Automatisierte Abhängigkeitsanalyse von Anforderungen zur Wirkkettenmodellierung

Effect chain modelling is a method for creating information models for impact analyses of changes in system elements. For the estimation of change propagation, dependencies between requirements must be detected. The high number of require- ment dependencies in the engineering of complex technical systems results in the need for automation. In a study, it was shown that transformer models (BERT) are suitable for the automated dependency analysis of requirements. However, there are currently deficits in the applicability of the models for different projects without an extensive and heterogeneous training database. This paper investigates how active learning can be used to train BERT models (active-BERT) in order to increase the performance of the models for classifying requi- rement dependencies of projects with heterogeneous require- ments. The results show that the performance of the models increases significantly through active learning. Through active- BERT, engineers are enabled to model effect chains efficiently and to handle requirement changes effectively.