PhD Defense: 'Drawing Digital Lines: Pattern Analysis of Divisive Rhetoric in Social Network Discussions'

The early internet promised to widen public discussion and improve collective reasoning. Two decades later, online debate increasingly resembles a “dark forest”, where moderate voices withdraw and the most adversarial registers dominate. Dominant explanations—algorithmic radicalization, disinformation and echo chambers—cast users as passive recipients of external forces.

This thesis restores participants’ agency by recasting online controversy as a peer-to-peer rhetorical phenomenon. Drawing on social identity theory, it treats contested discussion as identity work carried out through divisive rhetorical techniques (DRT), embedded in rhetorical ecologies shaped by both platforms and users.

To test this account, the thesis develops a computational apparatus combining large language models and fine-tuned encoders, applied to YouTube comments and transcripts on immigration. The results show that DRT track controversy and carry information not reducible to stance or semantics. They also show that audiences align rhetorically with the sources they engage with, forming four coherent communities.

The work sits at the intersection of philosophy, computational linguistics and social science, and was carried out within the Marie Skłodowska-Curie HYBRIDS network.

Supervisors: Martín Pereira-Fariña and Renata Vieira