A model-driven approach to generate relevant and realistic datasets

Abstract

Disposing of relevant and realistic datasets is a difficult challenge in many areas, for benchmarking or testing purpose. Datasets may contain complexly structured data such as graphs or models, and obtaining such kind of data is sometimes expensive and available benchmarks are not as relevant as they should be. In this paper we propose a model-driven approach based on a probabilistic simulation using domain specific metrics for automated generation of relevant and realistic datasets.