Text analysis for requirements engineering

Requirements Engineering is the Achilles’ heel of the whole software development process. It involves many stakeholders and includes not only technical but also sociological and psychological activities. Even when all the stakeholders come to a consensus, the produced requirements are rather informal. In the early project phases the functionality of the prospective software is not yet understood in the precision necessary for formalization, which makes requirements formalization not only a refinement, but also a learning process. As the survey by Mich et al. [MFN04] shows, the overwhelming majority of requirements are written in natural language. In practice these documents are often vague and contain a lot of ambiguities, which causes misunderstandings between project stakeholders. Misunderstandings and errors of the requirements engineering phase propagate to later development phases and can potentially lead to a project failure. To alleviate misunderstanding and to support the step from informal requirements to a formal model this thesis proposes a novel approach to the extraction of a domain ontology from requirements documents in order to establish a common language for the project stakeholders. An ontology consists of a set of terms and relations between these terms. As compared to a glossary, a domain-specific ontology gives a more explicit definition of terms and relations between them. When the ontology is extracted, a domain expert validates it. The validated ontology becomes both the common language for all the project stakeholders and a valuable resource for later development steps. The thesis makes two key contributions to ontology extraction as a part of requirements analysis: • It implements a semiautomatic method, extracting an ontology from a requirements document and validating the extracted ontology. • It shows how traditional requirements analysis process should be modified to include ontology extraction and validation. The feasibility of the proposed approach was evaluated on three comprehensive case studies.