Dogpile compares the results from leading search engines and results of top
search keywords are surprising, only 1.1 % of results are common.
Thanks,
Toufeeq
Yes, this is interesting. Although the main motivation for dogpile seems to
be to say it is better than other engines,
the scenario does bring up several related issues:
1. Whose top-10 list is actually more relevant on the average? (I
understand there is a
search engine comparison metric--a sort of Nielsen of search
engines--which does this
with a random keyword list and a random set of users. Hemal--which is
this metric?)
2. Is it really the case that combining all top-10 lists and give a top-20
list will improve relevance?
(Think in terms of precision..)
[Back in the pre-history of search engines, an engine called info-seek
will give you top 50 results
free and will ask you to pay money to get the next 1000 or so
results. Do you think info-seek made any money?]
3. Suppose--as is often the case--that different search engines are better
at providing relevant results for different types of
queries. Suppose you are a meta search engine who gets to look at the
query and decide to call upto k search engines
(for some small k) such that you are likely to get the most number of
relevant results in the union of their first pages.
How would you go about picking the k engines?
If this problem interests you, take a look at
http://rakaposhi.eas.asu.edu/icde05.pdf