Hierarchies of Power: Identifying Expertise in Anonymous Online Interactions

This paper sets the stage for our primary objective, which is to identify and examine various forms of claimed expertise in anonymous online interactions. By building upon the findings and incorporating the proposed enhancements, we aim to gain a deeper understanding of the nature and implications of different expertise claims within the context of power hierarchies. A combination of various machine learning techniques is employed in this work, including classical methods, deep learning models, and transformer-based approaches to create classification models, while using three datasets collected by specialists and annotated by linguistics experts. The first experiments’ results in binary classification, indicating whether a given post reflects expertise or not, are particularly promising, especially when utilising transformer-based approaches. The second set of experiments, focusing on the classification of different types of expertise, produced a diverse range of results with the less favourable results primarily caused by an imbalance in labelling between different classes.