|
Reading the digital public sphere: toxicity, language evolution, and AI in online conversations | title | Reading the digital public sphere: toxicity, language evolution, and AI in online conversations |
|---|
| start_date | 2026/05/29 |
|---|
| schedule | 11h |
|---|
| online | no |
|---|
| location_info | visioconférence Big Blue Button |
|---|
| summary | One of the most significant contemporary technological transformations is the overabundance of digital data produced by online interactions. This data has enabled the training of large-scale language models and opened new possibilities for studying the digital public sphere empirically and at scale. A central challenge, however, is distinguishing what genuinely changes across platforms and time from what remains invariant.
Drawing on a series of computational social science studies based on large-scale social media data, this talk examines persistent and evolving patterns of online public discourse across five papers.
A comparative analysis of approximately 500 million comments across eight platforms and 34 years (Avalle et al., 2024) shows that toxic content does not reduce participation contrary to its operational definition, and that ideological controversy is the primary driver of hostile exchanges. A second study (Di Marco et al., 2024) investigates long-term changes in linguistic complexity across platforms, identifying a gradual reduction in text length and lexical richness over time. Another paper (Loru et al., 2025) shows that ideological alignment is a stronger predictor of engagement in global online debates than actor category or institutional authority. Research on the 2024 U.S. presidential election (Bonetti & Maiorano, 2025) further analyzes Truth Social as a highly homogeneous ideological platform characterized by extensive circulation of low-reliability information sources. Finally, ongoing work examines the integration of GrokAI into X conversations, exploring how platform-embedded large language models may influence conversation dynamics.
Across all five studies, a common question remains: what do digital traces of language say about human behavior, and what changes — or persists — as the digital public sphere evolves?Bonett |
|---|
| responsibles | Bawden |
|---|
Workflow history| from state (1) | to state | comment | date |
|---|
| submitted | published | | 2026/05/26 07:32 UTC |
| |
|