Automatising defect management tasks for app developers: a Machine Learning approach

User feedback is a valuable source of information for app developers. in this context reviews on app stores can often give insights on problems experienced by the userbase as well as requests for new features. However the usually immense amount of these reviews poses a relevant challenge to the knowledge extraction process. Therefore the proposal of this work is to exploit Natural Language Processing and Machine Learning techniques to help developers in the identification and assessment of anomalies and feature requests signaled by their users via app reviews. To this end the first part is dedicated to review the current state of the art in terms of methods and results. After this initial step, different promising techniques are implemented and tested for the specific use case of this work and notably using a dataset containing text data in the Italian language. The final result is a defect management dashboard designed to be integrated into an online platform to help automatize the information mining process. The dashboard presents several indicators useful to monitor the app status as perceived by its customers. The project was carried out in collaboration with Iriscube Reply, a company active in the development and maintenance of mobile applications for financial institutions, for this reason the app on which the analysis is conducted is one of the most famous online banking apps in Italy. The results show that the dashboard can indeed be a promising tool, potentially useful to help domain experts in their everyday job of maintenance and assistance, by reducing the time spent manually analyzing the reviews and by giving clear directions to prioritize relevant problems which affects the largest amount of users.