Digital library access is driven by features, but the relevance of a f
eature for a query is not always obvious. This paper describes an appr
oach for integrating a large number of context-dependent features into
a semi-automated tool. Instead of requiring universal similarity meas
ures or manual selection of relevant features, the approach provides a
learning algorithm for selecting and combining groupings of the data,
where groupings can be induced by highly specialized features. The se
lection process is guided by positive and negative examples from the u
ser. The inherent combinatorics of using multiple features is reduced
by a multistage grouping generation, weighting, and collection process
. The stages closest to the user are trained fastest and slowly propag
ate their adaptations back to earlier stages, improving overall perfor
mance. (C) 1997 Pattern Recognition Society.