Andreu Casas
@andreucasas.bsky.social
Associate Professor of Political Communication and Computational Social Sciences at Royal Holloway University of London. Director of the London Social Media Observatory. UKRI Future Leader Fellow.
And as usual, the highlight of the week was to learn about the participants’ projects and to discuss with them how to take them to the next level. Here a few pictures from our traditional final session “Slides with Slices” 🍕 where participants presented the projects they worked on during the week.
September 19, 2025 at 4:32 PM
And as usual, the highlight of the week was to learn about the participants’ projects and to discuss with them how to take them to the next level. Here a few pictures from our traditional final session “Slides with Slices” 🍕 where participants presented the projects they worked on during the week.
Do you want to use large amounts of visual data (or text + visuals) in your research but don't know how? Join my "Computer Vision for Image and Video Analysis" course at GESIS this September (15-19th) to learn about machine learning for automatic visual analysis: training.gesis.org?site=pDetail...
July 31, 2025 at 1:33 PM
Do you want to use large amounts of visual data (or text + visuals) in your research but don't know how? Join my "Computer Vision for Image and Video Analysis" course at GESIS this September (15-19th) to learn about machine learning for automatic visual analysis: training.gesis.org?site=pDetail...
Friday (11:20-13:00: 1A.12) Large Visual Language Models for Supervised Multimodal Classification in Political Science Research. We compare 2 VLMs to 6 other supervised classifiers (incl. LLMs: Llama2&3), across 2 datasets and 10 tasks. VLMs outperform all models across all tasks, w. some caveats.
June 25, 2025 at 8:52 AM
Friday (11:20-13:00: 1A.12) Large Visual Language Models for Supervised Multimodal Classification in Political Science Research. We compare 2 VLMs to 6 other supervised classifiers (incl. LLMs: Llama2&3), across 2 datasets and 10 tasks. VLMs outperform all models across all tasks, w. some caveats.
Thursday (13:10-14:50: 1A.13): Moderation of Political Content on YouTube during the 2024 US Election. I tracked 20k channels and 6mil videos. Conservatives suspended more often, but also post most of hateful/misinfo. No difference in visibility, w. higher engagement for conservatives.
June 25, 2025 at 8:52 AM
Thursday (13:10-14:50: 1A.13): Moderation of Political Content on YouTube during the 2024 US Election. I tracked 20k channels and 6mil videos. Conservatives suspended more often, but also post most of hateful/misinfo. No difference in visibility, w. higher engagement for conservatives.
We discuss the problems that emerge for research, and put forward solutions. Beyond academia, the findings emphasize that AI models trained on human data can replicate cultural, ideological, gender, etc. biases; and call for closer attention and more diversity in AI development.
March 28, 2025 at 1:46 PM
We discuss the problems that emerge for research, and put forward solutions. Beyond academia, the findings emphasize that AI models trained on human data can replicate cultural, ideological, gender, etc. biases; and call for closer attention and more diversity in AI development.
We show systematic differences between annotators of different gender and ideology. These emerge even for fairly objective tasks (e.g. is a protest shown in the image?), may not be detected by inter-rater reliability statistics (e.g. Cohen Kappa), and can bias downstream analyses
March 28, 2025 at 1:46 PM
We show systematic differences between annotators of different gender and ideology. These emerge even for fairly objective tasks (e.g. is a protest shown in the image?), may not be detected by inter-rater reliability statistics (e.g. Cohen Kappa), and can bias downstream analyses
🚨New publication @The_JOP on human biases in data annotation (w. Nora Webb Williams, @kevinaslett.bsky.social, John Wilkerson). Extremely important given the increasing societal reliance on AI tools often trained on human coders www.journals.uchicago.edu/doi/10.1086/...
March 28, 2025 at 1:46 PM
🚨New publication @The_JOP on human biases in data annotation (w. Nora Webb Williams, @kevinaslett.bsky.social, John Wilkerson). Extremely important given the increasing societal reliance on AI tools often trained on human coders www.journals.uchicago.edu/doi/10.1086/...
New report "Academic Access to Social Media Data for the Study of Political Online Safety" w. @georgidag.bsky.social @benoloughlin.bsky.social osf.io/preprints/so... As new threats reach the social media environment, platforms continue to obstruct independent research. It’s time to act!
January 17, 2025 at 10:17 AM
New report "Academic Access to Social Media Data for the Study of Political Online Safety" w. @georgidag.bsky.social @benoloughlin.bsky.social osf.io/preprints/so... As new threats reach the social media environment, platforms continue to obstruct independent research. It’s time to act!