On the Suitability of Generalized Behavioral Profiles for Process Model Comparison
Given two process models, the problem of behavioral comparison is that of determining if these models are behaviorally equivalent (e.g., by trace equivalence) and, if not, identifying how can the differences be presented in a compact manner? Behavioral profiles have been proposed as a convenient abstraction for this problem. A behavioral profile is a matrix, where each cell encodes a behavioral relation between a pair of tasks (e.g., causality or conflict). Thus, the problem of behavioral comparison can be reduced to matrix comparison. It has been observed that while behavioral profiles can be efficiently computed, they are not accurate insofar as behaviorally different process models may map to the same behavioral profile. This paper investigates the question of how accurate existing behavioral profiles are. The paper shows that behavioral profiles are fully behavior preserving for the class of acyclic unlabeled nets with respect to configuration equivalence. However, for the general class of acyclic nets, existing behavioral profiles are exponentially inaccurate, meaning that two acyclic nets with the same behavioral profile may differ in an exponential number of configurations.