This paper presents the STALKER knowledge base refinement system. Like its
predecessor KRUST, STALKER proposes many alternative refinements to correct
the classification of each wrongly classified example in the training set.
However, there are two principal differences between KRUST and STALKER. Fi
rstly, the range of misclassified examples handled by KRUST has been augmen
ted by the introduction of inductive refinement operators. Secondly, STALKE
R's testing phase has been greatly speeded up by using a Truth Maintenance
System (TMS). The resulting system is more effective than other refinement
systems because it generates many alternative refinements. At the same time
, STALKER is very efficient since KRUST's computationally expensive impleme
ntation and testing of refined knowledge bases has been replaced by a TMS-b
ased simulator.