On the Use of Linear Combination in PWCP-Nets

Abstract

Conditional preference networks (CP-nets) are a compact but powerful formalism to represent and reason with qualitative preferences using the notion of conditional preferential independence. However, they suffer from incomparabilities between possible outcomes. Several works have attempted to overcome this weakness by quantifying CP-nets. This paper proposes a new approach combining two of the most interesting extensions of CP-nets, namely Probabilistic CP-nets (PCP-nets) using probability distribution to model uncertainty in different preference statements and Weighted CP-nets (WCP-nets) adding weights to express the relative importance of some attribute values regarding others. The new model so-called PWCP-nets combines the two models by handling both uncertainty and weights. Experimental results show the efficiency of this rich extension of CP-nets compared to PCP-nets and WCP-nets.