On Sun, Apr 11, 2010 at 6:19 PM, Ziko van Dijk <zvandijk(a)googlemail.com> wrote:
Hello,
Gregory (? if I remember well) mentioned in August 2009 this:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1446862
All examined sites spy on their visitors, but Wikimedia and Wikipedia.
It's possible to track click progress without setting tracking
cookies. However.
You can form an {IP address, Useragent} tuple for every search then
make the assumption that subsequent page loads are the same client.
This is less accurate than cookie based full tracking, but it should
be sufficient for training a machine learning system for predictive
search results. Especially if we make the reasonable assumption that
users at the same location are already more likely to be looking at
similar materials.
You could also insert a tracking token in the search result HTTP get,
which would give you very accurate data but only for the clicks
directly off the search page.
Again, not a maximum amount of information, but likely sufficient and
it doesn't involve any deep privacy violating tracking.