Similarities in Sentiment in the Comments to Far-Right YouTube Videos

The second paper in this Social Media & Society 2024 session is by Claudia Zucca, whose interest is in a network analysis of selective exposure on YouTube. The case study here is a Dutch far-right channel on the platform, and focusses on a sequence of five randomly selected videos, for which the project scraped the comments.

Based on this dataset, the project constructed a network of interactions from @mentions in the comments, and used ChatGPT for zero-shot sentiment analysis (using simple positive, neutral, negative labels). Most of the sentiment turned out to be negative, perhaps unsurprisingly.

Using exponential random graph models, this explores how similar opinions (or sentiments) disseminate across the network of comments. (There is a very complicated model here by which this analysis is being shoehorned into a distinction between ‘echoes’, ‘chambers’, and ‘full echo chambers’, which frankly makes very little sense to me.)