005 | Main | September 2005 »

Inline related searches being tested at Google?

First brought to my attention by Mike Levin, who's quoted in this Clickz article is what appears to be some results page experimentation by Google.  My analysis of this phenomenon, based on the article's own "on demand" example, is that Google appears to have employed clustering technology of the type that Teoma uses.  The key difference though, and it's a big difference, is that Google isn't clustering related searches off to the side under a drilldown link.  Instead, Google is inlining the top three results from the related search "comcast on demand" in positions 6,7, and 8 of the results for "on demand".

Google already suggests search revisions at the top of the page if their engine thinks you've misspelled a word. 

Further evidence that "related search" technology is being used here was found when I went over to Google Suggest and entered "on demand".  One of the suggested searches was "on deman comcast" which yields the same top three results.

The Clickz article alludes to some kind of results QA going on.  That's pretty  wide open to interpretation- are they looking at the quality of the results for "on demand", or are they doing QA on this phenomenon as a standard feature of Google?

The click tracking of those three URLs is interesting as well but I'm scratching my head over their methodology.  The clickthrough URLs contain a parameter "oi=revisions_inline" implying that this is meant to suggest revised searches.  But the use of the horizontal rules to set apart these results visually may draw the users' eyes, and artificially inflate clickthrough with the result of distorting their QA results.

It will be interesting to see where Google's going with this. Are they inlining query revisions in order to learn how to figure out what users really mean when they enter a given query?  Are they simply testing an inline version of clustered topics ( if that's the case than clustering is no longer an appropriate term) to enhance their user interface? I'll keep an eye on where this discussion goes and will post updates as more information emerges.

Posted on August 19, 2005 | Permalink | Comments (0) | TrackBack

Google Alerts Employing Clickthrough Tracking

I haven't seen this discussed in the forums I frequent so I'm not sure
how long it's been going on. I receive several different Google Alerts
and the majority of them list out, in the email, direct links to the
page they are publicizing.

Today, however, I received an alert whose link was to a redirect app at
Google! The url contained several cryptic numeric parameters, along with
a parameter spelling out the alert's keyword search and of course the
destination URL (how else would a redirect work but to know the
destination).

Could Google Alerts be learning about site relevancy in this way? If
that's the case, does it apply back to regular search results? Surely
there must be an article somewhere...

Posted on August 4, 2005 | Permalink | Comments (0) | TrackBack