Visualizing with Text: Bringing Content to the Fore to Better Assess (Mis)information

In this paper, the role of text visualization to misnformation is explored. In an age of Large Language Models, online misinformation, plagiarism, hate speech, and more—the analysis of text is ever more important. Our traditional statistical visualization tools, however, have lagged behind—geared towards the visual display of quantitative information—rather than text-centric unstructured data. Visualizing with Text characterizes the design-space for directly integrating text and visualization in each other. It builds on the traditional visualization pipeline familiar in statistics and visualization: it adds a) literal text; b) visual attributes such as font weight, x-height and many more; c) mark types ranging from individual alphanumeric characters to paragraphs; and d) representations from extended traditional visualizations such as scatterplots, line charts and treemaps with text marks; to text-centric visualization such as tables, dictionaries and mind-maps. There are many examples of text-centric visualizations as applied to qualitative data and misinformation: including a scatterplot of social media vs. mainstream media; a line chart of tweet popularity; a treemap of human rights; a chart of LLM verbatim responses; mindmaps of LLM knowledge extents; and a table of text for comparison.