Inférer l’espace idéologique depuis les médias sociaux
Comment reconstruire les dynamiques politiques dans les nouveaux espaces numériques?
Les traces numériques publiques des médias sociaux, en particulier Twitter, permettent d’inférer l’espace idéologique que leurs utilisateurs composent et occupent.
The public digital traces of social media, in particular Twitter, allow us to infer the ideological space that their users compose and occupy.
Inspired by network scaling methods (Pablo Barberá et al. 2015), we rely on the topological structure of French parliamentarians' followers to infer the structure of political "attitudes" of Twitter users in France. The space that emerges from the Correspondence Analysis is composed of two main dimensions: the classical right/left divide and a second axis opposing supra-nationalist and local views. It is possible to embed actors whose political orientations are known to validate the interpretation of the latent ideological space. In the figure below, the French deputies and senators are positioned, as well as the average position of their party.
This technique allows us to position hundreds of thousands of users. From then on, this signal can be propagated to artifacts manipulated by these users; for example, URLs shared on Twitter. By successive aggregations, the center of gravity of the URLs published by a media allows to build the political landscape of the French media space (Cointet et al., 2021) and, why not, to measure its degree of polarization. This signal can also be propagated to other accounts or other social networks (Ramaciotti Morales et al., 2020). It was through these methods that we were able to map the political space occupied by the Yellow Vests groups on Facebook (Cointet et al., 2021). Beyond Twitter and Facebook, these projections allow us to position other web entities, such as YouTube channels and websites in general. This allows for broad investigations of a digital public space; for example, the study of the deployment of the Yellow Vests movement on this multi-platform space (Ramaciotti Morales et al., 2021)