In this report, we describe investigations made in the context of user-adaptive in- 
formation systems. We verify the idea of combining two formalisms: description logics 
(DLs) and Bayesian networks in order to increase the effectiveness of information re- 
trieval. For this, we implemented the basic functionality of P-Classic which extends 
the DL Classic with probabilistic inferences. In P-Classic, the degree of concept 
subsumptions can be quantitatively expressed as a statistical value. We use this feature 
for the definition of the PLCS operator, a probabilistic "Least Common Subsumer" 
operator which allows for quantitative measure of concept overlap, and we show how 
this operator can improve the quality of information retrieval. 

The software package including the prototypical implementation of P-Classic in 
Common Lisp can be obtained from http://www.sts.tu- harburg.de/~r.f.moeller/band/ 
pclassic.zip.