PUB - Publikationen an der Universität Bielefeld

Requirements are an integral part of industry operation and projects. Not only do requirements dictate industrial operations, but they are used in legally binding contracts between supplier and purchaser. Some companies even have requirements as their core business. Most requirements are found in textual documents, this brings a couple of challenges such as ambiguity, scalability, maintenance, and finding relevant and related requirements. Having the requirements in a machine-readable format would be a solution to these challenges, however, existing requirements need to be transformed into machine-readable requirements using NLP technology. Using state-of-the-art NLP methods based on end-to-end neural modelling on such documents is not trivial because the language is technical and domain-specific and training data is not available. In this paper, we focus on one step in that direction, namely scope detection of textual requirements using weak supervision and a simple classifier based on BERT general domain word embeddings and show that using openly available data, it is possible to get promising results on domain-specific requirements documents.