AG Netzbasierte Informationssysteme (Networked Information Systems) | Freie Universität Berlin

MOMA Framework

FU Berlin

MOMA-A

The (automated) MOMA Framework (MOMA-A) allows a more flexible, automatically triggered usage of various matching algorithms, depending on their suitability to particular phases of the ontology management process. Due to its generic and automatic character the approach can be applied in a service-oriented context, in order to enable the discovery and operation of appropriate matching services required to deal with specific, (previously unknown) ontologies.

The MOMA-A Framework uses semantical descriptions of both single matching algorithms and Web ontologies, which are then related by means of rules to detect the suitable (with respect to the given ontologiecal input) matching approach.

Our MOMA-A Framework consists of the following components (see MOMA-A Architecture):

  • matcher repository (internal matcher repository + external matcher provider) with reusable matching components

  • matcher metadata describes the properties of single matchers, e.g. information concerning the ontological formats served by particular matches, information regarding match results that can be delivered or the natural language that the algorithm can handle
  • ontology repository manages the matching inputs defined by the ontology metadata

  • ontology metadata describes the ontologies to be matched

  • rule repository links ontology and matching properties and helps to determine which matching algorithms is to be used for which type of ontologies

  • selection engine is responsible for the process of determination which algorithms are applicable to a specific set of inputs

Matcher Metadata The MOMA-A Matcher Metadata captures information about existing ontology matchers. For the classification of the algorithms we rely (in the first version) on the work of Rahm & Bernstein who make the distinction between individual and combining matchers.

Beside this classification, the metadata model includes matching characteristics like input type (instances or schemas, additional input in the form of e.g. numerical values), matching level (atomic level, e.g. attributes in an XML schema and non-atomic level e.g. XML elements) and cardinality (i.e. whether a matcher compares one or more elements in one schema with one or more elements in a second schema).

The preliminary version of the matcher metadata (matchingmetadata.owl ) is available online.

Ontology Metadata The matcher inputs (i.e. ontologies) are described using the metadata model see "Using Context Information to Improve Ontology Reuse", pdf), which can be applied to describe ontologies in various phases of their life-cycle. Accounting for the fact that matching algorithms cannot be applied with the same success expectations regardless of any dimension of the ontology metadata model, we have identified the following ontology features as relevant for matching tasks:

  • syntactic features such as the number of specific ontological primitives that affect the matching execution performance and quality of the structured-based matchers that typically perform better on simple graph structures.

  • semantic features such as modelled domain, representation and natural language, level of formality, type of model, domain generality that restrict the number of applicable matching algorithms, which might be adequate for a subset of these features.

Rules In order to automatically infer which algorithms suit to certain inputs, it needs explicit knowledge regarding the dependencies between these algorithms and the structures on which they operate. We formalize this knowledge into dependency rules—statements that determine which elements (in this case which matchers) are to be used or excluded.

The MOMA-A Rules are for example:

  • Apply only matchers which are able to deal with the representation language of the inputs
  • Use structure-based matchers for ontologies with different natural languages
  • Use constraints-based matchers only for formal ontologies and only if ontologies contain axioms
  • Apply only scheme matchers if no instance data is available

Selection Engine For a given pair of ontologies to be matched, the MOMA selection engine has to decide which matching algorithms should be applied to obtain the desired outputs.

The engine is aware of background information describing the available matching services and the properties of the input ontologies. However, in order to automatically infer which algorithms suit to concrete inputs, it needs explicit knowledge about the dependencies between these algorithms and the structures they operate on. We formalize this knowledge in terms of generic dependency rules - statements that determine which elements (in this case which matchers) are to be used or excluded.

©2006 Freie Universität Berlin