8 Related Work

Subrahmanian 14] uses annotated logic 2] to integrate a set of deductive databases in an unique amalgamated database called amalgam. The amalgam, in addition to the disjoint union of the databases, contains a supervisory database. The supervisory database is a set of clauses (called amalgamated clauses) which resolve connicts due to inconsistent facts and compose uncertain information of diierent database sources. 14] investigates the relation between the models of the amalgam and the models of its components. Subrahmanian takes a more general approach then ourselves as he considers formulas with complex sets of truth values and time intervals. However the intuition behind amalgamated clauses (contained in the supervisory database) is very close to that of a generalization of view constraints described in 6]. From our perspective adopting a global amalgamated database is the reason of the main drawback of Subrahmanian's approach. Indeed global amalgamated database prevents one to associate distinct deductive mechanism to each database in the system. Furthermore amalgamated database doesn't support local inconsistency. I.e. the inconsistency of a local database forces the inconsistency of the whole amalgamated database. Diierently, our approach allows an integration information state in which the i-th database is inconsistent while the other are consistent. Vermeer et al. 16] exploit the semantic information provided by the integrity constraints of the single databases to achieve interoperability among them. The spirit of this approach is similar to ours, although they mainly address a diier-ent problem. In 16] diierent databases are integrated in an unique integrated view. The consistency of such an integrated view is checked by using integrity constraints of component databases. Vermeer et al. argue that semantic relations are expressed by relationships between objects (cf. domain relations) and relations between classes are the result of object relationships (cf. deenition of satissability of domain/view constraints w.r.t. a domain relation). 9 Conclusions In this paper we have provided a formal semantics, called LMS, for the integration of information of diierent agents. We have provided two examples from the electronic commerce scenario which involve many of the challenging aspects of information integration. We have deened LMS introducing the key concepts of domain constraint and view constraint which formalize how the information of diierent agents is integrated. Then we have argued that LMS is an adequate formalism for information integration in electronic commerce by formalizing the examples presented in the paper. Finally we have compared LMS with the main formalisms …