Search Engine Technologies name ______________
CS 5319, Spring 2007
Nigel Ward
Inclass Exercise 3: Rank Merging
Suppose you are working for an information exploitation company, and one day your boss comes to
you excited. “Look at this”, she says, “We’ve discovered a new algorithm that is good at classifying
user queries: it can reliably separate them into type X and type not-X. This is good news, because
type X queries are important to our business, and since we also know that queries of type X are
generally handled better by mamesearch.com, than by any other search engine!”
“Great,” you say. “Then we can just route all X-type queries to mamesearch.com and make our
analysts happy. It seems too good to be true.”
“It is,” she replies. “On most type X queries, google still returns some useful documents not seen in
mamesearch, and ranks some good ones higher. We think that google is doing a better job of finding
value in long documents. So here’s your next assignment: find a way to order the results for type X
queries that uses all the information. Here’s some data to get you started. Give me a first-pass
solution in 20 minutes. Oh, and by the way, whatever solution you come up with, give me a rough
analysis of how likely it is to work for general type X queries, and tell me what you could do to
improve the solution if we had more time.”
She leaves, dropping the following data on your desk, and saying over her shoulder “sometimes
people say that utility is proportional to the reciprocal of rank, for some queries on some engines, in
case that helps.”
query: “snowboarding vacations near West Texas” (a typical type X query)
mamesearch google desired document
results results results lengths (kilobytes)
1. A V A A 1
2. B Y B B 2
3. C D V C 9
4. D A D D 6
5. E F Y E 1
6. F G F F 1
7. G B G G 2
8. H Q E H 2
9. I N Q I 1
10. J Z C J 1
N 4
Q 3
V 9
Y 26
Z 1