Projects by Faculty Member: Jessica Hullman
Expert journalists and designers often present visualized data related to an article to help people gain context for what they are reading (e.g., a locator map to help readers place a foreign location). The systems we are developing analyze the text of a news article, identify relevant datasets, and produce automated, annotated visualizations to help readers better understand the context of the article.
Visualizations like scatterplots and bar charts are common ways of presenting data for analysis but in contrast to statistical analysis, visual analysis introduces perceptual errors and cognitive biases. This project explores what factors, related to both the data set and the visual presentation of the data, most impact a person's judgments about the data.
Variation and uncertainty are unavoidable in data analysis but most people have trouble incorporating uncertainty into their interpretations of a data set. This project aims to identify the design constraints and experimentally evaluate the usefulness of a technique that depicts uncertainty as a set of alternative possible outcomes. The outcome plots are presented in an animated or interactive format, so that the user can gain a better sense of the potential for variation in the data by watching possible outcomes "play out".
Measurements are ubiquitous yet can be challenging to understand when the unit (e.g., decaliters, tons) or magnitude (e.g., 320m, $5 mil) are unfamiliar to us. Strategies like re-unitization, in which an unfamiliar measurement is re-expressed using a new unit (e.g., 10kg is equal to the weight of 2 printers), can aid understanding but often require a skilled designer to realize. These re-expressions can also be personalized given some information about the user, such as their location (e.g., 11 miles is twice the distance from your house to the Space Needle). This project develops databases of familiar objects and landmarks and their measurements, drawing on web crawling techniques, semantic databases like WordNet and ImageNet, object databases like Amazon and Wikipedia, and crowdsourcing. We design automated algorithms for strategies like re-unitization and proportional analogy that rank re-expressions based on a number of dimensions. We apply these automated strategies in web applications that allow a user to get on-demand re-expressions of complex measurements.