When analyzing a large dataset, it can be overwhelming trying to figure out where to look for interesting insights. Visualizing the data can help, but manual chart specifications can also be a daunting and time-consuming task for users without extensive domain expertise. To help make datasets easier to understand and examine, in 2015, a team of researchers led by Allen School professor Jeffrey Heer introduced Voyager, a system that automatically generates and recommends charts and visualizations based on statistical and perceptual measures — allowing users to efficiently explore different parts of the dataset they may not have discovered before.
Since its release, Voyager has been called a “landmark development” and has helped transform visualization technology from human-led interactive visualization to a mixed-initiative approach. At IEEE VIS 2025 earlier this month in Vienna, Austria, Heer and his co-authors were recognized with the InfoVis 10-Year Test of Time Award for the Voyager paper’s lasting impact on the field. With the advancement and increasing popularity of artificial intelligence tools, the award committee noted, the system’s mixed-initiative approach is even more relevant today.
“The Voyager paper is exciting to me for many reasons. The work introduced a new approach to visualization recommendation and how to richly incorporate recommendations within user interfaces — all of which has proved influential to ongoing research,” said Heer, who holds the Jerre D. Noe Endowed Professorship in the Allen School. “To provide a solid representation for reasoning about visualizations and recommending charts, we also invented Vega-Lite, a high-level language for statistical graphics that went on to become a popular open source tool in its own right.”
Underlying Voyager is the Compass recommendation system which takes in user selections, data schema and statistical properties and generates suggested visualizations in the form of Vega-lite specifications. For example, an analyst looking at a dataset about cars can use Voyager to examine the effect of different variables such as horsepower, number of cylinders and acceleration. If the analyst is interested in looking at horsepower, they can browse charts with varied transformations of horsepower in the Voyager gallery and find the car with the highest horsepower. Voyager can also recommend visualizations with additional variables to help the analyst see if there are potential correlations or pursue other questions.
In a user study comparing Voyager to a visualization tool modeled after Tableau, the researchers found that Voyager helped users explore more of the data, leading to a significantly greater coverage of unique variable combinations.
Additional authors include Kanit Wongsuphasawat (Ph.D., ‘18), visualization team lead at Databricks; Dominik Moritz (Ph.D., ‘19), faculty at Carnegie Mellon University; Anushka Anand, director of product management at Salesforce; Jock Mackinlay, former technical fellow at Tableau; and Allen School adjunct faculty member Bill Howe, a professor in the University of Washington Information School.
Read the full paper about Voyager here and read more about the InfoVis Test of Time Award here, as well as a related story from Carnegie Mellon University.