Recommendations about improving the quality of requirements. In this case, the recommendations are related to:
1. Measuring the quality of requirements to detect bad quality requirements. Given the state-of-the-art presented in Section 2.2, we will focus on measuring quality aspects related to: i) textual properties of the requirements (e.g., ambiguity, incompleteness, redundancy), ii) structural properties of the requirements (e.g., conformance to template), and iii) stakeholder-related properties (e.g., engagement and motivation).
2. Improving the quality of requirements. For instance, by offering suggestions about rewording and standardized glossary to remove ambiguities and fill-in incomplete information, corrective actions to structure the requirements according to templates (if in use), and fostering stakeholders engagement through, for example, gamification elements. To accomplish both points we will use information available from the current and past projects, such as the textual representation of the requirements, their history, the template used (if any), and stakeholders preferences.
Recommendations about improving the quality of requirements. 1. Measuring the quality of requirements. A set of rules reflecting quality properties will be identified from existing related work (see Section 3.1) and adapted to OpenReq. These rules can then be used against requirements to check their quality, compounding them to calculate a quality score. The majority of the approaches to identify quality concerns leverage raw textual data (i.e., the requirement text itself). Therefore, NLP techniques will be heavily utilized. Among those, text chunking (Arora 2015) and part- of-speech tagging (Arora 2014) will be used to identify basic units of text. Later, the units will be analysed to find, for example, usage of passive voice (considered a bad quality indicator) (Xxxxxx 2014), or missing information (e.g., an active verb missing a direct object) (Xxxxxx 2016). Another possible use of NLP is the resolution of anaphora in the requirements text (Ferrari 2016) —i.e., identify references to the same entity, but using different names, within the text. Similarly, NLP and statistical machine learning algorithms (such as SVM and bayesian network) (del Sagrado 2017) (Xxx 2013) can be used to check conformance to requirements template. In addition, the quality of the requirements can be further analysed through personal ratings of stakeholders via several properties of the requirements, which could be used to improve stakeholders engagement and motivation. However, this special aspect of quality will be also dealt with in WP4.
2. Improving the quality of requirements. Simple word lists and thesauruses, together with NLP techniques can be used to build recommenders that can point out improvement actions on the basis of the previously defined rules. These improvement actions vary accordingly to the complexity of the task. For the case of passive voice a rewording in active voice can be done automatically (Xxxxxxxx 2016) (e.g., for the requirement in passive voice “Users should be alerted by the system when new documents are assigned to them.”, the following rewording will be automatically proposed “The system should alert users when new documents are assigned to them.”). In contrast, for other cases (e.g., ambiguities resolution, or incomplete information about a requirement) the recommender can only hint at the presence and location of a quality concern but human intervention is likely to be required (e.g., filling-in information gaps) (Xxxxxxxx 2016). NLP techniques (e.g., text chunking) are used to suggest how to fill-in mi...
