2.3 Ontology-Driven Interoperability
2.3.6 Ontology Mapping
The continuing diversity of ontologies is partly related to ontologies being aligned with particular views of the world, resulting in biases and subjective features (Hameed et al, 2004). Ontology heterogeneity in design and manufacture also occurs as a result of interspersed knowledge at different stages of the product lifecycle. The examples of ontologies discussed in the previous section reveals this ongoing semantic heterogeneity. If these ontological models are to semantically interoperate, methods need to be devised to reconcile disparate ontologies.
The area of ontology mapping has been a key direction to tackle semantic heterogeneity issues across ontologies, with the intention of promoting semantic interoperability. Several overlapping views over categories of ontology mapping methods have been suggested (Kalfoglou and Schorlemmer, 2003; Noy and Musen, 2003; Euzenat and Shvaiko, 2007;
Liping et al, 2007). There is almost general consensus over the types of methods that can be applied in ontology mapping. Figure 2-5, partly adapted from Noy and Musen (2003), identifies and summarises these methods.
Ontology mapping methods include (1) techniques that focus on combining (merging) two ontologies to construct a new ontology from the individual ontologies, (2) tools that compile a transformation function that transforms a given ontology into another based on the transformation rules specified (Noy
and Musen, 2003), (3) methods that concentrate on establishing a collection of binary relations between the vocabularies of two ontologies (alignment) (Kalfoglou and Schorlemmer, 2003) and (4) methodologies that enable specific portions of two ontologies to be reconciled, through the definition of mappings via an intermediate articulation ontology. It is to be noted that although some researched ontology mapping methods fit very well into this category, others occur as hybrids of the common ontology mapping methods identified in Figure 2-5.
Comprehensive available literature reviews on ontology mapping and the related methods (Kalfoglou and Schorlemmer, 2003; Euzenat and Shvaiko, 2007) point to a large number of ontology mapping tools that have been either theoretically proposed or fully implemented and tested (Kent, 2000;
McGuinness et al, 2000; Maedche and Staab, 2000; Kiryakov et al, 2001b;
Stumme and Maedche, 2001a; Kalfoglou and Schorlemmer, 2002; Madhavan et al, 2002; Noy and Musen, 2003; Bach et al, 2004; Euzenat and Valtchev, 2004; Mitra et al, 2004). In the literature review exposed in this work, only the most outstanding and pertinent ontology mapping methods are documented.
The ontology MApping FRAmework (MAFRA) developed by Maedche and Staab (2000) is an ontology mapping method used for the reconciliation of distributed ontologies on the Semantic Web. MAFRA consists of five horizontal dimensions which relate to the implementation structural aspects of MAFRA and four vertical dimensions which focus on the more strategic
X Y
Merged XY Merging
X Y
Articulation Articulation
X Y
Transformation
X Y
Alignment
Figure 2-5 Common Methods Used for Ontology Mapping (Based on Noy and Musen (2003))
perspectives pertaining to the framework (see Figure 2-6). Following the MAFRA approach, the first step in ontology mapping is that of (D) lift and normalisation where all information to be mapped are set onto the same RDF(S) representation platform. Lexical similarities are analysed in stage (E) and, then, based on the similarities found between the source and target ontologies, the “Semantic Bridging” module (F) establishes necessary correspondences (Kalfoglou and Schorlemmer, 2003). These semantic bridges are then executed (G), verified and enhanced in the final stage (H).
The FCA-Merge (see Figure 2-7), presented by Stumme and Maedche (2001a), is another important ontology merging environment. Unlike similar ontology merging tools which tend to exclude instances during semantic reconciliation, it is said that FCA-Merge in fact extracts meaningful information from classified instances. The merging process realised in FCA-Merge comprises three vital steps. The first consists of the extraction of instances and the computation of two formal contexts where the ontologies reside. An information extraction technique known as SMES (I) (Saarbrucken Message Extraction System) (Neumann et al, 1997) is used for this purpose.
The fundamental infrastructure underneath the second phase of the mapping process is the generation of a single context and the computation of the pruned concept lattice (J). This is performed using the FCA-Merge algorithm, known as “Titanic” (Stumme et al, 2000), which is attuned to fit the needs of the FCA-Merge environment. Both the first and the second stages are claimed
(D) Lift and Normalisation (E) Similarity (F) Semantic Bridging
(G) Execution (F) Post-Processing
Domain Knowledge and Constraints
Evolution Cooperative Consensus Building GUI
Figure 2-6 Conceptual Architecture of MAFRA (Redrawn from Maedche et al (2002))
to be fully automatic processes. The third stage, which is semi-automatic, involves an interactive user interface built on top of the OntoEdit tool (K). In order to support the knowledge engineer in the different steps, there is a number of queries for focusing his attention to the significant parts of the pruned concept lattice (Stumme and Maedche, 2001b).
Noy and Musen (2000) initially proposed an algorithm and tool to promote ontology merging and alignment. The authors have later exposed a complete suite of tools integrated in the “Prompt” suite (Noy and Musen, 2003), covering various functionalities for multiple-ontology management. The
“Prompt” suite comprises (1) “IPrompt” for interactive ontology merging, (2)
“AnchorPrompt” for graph-based mapping, (3) “PromptDiff” for ontology versioning management and (4) “PromptFactor” for factorising out semantically independent sub-ontologies.
“IPrompt”, which forms part of the algorithm-driven semi-automatic ontology merging feature of “Prompt”, is responsible for providing suggestions for merging, identifying inconsistencies, resolving potential problems and exposing strategies to solve these (Noy and Musen, 2003). During the
Figure 2-7 FCA-Merge Interaction Environment (Based on Stumme and Maedche (2001b))
comparison of two ontologies, “IPrompt” analyses small segments of the ontology graph around each concept prior to proposing appropriate merging decisions. Overall, the “Prompt” suite remains a comprehensive semi-automatic toolkit for coping with semantic reconciliation.
Researched and validated ontology mapping tools indicate that there is currently no ontology matching technique that uses the semantics of logic-based systems that employ upper ontologies (Euzenat and Shvaiko, 2007).
Moreover, it is evident, from experiments based on current ontology mapping methods, that ontology mapping has not been given due attention in design and manufacture primarily since the latter remains an expert domain with very specific content and issues (Chungoora and Young, 2008b). Hence, this work additionally addresses the relevance of semantic-based mapping methods for aiding semantic interoperability in product design and manufacture.