From an Isolation to a Conflict Paradigm for Understanding Polarisation in Social Media Spaces

Day two at the Indicators of Social Cohesion symposium begins with the great Petter Törnberg, who begins with a brief review of the changing understanding of the public sphere. With the arrival of the Web and (later) social media, there was early optimism about a new democratic renaissance – an opportunity for more inclusive and diverse public debate after the mass mediatisation of public debate through commercial print and broadcast media.

This was true to some degree, and social media did become a discursive engine of our political public sphere – yet the discourse there wasn’t particularly cross-cutting or inclusive. The concern here is about isolation leading to radicalisation and a divergence of opinions and ideological positionings – and this has led us to draw substantially on network analysis as a mechanism for studying these patterns. In short, social cohesion has been understood as network cohesion – but this is problematic.

This understanding has been challenged recently by the fact that empirical studies have not found any evidence for echo chambers or filter bubbles as determined by network fragmentation; indeed, polarisation appears to be driven often by people encountering counter-attitudinal content, and not by their descent into isolated information cocoons. How can we develop new approaches to make sense of this, then?

Petter’s projects have explored this question using a number of case studies – and as it turns out, retweet or @mention networks (or more broadly, networks of information flows) are not sufficient for analysing these structures. We must also consider the content and valence of the connections between participants: are they in agreement or disagreement; are they engaging civilly or uncivilly; etc.

Petter suggests that this leads us to the concept of the ‘trigger bubble’: conflict rather than disconnection is the mechanism for polarisation, as we are thrown into partisan disputes and are forced to take sides – and social media logics explicitly encourage this process.

Does this apply to platforms other than Twitter, though – what about, for instance, fringe platforms like Stormfront? Analysis of 20 years of scraped messages from that platform show that this is a deeply social space for its fascist participants, and that they are acutely aware of the arguments of their opponents; they are not engaged in rational exchange, though, but in emotional therapy, storytelling, and (far-right) identity formation. Such spaces are ritualistic and assist in the creation of tribes with a sense of collective self. (This study is covered in Petter’s book Intimate Communities of Hate.)

To understand this, we need a new paradigm: clearly, the Internet provides spaces both for group formation and for conflict, and interaction in these can lead to both intra-group cohesion and inter-group polarisation at the same time. Compared to the isolation paradigm represented by the ‘echo chamber’ thesis, this requires a conflict paradigm, where politics is about identity and emotion, polarisation is about inter-group conflict, the object of study should be the content of interactions (not just the existence of interactions themselves), and the method should be centred on natural language processing rather than network mapping alone.

Large Language Models have made such kind of analysis considerably easier, and Petter has demonstrated some very promising approaches to using these tools, but there is also a considerable need for further quality control and validation of the outcomes they produce. They might be used, for instance, to examine the in-group and out-group definitions presented by partisans in social media debates; and this also needs to be done across social media platforms rather than only for any one of them in isolation. LLM-based simulation of social media activity may also be valuable here to explore possible polarisation mechanisms and identify potential mechanisms for bridging and depolarisation between opposing groups.

We must be very careful before we implement such possible mechanisms in the wild, of course; but they push us to consider how to improve social media platforms and their affordances, and scholarly work rather than further in-house development is critically needed to improve these spaces.