Politics through machine eyes: What computer vision allows us to see

Abstract

Computer vision techniques have recently burst onto the scene in visual political communication research, dramatically extending the scope of digital image and video analysis. Over the last five years, in particular, papers featuring computational methods have grown in frequency and breadth, appearing in a variety of journals across different disciplines, ranging from the digital humanities and social semiotics to political communication and data science. Topics have been equally expansive and include large scale examinations of political protest, candidate self-presentation style and partisan symbolism, affective polarization, gender stereotypes, debate performances and voter impression formation, public policy, electoral fraud, and topics outside of conventional politics such as extremist propaganda. In this article, each of these areas of investigation are briefly reviewed for the insights they are revealing about political visuals as well as the tools and techniques employed. Ethical considerations are also explored to highlight some of the tradeoffs inherent in research involving computational analysis of political visuals.