A logic for reasoning about relative similarity (Q677086)

In AI we often classify objects of a given universe \(\mathcal E\) with respect to a fixed set of properties PROP. We can talk about similarity of such objects with respect to a given subset \(P\) of properties, understood as a reflexive and symmetric binary relation parametrized by the above set \(P\). The approach to similarity adopted in the paper stems from the rough set approach. We consider the operations of lower and upper approximations \(\underline{\text{sim}}(P)E\), \(\overline{\text{sim}}(P)E\) of a set of objects \(E\) with respect to a set of properties \(P\), with \(\underline{\text{sim}}(P)E= \{e'\in{\mathcal E}: (\forall e\in{\mathcal E})\text{sim}(P)(e',e)\to e\in E\}\), \(\overline{\text{sim}}(P)E= \{e'\in{\mathcal E}: (\exists e\in E)\text{sim}(P)(e',e)\}\). The above operations give rise to a polymodal logic with modalities corresponding to these operations, whose formulae are interpreted as sets of objects, and said to be true if they evaluate to the whole universe. A sound and complete deduction system for the logic is developed. Technically, the ideas underlying the system express the set-theoretical membership relation in the logic, making use of a syntactic counterpart of the idea of components of a family of sets in order to deal with interdependent modalities.