Automating Feedback Analysis to Support Requirements Relation and Usage Understanding

Software development often faces a gap between developers' assumptions and users' real needs. While direct user involvement is valuable, it is often impractical, making online user feedback a crucial but challenging resource due to its unstructured nature. This dissertation addresses two main challenges: identifying which functionalities users discuss in their feedback and understanding how users interact with them. To tackle these, two machine learning–based approaches are proposed: one relates user feedback to existing software requirements, and the other extracts detailed usage information using the TORE framework. Following a Design Science methodology, the thesis includes systematic mapping studies, the design and evaluation of automatic classifiers, and the development of a supporting software prototype, Feed. UVL, along with a Jira plugin to integrate into existing workflows. The contributions include new methods for feedback analysis, evaluated classifiers, annotated dataset s, and insights into current research in the field.