Let's look now at the user side of SEs. Assume that a user needs some
information on a certain topic. So he summons on a whim of a second his
favorite SE to search for any relevant information. The SE is invoked with an
ad-hoc query composed of a supposedly appropriate combination of keywords.
The SE will certainly return a lot of information (with low precision and recall),
which is bound to overload the user. He will then have to tediously sift through
it all and manually filter the supplied references. The relevant information
found will then be immediately used or temporarily kept in a cache for a
short-term period of use