Presented at the AI & Ideology Workshop, Sciences Po Paris. This talk covers the computational pipeline from LLM-based text annotation to ideological drift detection, with applications on two original databases.
YouPol

24,700 political videos from YouTube and TikTok across France and Quebec, with 2.95 million annotated sentences. A panel analysis across 19 channels and 36 quarters reveals a statistically significant neo-reactionary drift of +1.13 pp/decade in far-right YouTube content.

CCF Database

266,000 articles from 20 Canadian newspapers (1978-2024), with 9.2 million sentences annotated across 65 categories. Media cascade detection and thematic frame evolution over nearly five decades.

Computational Methods for Social Sciences
PDF Open
Scroll Navigate between slides Click Interact with data Fullscreen for a better experience
Use arrow keys or click to navigate through the slides.
Open in light mode

Projects and resources

All tools and platforms presented are open source.