Compare Clouds: Visualizing Text Corpora to Compare Media Frames

Media frames represent distinct ways of communicating about issues that are reflected in choices of key words and phrases. In this paper we develop a visualization technique and visual analytic system that enables the study of media frames across text corpora. In particular our system allows scholars or other analysts to compare media frames in a visualization called the Compare Cloud, which explicitly maps word prevalence and context information between two corpora. We assess the error profile of the visualization layout and demonstrate the utility of the system by comparing the media discussion between mainstream media and blogs on the topic of surveillance. We report salient observations that the visualization made possible and discuss future challenges related to scalability and effective filtering to support visual frame analysis. Author