Voyager++
Voyager++ is an experimental website prototype designed to be used by data analysts while they examine large datasets. It is built on top of the existing Voyager 2 software, which is a data exploration tool, using React and TypeScript. Voyager++ was created under the direction of Dr. Emily Wall and Dr. Jessica Hullman as part of the research lab Midwest Uncertainty Collective.
Voyager++ allows users to record their preconceived beliefs about the distribution a dataset will take before they examine the true data. Within the data science community, it is well known that having users write down their beliefs before examining data often causes a deeper examination of the data. After many rounds of brainstorming and evidence gathering for the most effective ways for users to log their beliefs about different types of data, Voyager++ was designed as a one-stop site that integrated the benefits of logging beliefs into a realistic data exploration workflow. Users can organically explore a dataset, log their pre-existing beliefs, and examine how the distributions they predict match up against the true data without adding an excessive amount of time to their normal process of exploring and analyzing datasets.
As part of the design process, I drew many wireframes of how users could interface with the site and visually share their beliefs about datasets. These were presented at lab meetings and iterated on based on feedback.
See below an image of how a user would share their beliefs about the distribution of IMDB ratings for different movie genres in a dataset about movies. After they share their beliefs, their predictions are displayed alongside the true data distributions in different colors so they can examine the difference as shown in the second image. The methods of belief elicitation and display vary by the type of data being examined.
