| | |
Summary: Document Filtering as an
Adaptive and Temporally-dependent Process
Avi Arampatzis Th.P. van der Weide
Faculty of Mathematics and Computing Science,
University of Nijmegen, Toernooiveld 1, NL-6525 ED Nijmegen, The Netherlands.
Tel: +31 24 3653147, fax: +31 24 3553450.
avgerino@cs.kun.nl
http://www.cs.kun.nl/avgerino
Abstract
The ltering task has traditionally been de ned as a special case of the information retrieval task, and
undeniably, it can be performed by applying retrieval techniques. This theoretical study summarizes
our experiences in viewing ltering as an adaptive and temporally-dependent process. A process that,
in contrast to traditional retrieval, takes into account the dynamic nature of relevance and its temporal
aspects. We investigate the nature of user interests, formulate useful types of adaptivity, and discuss
the e ectiveness of those types in relation to user interests. To deal with drifts, we introduce the
notion of the half life of documents. Furthermore, we discuss potential dangers for e ectiveness such as
selectivity traps. We pay special attention to practical eÆciency issues by discussing term selection and
incrementality.
1 Introduction
The digital and networking revolution over the last decade has made large amounts of digital information
|
