Towards structural criteria for ontology modularization
Anne Schlicht, Heiner Stuckenschmidt · MADOC (University of Mannheim) · 2006
Recently, the benefits of modular representations of ontologies has been recognized by the semantic web community.Existing methods for splitting up models into modules either optimize for completeness of local or for the efficiency of distributed reasoning.In our work on semantics-based P2P systems, we are also concerned with the additional criteria of robustness or reasoning in cases where peers are unavailable and with ease of maintenance.We define a number of structural criteria for modularized ontologies and argue why these criteria are suitable for estimating efficiency, robustness and maintainability.We apply the criteria to a number of modularization approaches and discuss the trade-offs made.Based on the discussion we propose a general quality measure for modular representations in the context of our use case. MotivationThe problem of modularizing ontologies in the sense of splitting up an existing ontology into smaller, interconnected parts has recently been discussed by a number of researchers (see for example [10,1,9]).These approaches differ significantly in terms of the concrete goal of the modularization and consequently in terms of the criteria used to determine a good modularization.In our work we are concerned with the automatic modularization of ontologies to support a distribution of knowledge in a P2P network in such a way that following requirements are fulfilled:Efficiency Reasoning in the distributed system should be efficient.This also requires that communication costs, which are known to be a major bottleneck in distributed systems are minimized.Robustness In a P2P network, single peers can be temporarily or permanently unreachable.The impact of such failures on the completeness of reasoning should be minimized.Maintainability Ontologies evolve over time and the corresponding changes need to be propagated through the network.The number of changes and the required costs for computing necessary changes should be minimized.It is obvious that these are conflicting requirements that need to be balanced in order to determine an optimal strategy for modularization.Similar problems