Diverse Scenario Exploration in Model Finders Using Graph Kernels and Clustering

Abstract

Complex software systems can be described using modeling notations such as UML/OCL or Alloy. Then, some correctness properties of these systems can be checked using model finders, which compute sample scenarios either fulfilling the desired properties or illustrating potential faults. Such scenarios allow designers to validate, verify and test the system under development. Nevertheless, when asked to produce several scenarios, model finders tend to produce similar solutions. This lack of diversity impairs their effectiveness as testing or validation assets. To solve this problem, we propose the use of graph kernels, a family of methods for computing the (dis)similarity among pairs of graphs. With this metric, it is possible to cluster scenarios effectively, improving the usability of model finders and making testing and validation more efficient.