Project Team

Manuel Wimmer,Horst Kargl,Martina Seidl,Petra Brosch

Abstract

Existing matching tools used for automating integration tasks are mostly based on schema-based information such as structure or name similarities. However, in various integration scenarios the schemas to be integrated do not use the same language, naming conventions, or jargons as well as equivalent structures, although they often represent the same semantics. In particular, when integrating domains-specific languages, these heterogeneity problems are quite common, because the main idea behind domain-specific languages is to employ domain-specific terms or jargons. For supporting also such integration scenarios, the SmartMatching approach has been developed. Summarized, the SmartMatching approach extends existing matching tools by a self-tuning component for improving the quality of automatically computed alignments. The self-tuning mechanism is realized by combining the idea of test-driven development and iterative machine learning approaches. Before actually starting with the computation of alignments, first the user has to define semantically equivalent test instances for the schemas to be integrated which can then be used for evaluating and adapting alignments until an appropriate solution for a given set of test instances is reached.

SmartMatcher Prototype

In the course of the ModelCVS project, a prototype for the SmartMaching approach has been developed which is called SmartMatcher. The prototype comprises a simple matching component for computing an initial alignment model, a mapping engine for adapting mapping models, and a fitness function based on the diff-operator for evaluating mapping models. The implementation is based on the Eclipse Modeling Framework (EMF). This means, schemas implemented in Ecore and instances expressed as EMF models are the required input formats. The mapping engine is highly configurable by a state machine for allowing experiments with various mapping strategies. An example mapping strategy for the CAR mapping language is shown in Figure 1. For observing the behavior of mapping strategies, we have developed a matrix-based view on the mapping problem (see Figure 2), where the first row and the first column represent the schema elements. The body of the matrix dynamically displays which alignments are evaluated to true or false as well as which alignment has been currently applied.

Core Features of the Prototype

  • Simple Initial Matcher (Name Similarities)
  • Import Function for Alignments in the INRIA alignment format
  • A configurable Mapping Engine (State-machine based)
  • Fitness Function (Diff-Operator for models)
  • Matrix Viewer for observing the behavior of the mapping strategies

Additional Features of the Prototype

  • Measuring the execution time
  • Export of CAR mapping models into Colored Petri Nets
  • Ecore2OWL Converter

depth-first-search

Figure 1: State Machine for default CAR mapping strategy.

mappingmatrix

Figure 2: MatrixViewer GUI

The UML 1.4.2 to 2.0 Case Study

In order to evaluate the Smart Matching approach a case studies has been conducted. Thereby a horizontal transformation scenario has been chosen for clarifying the value of the approach and to test the implementation. The horizontal scenario covers the integration of UML 1.4.2 with UML 2.0. In particular, this integration is needed to transform UML 1.4.2 models into UMl 2.0 models which can be loaded in tools only supporting the newest version of the UML standard.

Metamodels

Models

Mappings between UML 1.4.2 and UML 2.0 (class diagram part)

A summary of the necessary mappings between UML 1.4.2 and UML 2.0 for transforming existing UML models to the new standard can be found here.

Publications

  • H. Kargl, M. Wimmer: "SmartMatcher - How Examples and a Dedicated Mapping Language can Improve the Quality of Automatic Matching Approaches";
    in: "The Second International Conference on Complex, Intelligent and Software Intensive Systems (CISIS 2008)", IEEE Computer Scociety, 2008, pp. 879 - 885.
  • G. Kappel, H. Kargl, T. Reiter, W. Retschitzegger, W. Schwinger, M. Strommer, M. Wimmer: "A Framework for Building Mapping Operators Resolving Structural Heterogeneities";
    in: "Information Systems and e-Business Technologies", Springer, Lnbip 5, 2008, pp. 158 - 174.
  • T. Reiter, M. Wimmer, H. Kargl: "Towards a runtime model based on colored Petri-nets for the execution of model transformations";
    3rd Workshop on Models and Aspects - Handling Crosscutting Concerns in MDSD, in conjunction with ECOOP 2007, Berlin, Germany, 2007.
  • G. Kappel, H. Kargl, G. Kramler, A. Schauerhuber, M. Seidl, M. Strommer, M. Wimmer: "Matching Metamodels with Semantic Systems - An Experience Report";
    Model Management und Metadaten-Verwaltung Workshop, in conjunction with BTW 2007, Aachen, Germany, 2007.

Last Updated: 27.10.2008