Aller au contenu principal

PONs - People-Opinion Networks: A study of polarization in word embeddings and social networks in Switzerland and Southern Africa. Davide Morselli
Kevin Durrheim

This multidisciplinary research collaboration brings insight and methodological development on how opinions get polarized in the media and social media in Switzerland and Southern Africa. We propose a novel method called speaker landscape, using Natural Language Processing and grounded in social psychological theory, for studying how social networks develop around opinions and become polarized.

Our overall objective is to study polarization in a framework that jointly considers opinion-based and social structure definitions of polarization. The way to do this is to link two networks, one derived from each approach. While the social-structure network is straightforward to construct, the methods for understanding opinion-based networks are less clear. To this goal, we have developed the speaker landscape method, an opinion-based network extracted from word embeddings.

Speaker landscapes are network representations of social opinion that can be used to study polarization. This method extracts opinion-based networks from word embeddings trained on media and social media data. Our second aim is to link social networks and opinion networks in multilayer representations that we call People-Opinion Networks (PONs). The social networks can be developed from interactions between agents on social media. Our third aim is to use our multilayer PONs model in a comparative investigation of polarization different socio-political contexts, like Switzerland and South Africa.

In this project, we are performing cross-sectional and longitudinal analyses of polarization in both media and social media. These investigations were informed by social psychological view that polarization arises from intergroup comparisons, and that the emergence of polarization depends on the spread of opinions between groups. These investigations exploits the full potential of our multilayer PONs representation to examine how polarization arises from potentialities in both social and opinion structures. Our investigation has been focusing on the debate around Covid-19 vaccination in South Africa, as well as other settings including gendered opinions on parenthood, the semantic influence of social movements on the press, and processes of radicalization.


Important links

Project page on the SNSF website

Institutions liées
Projet soutenu par