Explicit Stance Detection in the Political Domain: A New Concept and Associated Dataset

Stance detection, defined as the task of classifying an individual's attitude towards a target person or concept, offers the potential to understand political opinions at scale using social media data. However, recent studies have questioned the robustness and accuracy of current stance detection methods, highlighting issues such as generalizability in time and inconsistencies in annotations driven by subtle differences in annotation task design. We argue that central to these challenges is the unresolved question of what constitutes an expression of stance. To address this, the present work introduces a distinction between explicit and implicit stance expressions, and argue that a focus on explicit stance detection addresses many of the existing concerns with modern stance detection methods. To facilitate research on explicit stance detection, we then present a novel (and public) dataset of over 1000 tweets across 13 stance targets for explicit stance detection and evaluate baseline models to establish a foundation for future research in this area.