Mining requirements arguments from user forums

In order to sustain, software systems have to evolve in favor of its main target users. Due to the pervasive adoption of online user forums and social media, collecting users feedbacks and comments become possible. However, such crowd generated data are often fragmented, with various viewpoints mentioned during a series of message exchange. The aim of the thesis is to propose an argumentation-based CrowdRE approach, which represents such group conversations as a user argumentation model with the original conversation structure reserved. Based on the argumentation model, we are able to identify new features proposed by the crowd-users or issues encountered, and their supporting and attacking arguments using argumentation theory. To accomplish this research, we adopted an abstract argumentation, bipolar argumentation framework, and coalition-based meta argumentation framework. In addition, to provided automated support to our proposed approach, algorithms will be developed for bipolar argumentation, coalition-based meta argumentation, and end-users voting mechanism. Finally, this thesis employees different machine learning algorithms to automatically classify crowd-users comments into rationale elements and identify conflict-free features or claims based on their supporting and attacking arguments. Initial results show that the proposed approach can identify features, issues and their supporting and attacking arguments with acceptable performance.