Synthesis of Model Transformations from Metamodels and Examples

Student thesis: Doctoral ThesisDoctor of Philosophy

Abstract

Model transformations are central elements of model-driven engineering (MDE). However, model transformation development requires a high level of expertise in particular model transformation languages, and model transformation specifications are often difficult to manually construct, due to the lack of tool support, and the dependencies involved in transformation rules.
In this thesis, we describe techniques for automatically or semi-automatically synthesising transformations from metamodels and examples, in order to reduce model transformation development costs and time, and improve model transformation quality.

We proposed two approaches for synthesising transformations from metamodels. The first approach is the Data Structure Similarity Approach, an exhaustive metamodel matching approach, which extracts correspondences between metamodels by only focusing on the type of features. The other approach is the Search-based Optimisation Approach, which uses an optimisation algorithm to extract correspondences from metamodels by data structure similarity, name syntax similarity, and name semantic similarity. The correspondence patterns between the classes and features of two metamodels are extracted by either of these two methods. To enable the production of specifications in multiple model transformation languages from correspondences, we introduced an intermediate language which uses a simplified transformation notation to express transformation specifications in a language-independent manner, and defined the mapping rules from this intermediate language to different transformation languages.

We also investigated Model Transformation by Examples Approach. We used machine learning techniques to learn model transformation rules from datasets of examples, so that the trained model could generate target model from source model directly.

We evaluated our approaches on a range of cases of different kinds of transformation, and compared the model transformation accuracy and quality of our versions to the previously-developed manual versions of these cases.

Key words: model transformation, model-driven engineering, transformation syn-thesis, metamodel matching, model transformation by examples
Date of Award1 Mar 2024
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorKevin Lano (Supervisor) & Steffen Zschaler (Supervisor)

Cite this

'