Clickthrough, Themes and Factors that Affect Contextual Relevancy

[Marcia]

Reference has been made in a member's blog entry with regard to Google's use of clickthrough data in scoring, with recollections described of having seen a photograph of such a couple of years ago, with a detailed description of the photo:

Google Considers Visitors in Algo

As a matter of fact, I personally have a recollection of this, but something is mentioned by grnidone that I think warrants further thought, particularly with algo developments and changes in recent years - or months, that relates to seasonality and contextual relevancy.

One reason could be that the same search used in different times of the year may need to give different results. Ever notice that in early January a search for “rose” will get you results about the Rose parade while a search for “rose” in February will get you a flower company?

We all know what a problem it has always been at every search engine, when searching, to get back results with "contextual relevancy" for the specific topic of our own interest at the time. We can do a search for bows and find that there will be "buttons and bows" for square dancing groups, violin bows, hair bows, cheerleading bows, bows as in "curtsey to the queen" as well as bows and arrows as in archery.

A few questions arise:

1. If clickthrough data is collected and used in Google's algo, how long would it take for the SERPs to reflect relevancy based on seasonal factors?

2. Considering the alleged "sandbox" which may be nothing more than a collection of filters with a resultant phenomenon, how would seasonal relevancy in the SERPs be affected by the absence of sites due to delay?

3. How much of a role would sitewide considerations play in Google determining the context of a site as it relates to seasonality and traffic statistics, and returning results appropriate to context?

Then there's exact match or partial match. If someone is looking for a punch bowl for a wedding in April or May, will they still be getting Rose Bowl in the SERPs if there happen to be vague references on sites that don't relate to the theme?

What I'm getting at is that if clickthrough data were to be combined with historical data of search terms used for sites, and semantic similarities sitewide that reveal a site's theme were combined, could that at some time improve results and eliminate incorrect sites from being returned - based on semantic analysis, theming, historical and clickthrough data all combined?

Added:
Here's the Cornell paper on

Optimizing Search Engines using Clickthrough Data