Stance detection on digital social media ; Détection des points de vue sur les médias sociaux numériques
Many domains have an interest in studying the stances expressed online, whether for marketing, cybersecurity or research with the rise of the digital humanities. In this manuscript, we propose two contributions to the field of stance detection, focusing on the difficulty of obtaining quality annotated data on social media. Our first contribution is a large and complex dataset of 22,853 active Twitter profiles during the 2017 French presidential campaign. This is one of the few datasets considering more than two stances and, to our knowledge, the first with a large number of profiles and the first proposing overlapping political communities. This dataset can be used as-is to study campaign mechanisms on Twitter or to evaluate stance detection models or network analysis tools. We then propose two semi-supervised generic stance detection models, using a handful of seed profiles, for which we know the stance, in order to categorize the rest of the profiles by exploiting different inter-profile proximities. Indeed, the current models are generally based on the specificities of certain social platforms, which does not allow the integration of the multitude of available signals. By building proximities from different types of elements available on social media, we can detect profiles close enough to assume that they share a similar stance on a given subject, regardless of the platform. Our first model is a sequential model propagating the stances thanks to a multilayer graph representing the proximities between the profiles. Using datasets from two platforms, we show that by combining several types of proximity, we can correctly label 98% of the profiles. Our second model allows us to observe the evolution of the profiles' stances during an event, with only one profile-seed per point of view. This model confirms that the vast majority of profiles do not change positions on social media, or do not express their change of opinion. ; De nombreux domaines ont intérêt à étudier les points de vue exprimés en ligne, que ce ...