On the possibilistic-based decision model: Characterization of preference relations under partial inconsistency (Q5945813)

The authors extend the classical Dubois and Prade's decision model to cope with partially inconsistent descriptions of belief states, represented by non-normalized possibility distributions. Such descriptions appear frequently in the case-based decision theory, which assumes that the decision maker is provided with a set \(M\) of decision problem instances as triples (situation, decision, consequence) and with some similitarity measure Sim between situations. Given a new situation \(s_0\), possible decisions to take are proposed by the evolution \(\min_{(s,d,x)\in M} (\text{Sim}(s,s_0\rightarrow u(x)))\), where \(u\) is a preference function on consequences. The authors extend the possibilistic model based on this evaluation to take into account also belief states which are partially inconsistent. Their approach consists in suitably transforming the non-normalized distributions \(\pi\) into normalized ones. If \(\mathcal H(\pi)\) is the height of \(\pi\) and \(\mathcal N(\pi)\) is (standard) normalization of \(\pi\) (\(\mathcal N(\pi)(x)=1\) if \(\pi(x)=\mathcal H(\pi)\) and \(\mathcal N(\pi)(x)=\pi(x)\) otherwise), this normalization \(\pi^*\) is such that \(\pi^*(x)=\max(\mathcal N(\pi)(x),n_V(\mathcal H(\pi)))\), where \(n_V\) is a suitable involution map. The authors prove several representation theorems concerning preference orderings on the set of non-necessarily normalized distributions by pessimistic and optimistic quantitative utility functionals \(QU^-,QU^+\).