Using Graphical Perception in Visualization Recommendation

different encoding choices, which could inform the development of visualization recommendation tools. However, when we surveyed current tools [2], we noticed a surprising pattern: They seem to reference few if any findings from graphical perception when recommending visual encodings. This result led us to another important question: Why aren’t current visualization recommendation tools incorporating experiment results and guidelines from graphical perception research? A natural starting point is to review the graphical perception literature and figure out which parts are most relevant to visualization recommendation tools. This led us to review 132 interesting works in graphical perception [3], from visualization textbooks to decadesold experiments of how people perceive bar charts to studies of what happens when you add iconography or other embellishments to visualizations, among others. The sheer breadth and depth of work was at times overwhelming, and we started to see the problems that developers were running into. For example, it’s a struggle to separate the papers (and textbooks) that are relevant to visualization recommendation from those that are A s data continues to grow at unprecedented rates, we encounter unique challenges in helping analysts make sense of it. A prime example involves visualizing the data, where an analyst may have to reduce thousands of data columns and billions of data records to a single visualization. This often involves selecting which columns to visualize; sampling, filtering, or aggregating the data down to a manageable number of records; and mapping the results to intuitive visual encodings such as positional axes, bar heights, or color hues. Every step of the way, the analyst must grapple with what to focus on and how to translate the focus into a compelling image. We see a small slice of this problem in Figure 1: We can generate many different visualizations for a movie dataset, but the default design choices can be problematic. For example, the line chart in Figure 1 is just a blob of blue pixels. How can visualization tools help analysts navigate this complex and even frustrating web of interconnected design decisions? We have seen an explosion of visualization recommendation tools responding to this challenge. These tools aim to reduce decision fatigue by automating part or even all of the visualization design process. We summarize how these tools behave based on what they aim to automate [2]: which parts of the data to focus on (recommending data columns, rows, queries, etc.), which visual encodings to apply (recommending scales, colors, shapes, etc.), or both. Graphical perception research Why aren’t current tools incorporating experiment results and guidelines from graphical perception research? Using Graphical Perception in Visualization Recommendation