| | |
Summary: Unbiased S-D Threshold Optimization,
Initial Query Degradation, Decay, and Incrementality,
for Adaptive Document Filtering
Avi Arampatzis
Information Retrieval and Information Systems, University of Nijmegen,
Postbus 9010, 6500 GL Nijmegen, The Netherlands.
avgerino@cs.kun.nl, http://www.cs.kun.nl/avgerino
Proceedings of the Tenth Text REtrieval Conference (TREC-10).
Abstract
We develop further the S-D threshold optimization
method. Speci cally, we deal with the bias problem
introduced by receiving relevance judgements only
for documents retrieved. The new approach esti-
mates the parameters of the exponential{Gaussian
score density model without using any relevance
judgements. The standard expectation maximiza-
tion (EM) method for resolving mixtures of distri-
butions is used. In order to limit the number of doc-
uments that need to be bu ered, we apply nonuni-
form document sampling , emphasizing the right tail
|
