Detecting Bad Smells Of Software Requirements Defects

With the rapid development of computer science and information technology, the impact of software on people’s lives continuing to expand, providing them with a lot of convenience, people will be more and more dependent on software. In recent years, users have more and more demands on software they use, which results in the growing complexity and sizes in software. All sorts of problems and defects will eventually emerge, and these problems may bring damage to people’s lives and properties. Many case studies and investigations show that the most significant contributing factor to software poject failures is made at the Requirements stage. There may be many factors that can lead to software defects at the requirements stage,, including imperfect requirements elicitation, improper requirements analysis and management methods, or errors in the requirements specificaction documents, etc. During the entire software lifecycle, requirements specification plays an important role, and if there exists defects in requirements specification documents, the entire software development process will be affected, for example, a lof of time and financial resources, material and human resources will be wasted, which will eventually lead to the poor quality of software. Therefore, research on defects in software requirements is of greate importance in requirements engineering.Requirements specification usually involves natural language descriptions and graphical modelling. Natural language descriptions in requirements documents can be inherently vague or ambiguous, thus its meaning depends a lot on the background of the reader and there is no absolute standard for judging correctness, so these all thes factors may potentially create requirements defects. Researchers have found that these defects are difficult to detect, which will results in incorrect descriptions of functional requirements, and other members of the development team having difficulty understanding the real requirements. To face these challenges, researchers have proposed an approach that is based on what we call "requirements bad smells", which are concrete symptoms for requirements defects. There are eight types of bad smells in natural language specifications. A common graphical modelling technique in requirements specification is the Unified Modeling Language(UML), which contains no measure of right and wrong itself, so the degrees of freedom is bigger when designing a UML diagram. UML directly effects the coding of software. Detecting defects in UML is difficult, especially in the process of transforming natural language descriptions into UML models. In this paper, we firstly investigate how requirements bad smells in natural language are detected and then proposes four criteria for detecting bad smells in UML notations. An evidence-based research method known as the Systematic Literature Review (SLR) is adopted to collect evidence on reported defects in software requirements specifications. SLRs are more rigorous than traditional literature reviews, because they can help eliminate bias caused by subjective factors and ensure the relative completeness in understanding the relevant literature. In this dissertation, we define the UML bad smells based on the defects found by the SLR.In this dissertation, we expand the notion of bad smells in requirements specification, and provide a method of detecting bad smells of requirements defects in UML. The work in this dissertation can potentially improve the quality of requirements specification, so the work is valuable. The main contribution of this work can be summarized from three aspects below:(1) Providing an evidence-based systematic review on software requirements defects, which lays a solid foundation for our definition of UML bad smells.(2) Defining the UML bad smells by combing with the requirements specification and software engineering design rules.(3) Evaluating the UML bad smells by a case study and illustrate the method is feasible.