Measurement-based Quality Assessment of Requirements Specifications for Software-Intensive Systems
Jakub Mikołaj Mund · mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) · 2017
Requirements engineering (RE) is the process of discovering, negotiating, analyzing, communicating and managing the purpose of an envisioned software-intensive system and is widely recognized as a success factor.To prevent project failures, the suitability of its results, predominantly in terms of the requirements specication, must be ensured.One promising and frequently advocated means to this end is the use of measurement.Despite the importance of RE and the eorts spent on developing metrics, measurement often fails in practice.According to our understanding and experiences, we see three fundamental reasons prohibiting the eective use of metrics for assessing the quality of the requirements specication, namely that we need a better understanding of (1) if, (2) what and (3) how to measure.This thesis contributes to those three areas.First, we provide empirical evidence on the signicance of requirements specication quality.Specically, we studied to what extent requirements specications are created and used for communication, and revealed that it depends on certain criteria, e.g., whether the system is safety-critical or the involved parties collaborated in the past.Furthermore, we observed that factual faults in specications led to awed test cases in at least 47% of all cases (depending on the ease of discovering the aw) while negated sentences, which are commonly purported to impair understandability, did have no eect.We concluded that assessment of the requirements specication is an eective means in specic situations, but we have to carefully select which properties to assess and must not neglect the context of use.Second, in order to understand what to measure, we studied what properties of the requirements specication are (i) relevant and (ii) measurable.Regarding the former, we contribute a novel and systematic method to dene quality based on the activities which depend on the requirements specication, and, based thereon, a comprehensive quality denition model.Regarding the latter, we propose a meta-model which encompasses the necessary concepts associated with measures.Furthermore, we identied 136 measures from 2397 scientic publications in an extensive systematic literature review, which we analyzed in-depth to consistently (re-)evaluate crucial aspects, such as the actually measured attribute, threats for validity, or necessary prerequisites for application.Based on the results of the literature survey and the subsequent analysis, we instantiated the meta-model to obtain a unied and comprehensive assessment model.Moreover, by contrasting the quality denition and assessment models, we contribute an evaluation of (the limitations of) measurability which also helps us to understand the utility of the assessment approach.Finally, we provide a process model for measurement-based quality assessment of requirements specications complementary to the assessment model and developed specically with requirements specication quality in mind.The process model covers planning, assessment and continuous improvement phases, and includes a tailoring procedure to customize it for a specic project.Our results conrm that measurement-based quality assessment is no silver bullet, but rather constitutes a particular means to provide useful estimations of certain aspects of quality in specic situations.v Firstly, I want to express my sincere gratitude to Prof. Dr. Dr. h.c.Manfred Broy, not only for providing guidance on this thesis but also for giving me the opportunity to experience research in an environment which provides both individual freedom and stimulating partners from academia and practice to collaborate with.As for my next superior, the bar is set quite high.Also, I