Shapes in Scatterplots: Comparing Human Visual Impressions and Computational Metrics

AbstractWe are currently in the process of designing and implementing a computational cognitive system that combines perception, memory, attention, and domain-specific semantic knowledge to perform data visualization tasks. While this work is still in early stages, we report here on one subset of this larger project that involves building a ``visual long term memory'' for the system. To constrain the problem, we assume a domain of astronomy, and we focus exclusively on scatterplot visualizations. In this paper, we present three of our initial steps along this path. First, we collected and analyzed a catalog of 74 scatterplots from real astronomy sources (papers, books, etc.), which we consider to be typical data visualizations that astronomers would frequently encounter during their education. Second, we asked a team of human raters to rate all 74 scatterplots along nine dimensions describing shape categories, taken from a computational approach originally suggested by John and Paul Tukey called scagnostics. Third, we calculated computer-based scagnostics for a subset of the scatterplots. We measured inter-rater agreements among the human raters and between the calculated and human ratings.


Return to previous page