By Rob Mitchum // October 20, 2014
Data visualization is among the hottest of science and technology terms right now, gathering more attention and hype as experts deal with larger and larger collections of data. But data visualization is actually a centuries-old process, with classic types such as the bar graph and the pie chart dating back to the 1700s. In his October 15th talk at the Computation Institute, Min Chen of the University of Oxford suggested that the recent wave of data viz may have lost sight of the original and perhaps most important purpose of the practice.
As a Professor of Visualization with the Oxford e-Research Centre, Chen has worked with researchers in several different fields on visual interpretations of their data, from glacier movement to poetry to rugby. Currently, Chen is collaborating with the CI and University of Chicago research center ARTFL on their Commonplace Cultures project, helping their researchers identify shared passages between 18th century texts. Across all of these projects, Chen said his team aimed to create visualizations where the primary goal was not necessarily to “gain insight,” but instead to help researchers conserve a precious resource: time.
To explain, Chen talked about how analytical aims are only one of the four levels of visualization, alongside disseminative, observational, and inventive forms. While some graphics attempt to reveal previously hidden relationships or correlations, many more serve a more humble purpose, helping observers quickly make sense of data that would be otherwise difficult to conceptualize and a bottleneck on discovery.
Chen’s primary example was stockbrokers, for whom the simple time series of a stock’s price is a key professional tool. The simple line graph of a stock’s performance is a dramatic improvement over a spreadsheet or other layout, helping brokers instantly spot patterns and anomalies that may drive their trading activity. These simple visualizations also act as a form of “external memorization,” Chen said, saving busy people from the difficult task of holding multiple numbers in their memory simultaneously. Altogether, these time series likely save stockbrokers 90 percent of their time, he estimated, in an industry where time is literally money.
The stockbroker case study was an example of visualization at the observational/operational level, where Chen’s group has done several innovative designs. A collaboration [pdf] with the Department of English at the University of Utah searched for ways to graphically depict poetry, arriving at a map of a poem’s “flow” — how it sounds when it is read aloud during a close reading. Another project helped scientists in Greenland study the movement of more than 200 glaciers over several years. Instead of a crowded time series or a hard-to-read map, Chen’s team developed a visualization that combines the two forms, presenting both time and location dimensions simultaneously.
Another intriguing form of data visualization was the fourth level, inventive visualizations, which allow researchers to evaluate the effectiveness of the model they are using on the data. One example Chen provided was in the young field of “facial dynamics,” taking on the difficult task of teaching computer vision algorithms to correctly read human facial expressions and emotion. By illustrating where the facial movements of various emotions differ across 300 different variables, the visualization helped researchers pick out which variables would be useful in developing a model to automatically read emotions.
With ARTFL, Chen is developing another inventive visualization system, helping their researchers comb through an enormous corpus of 18th century text to find identical passages and trace the transmission of knowledge. There are many different algorithms that one can use to find these overlaps, and Chen developed a simple drag-and-drop interface that allows a user to try out different combinations of search methods. Another visualization uses a pixel matrix to map the search results, drawing the user’s eye to interesting connections and patterns worth exploring further.
The system, still under development, embodies Chen’s concept of visualizations as tools that enable research, not just infographic summaries of final results. It also capitalizes upon the natural “data visualization” skills of human vision, perception, and thought — which, after all, uses 130 million retinal receptors and sophisticated neural processing to make sense of the “big data” of everyday life.
“The visual system is a manual for dealing with big data,” Chen said. “We are probably the most effective machine for dealing with big data, and there are a lot of things we can learn from ourselves.”
[You can read a 2013 paper by Chen that touches on some of these themes at arXiv, or view slides from a similar talk here.]