2025
By Roeland Dubèl, Mark Boukes, Sandra Jacobs, and Damian Trilling
“News websites closed their own comment sections as the volume of comments proved overwhelming, making manual moderation unfeasible. Contemporary developments in computational methods, such as large language models, offer the potential to automatically moderate comment sections and retrieve valuable user feedback. This project aims to develop such a pipeline. We compiled a dataset of 3,305,599 social media comments towards 18 Dutch news outlets, spread across Facebook, Instagram, LinkedIn, TikTok, Twitter, and YouTube. Currently, we use this dataset for two pilot studies: Testing whether the informational quality of comments can automatically be detected, and whether key themes can automatically be extracted.”