Semantic Coordination of Heterogeneous Classifications Schemas

One of the key challenges in the development of open distributed systems is enabling the exchange of meaningful information across applications which (i) may use autonomously developed schemas for organizing locally available data, and (ii) need to discover relations between schemas to achieve their users’ goals. Typical examples are databases using different schemas, and document repositories using different classification structures. In restricted environments, like a small corporate Intranet, this problem is typically addressed by introducing shared models (e.g., ontologies) throughout the entire organization3. The idea is that, once local schemas are mapped onto a shared ontology, the required relations between them is completely defined. However, in open environments (like the Web), this approach can’t work for several reasons, including the difficulty of ‘negotiating’ a shared model of data that suits the needs of all parties involved, and the practical impossibility of maintaining such a shared model in a highly dynamic environment. In this kind of scenarios, a more dynamic and flexible method is needed, where no shared model can be assumed to exist, and semantic relations between concepts belonging to different schemas must be discovered on-the-fly. In other words, we need a sort of peer-to-peer form of semantic coordination, in which two or more semantic peers (i.e. agents with autonomously developed schemas and possibly heterogeneous ontologies) discover relations across their schemas and use them to provide the required services. In this paper, we propose a general approach to the problem of discovering mappings across the schemas of two or more semantic peers. The method we propose is intrinsically semantic, as the mappings it discovers between nodes of different schemas are computed as a logical consequence of (1) the explicit representation of the meaning of each node in the schemas, and (2) additional background knowledge (if available). The method is illustrated and tested on a significant instance of the