On the Role of Interaction in Visual Data Science
Helwig Hauser
University of Bergen, Norway
Interactive visual data exploration – also referred to as exploratory data analysis, EDA – is common in data science, not only as part of hypothesis generation, but also when prototyping appropriate data analysis workflows. Interactive data visualization is a key ingredient and a solid foundation of useful methods – both regarding appropriate visual representations as well as user interaction techniques – has been established and evaluated in the scientific field of visualization and visual analytics, including coordinated multiple views, linking and brushing, focus+context visualization, etc.

In this contribution, we take a closer look at the role of interaction in visual data science.  We examine potential benefits – for example, how interactive visualization becomes part of externalization, enabling and supporting complementary levels of cognition during data exploration – as well as possible downsides (human-in-the-loop processes are potentially costly, requiring the user’s time, for example).  We discuss, how interactive visualization can be understood as a form of dialogue between the analyst and her/his data and we analyze, which requirements – primarily in the temporal sense – are faced when establishing an effective and efficient interaction.

Besides an overview of selected examples from published work in visualization research, we also considered a few concrete examples from own prior work on interactive visual data exploration and analysis, including a demonstration of how deep learning can help to optimize user interaction in visualization.

Keywords: visual data science, interactive visualization, interaction.