||Bangkok, Berlin, Moscow, Sao Paulo, Tokyo
||Redi, Crockett, and Manovich. ACMMM '16
This is a "one-on-two" direct visualization of Instagram images. This visualization type maps a single variable onto two spatial axes, so that values for that variable correspond with closeness to a central point. If the variable is ranked, this produces a tight circular cluster (and would be a sphere in a "one-on-three" visualization). Note that a ranked visualization is only informative if the plot elements have non-relational properties, as they do here, since the plot elements are images. Consider, alternatively, simple scatter points arranged in this fashion — they would carry no information other than giving the size of the dataset. I use plots like these to reveal the nature of visual variables. I didn't know what this particular visual feature was; this plot tells me. The data here are sampled such that images with very high arousal scores are heavily weighted. This helps draw a clearer contrast between high and low scores for the feature.