Exploration of Framing Biases in Polarized Online Content Consumption

The study of framing bias on the Web is crucial in our digital age, as the framing of information can influence human behavior and decision on critical issues such as health or politics. Traditional frame analysis requires a curated set of frames derived from manual content analysis by domain experts. In this work, we introduce a frame analysis approach based on pretrained Transformer models that let us capture frames in an exploratory manner beyond predefined frames. In our experiments on two public online news and social media datasets, we show that our approach lets us identify underexplored conceptualizations, such as that health-related content is framed in terms of beliefs for conspiracy media, while mainstream media is instead concerned with science. We anticipate our work to be a starting point for further research on exploratory computational framing analysis using pretrained Transformers.