Automatic Requirement Classification: Tackling Inconsistencies Between Requirements and Regulations
Current Requirement Engineering research must face the need to deal with the increasing scale of today's requirement specifications. One important and recent research direction is handling the consistency assurance between large scale specifications and many additional regulations (e.g. national and international norms and standards), which the specifications must consider or satisfy. For example, the specification volume for a single electronic control unit (ECU) in the automotive domain sums up to 3000 to 5000 pages distributed over 30 to 300 individual documents (specification and regulations). In this work, we present an approach to automatically classify the requirements in a set of specification documents and regulations to content topics in order to improve review activities in identifying cross-document inconsistencies. An essential success criteria for this approach from an industrial perspective is a sufficient classification quality with minimal manual effort. In this paper, we show the results of an evaluation in the domain of automotive specifications at Mercedes-Benz passenger cars. The results show that one manually classified specification is sufficient to derive automatic classifications for other documents within this domain with satisfactory recall and precision. So, the approach of using content topics is not only effective but also efficient in large scale industrial environments.