TU München - Fakultät für
Es sprechen Studenten Ã¼ber ihre abgeschlossenen Diplomarbeiten und Systementwicklungsprojekte.
Am Mittwoch, 14.03.18, ab 11:30 Uhr, im Besprechungsraum "PrÃ¤sident" am ZDB Garching:
Systematic Improvement of Requirements Smell Detection
In industrial practice, software requirements specifications (SRS) written in natural language are the most common kind of documentation. Quality defects contained in these SRS are the cause of increasing cost and delayed delivery time. Therefore, quality assurance (QA) is important and mostly done through reviews. Nevertheless, reviews are time consuming. Automated processes can support manual reviews of requirements engineers since they are cheap and can be integrated into daily work routine. However, the usage of automated quality detection tools is problematic since current research suffers from performance evaluation. We inspect the detection performance of Qualicen Requirements Scout (Scout) which is able to detect quality defects known as requirements smells. To evaluate the Scout, we created a text corpus manually annotated with respect to four different requirements smells. Thereby, we built a gold standard containing a total number of 281 findings for our study. We identified that the annotations of requirements smells are influenced by subjectivity. Using this gold standard, we calculated a precision of 14% and a recall of 26% combined over all smells. Afterwards, we analyzed the false predicted words and phrases for three of our requirements smells and updated the checklists of the Scout. For the forth requirements smell, we created a new detector using tf-idf and sentence lengths to identify complex sentences. With the improved Scout, we repeated our performance calculation and received a precision of 18% and a recall of 42%. In this thesis, we provide a detailed analysis of for requirements smells, evaluated an automated smell detection tool, improved this tool and created a gold standard for QA which is made available to the public.