The post-lunch panel at the 2026 Association of Internet Researchers conference in Mexico City is on polarisation, and starts with a paper by Marco Bastos, who begins by introducing the various understandings of polarisation that exist in the literature. These also differ by the unity of analysis, and the present study focusses on narratives as its unit of analysis. Can these be quantified, and their polarising qualities assessed?
Narratives, put simply, are stories about how and why things happen. There are various structural models of narratives, such as Greimas’s actantial model, involving a key subject aided by supporters and frustrated by opponents; in the Gaza conflict, for instance, the Israeli narrative is that Hamas is preventing the Palestinians from gaining their rights to statehood, while in the Palestinian narrative it is Israel which acts as the opponent.
This analysis can be operationalised via LLMs: they are proficient at extracting narratives from texts, though the manual verification of such extraction remains time-consuming. This project extracted such narratives from pro-Israeli and pro-Palestinian YouTube videos, and the comments attached to them, drawing on 100 videos from each side, expecting to identify narrative polarisation between the two sides.
In this analysis, violence emerges as a key element of these narratives, but such violence is portrayed in very different ways: pro-Palestinian narratives emphasise Israeli violence, and vice versa. The distance between these two measures the depth of polarisation between them.
This also finds a substantial divergence between narratives in transcripts and comments. Narrative polarisation in videos is a great deal more pronounced than in the comments; indeed, narratives in the comments tend to push back to some extent against the narratives in the videos themselves.
Such an analysis can also be applied to many other datasets, and may surface much more nuanced narrative motifs compared to the study of structural polarisation. Overall, too, it appears that the crowd operate as a depolarising force compared to the influencers.
