Jessica Ruth Hullman

Research - Selected projects

Storytelling and Framing in Information Visualization.

My work provides tools and theory for understanding and producing visualization-based context and storytelling. I led the development of a system called Contextifier, which supplements online news stories about companies with customized, automatically generated annotated visualizations. More recently, I collaborated with others to extend this approach to creating annotated maps of data that is thematically-relevant to a news article (NewsViews). These systems develop and evaluate generalizable criteria for contextualizing visualizations including relevance to the news article and visual salience or "interestingness."

An IEEE InfoVis 2011 paper investigates the often tacit, persuasive nature of narrative visualizations, those that subtly prioritize some data over others to "tell a story." An IEEE InfoVis 2013 paper examines the role of sequence and other structural techniques in rendering a set of visualizations to convey a narrative, drawing on work in decision science and evaluating design principles through crowdsourced studies. This work also proposes an approach to modelling sequence so as to offer semi-automated design support for visualization presentation creators. I am currently exploring integrating this approach with existing automated presentation techniques that optimize for the best singular data graphic.

Visual Analytics for Broad Audiences

Most of my work is driven by an interest in how people interpret visualizations on a perceptual and cognitive level. I've developed a method for visualizing uncertainty more directly for non-statisticians. A set of hypothetical data samples is generated and presented in an animated or interactive format. The approach generalizes to a number of data inputs and complex visualization types that lack uncertainty representations, like choropleth maps, treemaps, and network diagrams. Studies show that ppresenting uncertainty as hypothetical samples helps users better infer how reliable a data pattern they see is, and can reduce overconfidence in unreliable data.

Hypothetical samples are also beneficial when presenting visualizations to an online crowd to evaluate a hypothesis. For example, mechanisms that present each crowd member with several pieces of visual evidence and ask them to make an overall judgment are affected by order biases, while presenting a single visualization to a crowd can result in a systematic bias across the crowd. I am conducitng an analysis that compares this approach to distributing singular hypothetical data samples to each member of a crowd.

A CHI 2011 paper contributes experimental evidence of how social proof can cause a visualization user to rely on information about what others saw in the same visualization, even for simple judgments like comparing bars in a bar graph. We find that dynamics resembling information cascades are possible, where the erroneous judgments of a very small number of initial visualization users propogate across an entire community. These results motivate new mechanisms for presenting social information along with data representations online.

Visual Difficulties

Conventional guidance for creating useful data graphics advises to simplify visual clutter and in general make the graph user's interpretation process as efficient as possible. However, there is support in various psychological fields for an opposing possibility: making graphs less cognitively-easy or efficient to understand can actually be better for the user. An Honorable Mention paper at InfoVis works toward refining visualization theory to account for the fact that introducing 'visual difficulties' to graphical perception can increase understanding and/or recall under various conditions. This possibility rests on the potential for obstructions to learning to increase the user's active, deep processing of the information, as well as to often increase their motivation or engagement to continuing interacting.