Generating Examples for Knowledge Abstraction in MDE: a Multi-Objective Framework

Model-Driven Engineering (MDE) aims at raising the level of abstraction in software development and therefore relies on task automation. To foster automation, MDE promotes the use of specific domain languages (DSLs), essential to express ideas at the domain level. Furthermore, to ease communication between computer science and other fields, modelers employ model examples (i.e., selected metamodel instances) to illustrate and refine their conceptual ideas. But, if the use of model examples has shown its efficiency, it is still an ad hoc process which requires automation. In this paper, we briefly depict the thorough example-toknowledge learning process. Then, we present a framework that produces, from a metamodel, a representative model example set with regards to a given coverage definition. To find the best trade-off between coverage and a necessary minimality objectives, we use a non-dominated genetic algorithm (NSGAII). We illustrated our method by generating a near-optimal set of models for the peculiar constraint learning task. We evaluated its efficiency comparing the resulting generated set with the best one issued from a raw random generation. Our encouraging preliminary results let us envision a deep study of the relation between various types of coverage and their impact on our ability to abstract knowledge from examples.