Visual Politics: A Deep Learning Approach to the Spread and Stick of Political Ideas

This project addresses the visual dimension of power over public agenda in the digital age. Visual politics is increasingly prominent.

However, we know little about the actors and mechanisms that work through images to promote political ideas, and even less about the outcomes with respect to the spread and stickiness of political ideas. Under what conditions does visual information amplify content and engage citizens in ways that shape public discourse?

A key obstacle has been the inability to address complex information spread based on actual images.

The spread of ideas is typically studied via text, but images differ from text in what they amplify and convey, how they engage, and how much they are shared online.

The project therefore has two aims: (1) We combine qualitative approaches with advances in network science and pattern recognition via deep learning neural networks to develop image-based computational methods to study the visual spread of political ideas online; (2) We deploy the methods to identify and trace visual narratives around a controversial issue (climate change) across media platforms, to assess how and why some visual narratives become prominent and resonate across arenas, groups and time.

The project contributes state-of-the-art methodology that can be applied across issues, sheds light on the conditions for integrating AI into political research, and provides fresh traction on a classic phenomenon in the study of democracy: power over public agenda.