A Topic is Not a Theme: Towards a Contextualised Approach to Topic Modelling

There is a disconnect between an understanding of the processes of machine learning at a technical level and the societal impact of these technologies. This thesis considers the robustness of topic models. Researchers have increasingly applied topic models to social media datasets to study social phenomena. However, the application of topic modelling for social media analysis has cast doubt on the validity, usability, and direction of research on topic modelling. Given such shortcomings, this thesis addresses the theoretical and methodological practices and empirical aspects of topic modelling to identify opportunities to improve the development and application of topic models.