Next: Interface Components
Up: Extensions
Previous: Playback logs
One of the difficulties in normal
IR experiments is obtaining really comparable data. If the value of a
particular function is tested by giving it to one group of users and not
to another and then measuring their relative success, there is always a
possibility that other factors have obscured the difference which is the
subject of investigation. However some comparisons do lend themselves to a
''blind testing'' approach. In the CILKS project for example, searches
were done using both the original and the enhanced query and the results
were merged before being presented to the user for judgement, allowing an
objective test of the value of query enhancement. Code to support similar
dual query formulation and hit-list merging exists in the ENQUIRE query
layer, but it has not yet been used experimentally.
PAYNE A
Wed Jul 3 14:11:32 BST 1996