A unified ontological framework for semantic integration

Abstract

In order to understand, evaluate, use and develop further semantic integration approaches in the geographic domain, it is essential to establish a proper ontological framework. This paper presents such a framework putting emphasis on three processes: (a) semantic information extraction, (b) concept/ontology comparison with identification and resolution of heterogeneities, and (c) ontology integration. These processes are further analyzed on the basis of their principal characteristics. These include the assumptions made, the semantic level addressed, the method employed, the input/output components, the distortions caused, and the user involvement required (as an indication of objectivity). The framework is exemplified by the analysis and comparison of existing integration approaches. 2. A “lower” design/implementation perspective with an interest in formalizing, processing and associating existing information or data. Very important here are database issues, attributes, structural, schematic and syntactic issues, as they relate to explication differences. While the two perspectives can and should be complementary, there is a systematic distortion of the notions from the first perspective to the second. As a result, in various approaches and projects, the “ontological concepts” and “semantics” are often database elements with data type definitions. This is usually treated with some sort of formalization coupled with elementary to more sophisticated conversion utilities. The real needs of the first perspective (semantic integration) are not fulfilled. Based on several survey results (Uschold & Jasper 1999, Wache et al. 2001, Ding & Foo 2002), when concepts from different ontologies are to be associated (mapped, merged, integrated, etc.), most of what many approaches offer today, are editing tools for a manual mapping by experts based on brainstorming. Some (Hameed et al. 2002) accept the position that none of the available tools can resolve all types of discrepancies, focusing thereinafter in interoperability between the tools. On the other hand, there are also noteworthy approaches addressing more semantic issues. In order to understand better what these approaches offer, it is of utmost importance to establish an ontological framework for semantic integration. This is the objective and contribution of the present paper. Attempting to develop an overarching framework for every aspect of the problem would be an extremely difficult task, possibly of limited practical value. A more rational approach is to view such a framework from a specific view point. Several such frameworks have been presented in literature (Visser et al. 1998, Wache et al. 2001, Klein 2001, Denny 2002, Ding & Foo 2002). All these frameworks mainly focus on explication mismatches of ontologies, paying more attention to structural and relation differences, than to semantics of concepts as expressed in different resources such as terms, definitions and subsumption relations. Our experience with geospatial integration shows that it is very useful to distinguish three subprocesses and examine approaches on the way they deal with concept semantics in each subprocess. This results into the framework presented in Section 4. Some of the above realizations generally hold for various ontology types and concepts. We are however more interested in those pertaining to the geographic domain (Kokla & Kavouras 2001, 2002, Hakimpour & Timpf 2002, Fonseca et al. 2002, 2003). The work presented here draws from previous work by the OntoGeo Group (http://ontogeo.ntua.gr/) on various ontological issues presented elsewhere, such as: extraction of semantic relations and properties, integration of geographic ontologies, association to top-level and basic-level ontologies, semantic similarity, ontological vagueness, context/scale and granularity issues, and guidelines to ontology generation.