Model comparison with GenericDiff

Abstract

This paper proposes GenericDiff, a general framework for model comparison. The main idea is to separate the specification of domain-specific model properties and syntax from the general graph matching process and to use composite numeric vectors and pairup graph to encode the domain-specific properties and syntax so that they can be uniformly exploited in the general matching process. Our initial evaluation demonstrates that it is easy to deploy GenericDiff in a new application domain and GenericDiff is able to produce an accurate comparison reports for diverse types of models.