Claria(.com) has put out a new product called RelevancyRank


"The technology analyzes what happens after a search results page is displayed, incorporating metrics such as the location of each listing on the results page, click rate, and post-click consumer behavior - such as time spent viewing a site, number of pages viewed at a site, and number of return visits to a destination Web site."

I was wondering what you guys think of the new technology? Also, can I have your opinions on my RelevancyRank commentary?

I like the idea. I'd like to see it (among other things) incorporated into the SE. I think SE will have to adapt and find new ways to rank sites. Google is becoming a bit spammy in my opinion. More factors should equal more relevency.

Its been tried before, in the late 90's there used to be search engine called 'Direct Hit'. Here is the 1999 blurb from their website via the Internet Archive:

Rather than building an index from the "bottom-up" by adding sites one at a time, Direct Hit takes a "top-down" approach by refining the relevancy of any large body of data, such as that generated by conventional search technologies or an open site submission process where users are allowed to submit sites. To refine these results, the Direct Hit Popularity Engine anonymously monitors which web sites Internet searchers select from the search results list, how much time the searchers spend at these sites and a number of other metrics, such as the position of a site relative to other sites. The sites that are selected by searchers are boosted in their ranking, while the sites that are consistently ignored by searchers are penalized in their rankings.

There are lots of reasons why it doesn't work as a stand alone procedure not least that nearly half of all searches are unique, so there is no clickthrough data.

That is not to say that some kind of clickstream analysis is not being folded back into the Google algo. However the last time I asked a Google engineer if they used click data they said "not at the moment but they had looked at it." I took this to mean that they had experimented and not found it useful for improving the quality of the serps.

- Michael

Terrific answer Michael!

I didn't know this had been tried before. It would be interesting to see what Google's results had been when they tested it (assuming your interpretation of the engineer's words is correct).

Incidentally, I was thinking: won't such click analysis mean initially top ranking sites are reinforced? Given that most people click what's one page 1... You'd have to randomly present results with no other ranking factors to see what was actually a better site. And it would have to be done with paid focus groups kept in the dark as to who the engine was, because doing it with normal searchers would cause loss of market share due to poor results during testing. And if the focus group knew who the engine was, similar bad worth of mouth would spread.