A framework for automated quality assessment of software requirement specification based on part-of-speech tagging, multi-agent k-means clustering and case-based reasoning

• Main Author: Jubair, Mohammed Ahmed
• Format: Thesis
• Language: English; English; English
• Published: 2022
• Subjects: T Technology (General)
• Online Access: http://eprints.uthm.edu.my/8438/1/24p%20MOHAMMED%20AHMED%20JUBAIR.pdf; http://eprints.uthm.edu.my/8438/2/MOHAMMED%20AHMED%20JUBAIR%20COPYRIGHT%20DECLARATION.pdf; http://eprints.uthm.edu.my/8438/3/MOHAMMED%20AHMED%20JUBAIR%20WATERMARK.pdf; http://eprints.uthm.edu.my/8438/

Software Requirement Specification (SRS) is an imperative process in a Software Engineering (SE) cycle, where its role is to document functional and non-functional requirements and to establish the tasks that a particular system is set to accomplish. Because a badly written SRS has an expensive impact on the entire project, the success or failure of any software product depends on the quality of the SRS document. Recent advancements in the field have explored automated extraction of quality attributes in SRS documents such as the Reconstructed ARM and the Rendex models. However, automating the quality assessment process poses major challenges, which requires advanced Natural Language Processing (NLP) algorithms to extract the quality features, interpreting the context of the features, formulating the assessment metrics, and documenting the shortcomings as well as possible improvements. Recent automated models also attempted to assess the quality of the SRS based on a small number of quality attributes and indicators due to the limitation in extracting quality attributes that require specific indicators from the SRS. To address this gap, this thesis proposes an Automated Quality Assessment of SRS (AQA-SRS) framework by integrating NLP for feature extraction, Multi-Agent System (MAS) with K-means for features clustering, and Case-based Reasoning (CBR) for process management. This framework assessed the SRS documents by automatically extracted 11 quality attributes and their corresponding 11 quality indicators through a deep analysis of the SRS textual content. This process is performed through the Multi-Agent K-means (MA-K-means) model for handling the automatic evaluation of the AQA-SRS framework. The performance of the AQA-SRS framework is evaluated by comparing the results against the state-of-the-art techniques as well as human experts based on two standard SRS datasets. The results showed the AQA-SRS framework reliably handled the assessment of 11 quality attributes and their corresponding 11 quality indicators with Krippendorff’s Alpha 0.78 for the agreement with software engineering experts.